diff --git "a/speaker_embedder/pyannote-v3-pro/W8A16/SpeakerEmbedder.mlmodelc/model.mil" "b/speaker_embedder/pyannote-v3-pro/W8A16/SpeakerEmbedder.mlmodelc/model.mil" --- "a/speaker_embedder/pyannote-v3-pro/W8A16/SpeakerEmbedder.mlmodelc/model.mil" +++ "b/speaker_embedder/pyannote-v3-pro/W8A16/SpeakerEmbedder.mlmodelc/model.mil" @@ -1,12 +1,12 @@ program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3402.3.2"}, {"coremlc-version", "3402.4.1"}})] { - func main(tensor preprocessor_output_1, tensor speaker_masks) { + func main(tensor preprocessor_output_1, tensor speaker_masks) { tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; tensor var_82 = const()[name = tensor("op_82"), val = tensor([0, 2, 1])]; tensor input_1_axes_0 = const()[name = tensor("input_1_axes_0"), val = tensor([1])]; - tensor fbank_cast_fp16 = transpose(perm = var_82, x = preprocessor_output_1)[name = tensor("transpose_0")]; - tensor input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = fbank_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor fbank_cast_fp16 = transpose(perm = var_82, x = preprocessor_output_1)[name = tensor("transpose_0")]; + tensor input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = fbank_cast_fp16)[name = tensor("input_1_cast_fp16")]; tensor input_3_pad_type_0 = const()[name = tensor("input_3_pad_type_0"), val = tensor("custom")]; tensor input_3_pad_0 = const()[name = tensor("input_3_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_3_strides_0 = const()[name = tensor("input_3_strides_0"), val = tensor([1, 1])]; @@ -14,8 +14,8 @@ program(1.0) tensor input_3_groups_0 = const()[name = tensor("input_3_groups_0"), val = tensor(1)]; tensor const_5_to_fp16 = const()[name = tensor("const_5_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor const_6_to_fp16 = const()[name = tensor("const_6_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(704)))]; - tensor input_5_cast_fp16 = conv(bias = const_6_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = const_5_to_fp16, x = input_1_cast_fp16)[name = tensor("input_5_cast_fp16")]; - tensor input_7_cast_fp16 = relu(x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor input_5_cast_fp16 = conv(bias = const_6_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = const_5_to_fp16, x = input_1_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor input_7_cast_fp16 = relu(x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; tensor input_9_pad_type_0 = const()[name = tensor("input_9_pad_type_0"), val = tensor("custom")]; tensor input_9_pad_0 = const()[name = tensor("input_9_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_9_strides_0 = const()[name = tensor("input_9_strides_0"), val = tensor([1, 1])]; @@ -23,8 +23,8 @@ program(1.0) tensor input_9_groups_0 = const()[name = tensor("input_9_groups_0"), val = tensor(1)]; tensor const_7_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10112))), name = tensor("const_7_to_fp16_palettized"), shape = tensor([32, 32, 3, 3])]; tensor const_8_to_fp16 = const()[name = tensor("const_8_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10688)))]; - tensor input_11_cast_fp16 = conv(bias = const_8_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = const_7_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor("input_11_cast_fp16")]; - tensor input_13_cast_fp16 = relu(x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor input_11_cast_fp16 = conv(bias = const_8_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = const_7_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor input_13_cast_fp16 = relu(x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; tensor input_15_pad_type_0 = const()[name = tensor("input_15_pad_type_0"), val = tensor("custom")]; tensor input_15_pad_0 = const()[name = tensor("input_15_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_15_strides_0 = const()[name = tensor("input_15_strides_0"), val = tensor([1, 1])]; @@ -32,9 +32,9 @@ program(1.0) tensor input_15_groups_0 = const()[name = tensor("input_15_groups_0"), val = tensor(1)]; tensor const_9_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20096))), name = tensor("const_9_to_fp16_palettized"), shape = tensor([32, 32, 3, 3])]; tensor const_10_to_fp16 = const()[name = tensor("const_10_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20672)))]; - tensor out_1_cast_fp16 = conv(bias = const_10_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = const_9_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor("out_1_cast_fp16")]; - tensor input_17_cast_fp16 = add(x = out_1_cast_fp16, y = input_7_cast_fp16)[name = tensor("input_17_cast_fp16")]; - tensor input_19_cast_fp16 = relu(x = input_17_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor out_1_cast_fp16 = conv(bias = const_10_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = const_9_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor("out_1_cast_fp16")]; + tensor input_17_cast_fp16 = add(x = out_1_cast_fp16, y = input_7_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor input_19_cast_fp16 = relu(x = input_17_cast_fp16)[name = tensor("input_19_cast_fp16")]; tensor input_21_pad_type_0 = const()[name = tensor("input_21_pad_type_0"), val = tensor("custom")]; tensor input_21_pad_0 = const()[name = tensor("input_21_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_21_strides_0 = const()[name = tensor("input_21_strides_0"), val = tensor([1, 1])]; @@ -42,8 +42,8 @@ program(1.0) tensor input_21_groups_0 = const()[name = tensor("input_21_groups_0"), val = tensor(1)]; tensor const_11_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30080))), name = tensor("const_11_to_fp16_palettized"), shape = tensor([32, 32, 3, 3])]; tensor const_12_to_fp16 = const()[name = tensor("const_12_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30656)))]; - tensor input_23_cast_fp16 = conv(bias = const_12_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = const_11_to_fp16_palettized, x = input_19_cast_fp16)[name = tensor("input_23_cast_fp16")]; - tensor input_25_cast_fp16 = relu(x = input_23_cast_fp16)[name = tensor("input_25_cast_fp16")]; + tensor input_23_cast_fp16 = conv(bias = const_12_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = const_11_to_fp16_palettized, x = input_19_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor input_25_cast_fp16 = relu(x = input_23_cast_fp16)[name = tensor("input_25_cast_fp16")]; tensor input_27_pad_type_0 = const()[name = tensor("input_27_pad_type_0"), val = tensor("custom")]; tensor input_27_pad_0 = const()[name = tensor("input_27_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_27_strides_0 = const()[name = tensor("input_27_strides_0"), val = tensor([1, 1])]; @@ -51,9 +51,9 @@ program(1.0) tensor input_27_groups_0 = const()[name = tensor("input_27_groups_0"), val = tensor(1)]; tensor const_13_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40064))), name = tensor("const_13_to_fp16_palettized"), shape = tensor([32, 32, 3, 3])]; tensor const_14_to_fp16 = const()[name = tensor("const_14_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40640)))]; - tensor out_3_cast_fp16 = conv(bias = const_14_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = const_13_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("out_3_cast_fp16")]; - tensor input_29_cast_fp16 = add(x = out_3_cast_fp16, y = input_19_cast_fp16)[name = tensor("input_29_cast_fp16")]; - tensor input_31_cast_fp16 = relu(x = input_29_cast_fp16)[name = tensor("input_31_cast_fp16")]; + tensor out_3_cast_fp16 = conv(bias = const_14_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = const_13_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("out_3_cast_fp16")]; + tensor input_29_cast_fp16 = add(x = out_3_cast_fp16, y = input_19_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor input_31_cast_fp16 = relu(x = input_29_cast_fp16)[name = tensor("input_31_cast_fp16")]; tensor input_33_pad_type_0 = const()[name = tensor("input_33_pad_type_0"), val = tensor("custom")]; tensor input_33_pad_0 = const()[name = tensor("input_33_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_33_strides_0 = const()[name = tensor("input_33_strides_0"), val = tensor([1, 1])]; @@ -61,8 +61,8 @@ program(1.0) tensor input_33_groups_0 = const()[name = tensor("input_33_groups_0"), val = tensor(1)]; tensor const_15_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50048))), name = tensor("const_15_to_fp16_palettized"), shape = tensor([32, 32, 3, 3])]; tensor const_16_to_fp16 = const()[name = tensor("const_16_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50624)))]; - tensor input_35_cast_fp16 = conv(bias = const_16_to_fp16, dilations = input_33_dilations_0, groups = input_33_groups_0, pad = input_33_pad_0, pad_type = input_33_pad_type_0, strides = input_33_strides_0, weight = const_15_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor("input_35_cast_fp16")]; - tensor input_37_cast_fp16 = relu(x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor input_35_cast_fp16 = conv(bias = const_16_to_fp16, dilations = input_33_dilations_0, groups = input_33_groups_0, pad = input_33_pad_0, pad_type = input_33_pad_type_0, strides = input_33_strides_0, weight = const_15_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor input_37_cast_fp16 = relu(x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; tensor input_39_pad_type_0 = const()[name = tensor("input_39_pad_type_0"), val = tensor("custom")]; tensor input_39_pad_0 = const()[name = tensor("input_39_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_39_strides_0 = const()[name = tensor("input_39_strides_0"), val = tensor([1, 1])]; @@ -70,9 +70,9 @@ program(1.0) tensor input_39_groups_0 = const()[name = tensor("input_39_groups_0"), val = tensor(1)]; tensor const_17_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60032))), name = tensor("const_17_to_fp16_palettized"), shape = tensor([32, 32, 3, 3])]; tensor const_18_to_fp16 = const()[name = tensor("const_18_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60608)))]; - tensor out_5_cast_fp16 = conv(bias = const_18_to_fp16, dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = const_17_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor("out_5_cast_fp16")]; - tensor input_41_cast_fp16 = add(x = out_5_cast_fp16, y = input_31_cast_fp16)[name = tensor("input_41_cast_fp16")]; - tensor input_43_cast_fp16 = relu(x = input_41_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor out_5_cast_fp16 = conv(bias = const_18_to_fp16, dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = const_17_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor("out_5_cast_fp16")]; + tensor input_41_cast_fp16 = add(x = out_5_cast_fp16, y = input_31_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor input_43_cast_fp16 = relu(x = input_41_cast_fp16)[name = tensor("input_43_cast_fp16")]; tensor input_45_pad_type_0 = const()[name = tensor("input_45_pad_type_0"), val = tensor("custom")]; tensor input_45_pad_0 = const()[name = tensor("input_45_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_45_strides_0 = const()[name = tensor("input_45_strides_0"), val = tensor([2, 2])]; @@ -80,8 +80,8 @@ program(1.0) tensor input_45_groups_0 = const()[name = tensor("input_45_groups_0"), val = tensor(1)]; tensor const_19_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79232))), name = tensor("const_19_to_fp16_palettized"), shape = tensor([64, 32, 3, 3])]; tensor const_20_to_fp16 = const()[name = tensor("const_20_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79808)))]; - tensor input_47_cast_fp16 = conv(bias = const_20_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = const_19_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor("input_47_cast_fp16")]; - tensor input_49_cast_fp16 = relu(x = input_47_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor input_47_cast_fp16 = conv(bias = const_20_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = const_19_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor("input_47_cast_fp16")]; + tensor input_49_cast_fp16 = relu(x = input_47_cast_fp16)[name = tensor("input_49_cast_fp16")]; tensor input_51_pad_type_0 = const()[name = tensor("input_51_pad_type_0"), val = tensor("custom")]; tensor input_51_pad_0 = const()[name = tensor("input_51_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_51_strides_0 = const()[name = tensor("input_51_strides_0"), val = tensor([1, 1])]; @@ -89,7 +89,7 @@ program(1.0) tensor input_51_groups_0 = const()[name = tensor("input_51_groups_0"), val = tensor(1)]; tensor const_21_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116928))), name = tensor("const_21_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_22_to_fp16 = const()[name = tensor("const_22_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117504)))]; - tensor out_7_cast_fp16 = conv(bias = const_22_to_fp16, dilations = input_51_dilations_0, groups = input_51_groups_0, pad = input_51_pad_0, pad_type = input_51_pad_type_0, strides = input_51_strides_0, weight = const_21_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("out_7_cast_fp16")]; + tensor out_7_cast_fp16 = conv(bias = const_22_to_fp16, dilations = input_51_dilations_0, groups = input_51_groups_0, pad = input_51_pad_0, pad_type = input_51_pad_type_0, strides = input_51_strides_0, weight = const_21_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("out_7_cast_fp16")]; tensor input_53_pad_type_0 = const()[name = tensor("input_53_pad_type_0"), val = tensor("valid")]; tensor input_53_strides_0 = const()[name = tensor("input_53_strides_0"), val = tensor([2, 2])]; tensor input_53_pad_0 = const()[name = tensor("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; @@ -97,9 +97,9 @@ program(1.0) tensor input_53_groups_0 = const()[name = tensor("input_53_groups_0"), val = tensor(1)]; tensor const_23_to_fp16 = const()[name = tensor("const_23_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(117696)))]; tensor const_24_to_fp16 = const()[name = tensor("const_24_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121856)))]; - tensor var_231_cast_fp16 = conv(bias = const_24_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = const_23_to_fp16, x = input_43_cast_fp16)[name = tensor("op_231_cast_fp16")]; - tensor input_55_cast_fp16 = add(x = out_7_cast_fp16, y = var_231_cast_fp16)[name = tensor("input_55_cast_fp16")]; - tensor input_57_cast_fp16 = relu(x = input_55_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor var_231_cast_fp16 = conv(bias = const_24_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = const_23_to_fp16, x = input_43_cast_fp16)[name = tensor("op_231_cast_fp16")]; + tensor input_55_cast_fp16 = add(x = out_7_cast_fp16, y = var_231_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor input_57_cast_fp16 = relu(x = input_55_cast_fp16)[name = tensor("input_57_cast_fp16")]; tensor input_59_pad_type_0 = const()[name = tensor("input_59_pad_type_0"), val = tensor("custom")]; tensor input_59_pad_0 = const()[name = tensor("input_59_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_59_strides_0 = const()[name = tensor("input_59_strides_0"), val = tensor([1, 1])]; @@ -107,8 +107,8 @@ program(1.0) tensor input_59_groups_0 = const()[name = tensor("input_59_groups_0"), val = tensor(1)]; tensor const_25_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158976))), name = tensor("const_25_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_26_to_fp16 = const()[name = tensor("const_26_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159552)))]; - tensor input_61_cast_fp16 = conv(bias = const_26_to_fp16, dilations = input_59_dilations_0, groups = input_59_groups_0, pad = input_59_pad_0, pad_type = input_59_pad_type_0, strides = input_59_strides_0, weight = const_25_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("input_61_cast_fp16")]; - tensor input_63_cast_fp16 = relu(x = input_61_cast_fp16)[name = tensor("input_63_cast_fp16")]; + tensor input_61_cast_fp16 = conv(bias = const_26_to_fp16, dilations = input_59_dilations_0, groups = input_59_groups_0, pad = input_59_pad_0, pad_type = input_59_pad_type_0, strides = input_59_strides_0, weight = const_25_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor input_63_cast_fp16 = relu(x = input_61_cast_fp16)[name = tensor("input_63_cast_fp16")]; tensor input_65_pad_type_0 = const()[name = tensor("input_65_pad_type_0"), val = tensor("custom")]; tensor input_65_pad_0 = const()[name = tensor("input_65_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_65_strides_0 = const()[name = tensor("input_65_strides_0"), val = tensor([1, 1])]; @@ -116,9 +116,9 @@ program(1.0) tensor input_65_groups_0 = const()[name = tensor("input_65_groups_0"), val = tensor(1)]; tensor const_27_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196672))), name = tensor("const_27_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_28_to_fp16 = const()[name = tensor("const_28_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197248)))]; - tensor out_9_cast_fp16 = conv(bias = const_28_to_fp16, dilations = input_65_dilations_0, groups = input_65_groups_0, pad = input_65_pad_0, pad_type = input_65_pad_type_0, strides = input_65_strides_0, weight = const_27_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("out_9_cast_fp16")]; - tensor input_67_cast_fp16 = add(x = out_9_cast_fp16, y = input_57_cast_fp16)[name = tensor("input_67_cast_fp16")]; - tensor input_69_cast_fp16 = relu(x = input_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor out_9_cast_fp16 = conv(bias = const_28_to_fp16, dilations = input_65_dilations_0, groups = input_65_groups_0, pad = input_65_pad_0, pad_type = input_65_pad_type_0, strides = input_65_strides_0, weight = const_27_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("out_9_cast_fp16")]; + tensor input_67_cast_fp16 = add(x = out_9_cast_fp16, y = input_57_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor input_69_cast_fp16 = relu(x = input_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; tensor input_71_pad_type_0 = const()[name = tensor("input_71_pad_type_0"), val = tensor("custom")]; tensor input_71_pad_0 = const()[name = tensor("input_71_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_71_strides_0 = const()[name = tensor("input_71_strides_0"), val = tensor([1, 1])]; @@ -126,8 +126,8 @@ program(1.0) tensor input_71_groups_0 = const()[name = tensor("input_71_groups_0"), val = tensor(1)]; tensor const_29_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234368))), name = tensor("const_29_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_30_to_fp16 = const()[name = tensor("const_30_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234944)))]; - tensor input_73_cast_fp16 = conv(bias = const_30_to_fp16, dilations = input_71_dilations_0, groups = input_71_groups_0, pad = input_71_pad_0, pad_type = input_71_pad_type_0, strides = input_71_strides_0, weight = const_29_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("input_73_cast_fp16")]; - tensor input_75_cast_fp16 = relu(x = input_73_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor input_73_cast_fp16 = conv(bias = const_30_to_fp16, dilations = input_71_dilations_0, groups = input_71_groups_0, pad = input_71_pad_0, pad_type = input_71_pad_type_0, strides = input_71_strides_0, weight = const_29_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor input_75_cast_fp16 = relu(x = input_73_cast_fp16)[name = tensor("input_75_cast_fp16")]; tensor input_77_pad_type_0 = const()[name = tensor("input_77_pad_type_0"), val = tensor("custom")]; tensor input_77_pad_0 = const()[name = tensor("input_77_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_77_strides_0 = const()[name = tensor("input_77_strides_0"), val = tensor([1, 1])]; @@ -135,9 +135,9 @@ program(1.0) tensor input_77_groups_0 = const()[name = tensor("input_77_groups_0"), val = tensor(1)]; tensor const_31_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272064))), name = tensor("const_31_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_32_to_fp16 = const()[name = tensor("const_32_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272640)))]; - tensor out_11_cast_fp16 = conv(bias = const_32_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = const_31_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor("out_11_cast_fp16")]; - tensor input_79_cast_fp16 = add(x = out_11_cast_fp16, y = input_69_cast_fp16)[name = tensor("input_79_cast_fp16")]; - tensor input_81_cast_fp16 = relu(x = input_79_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor out_11_cast_fp16 = conv(bias = const_32_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = const_31_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor("out_11_cast_fp16")]; + tensor input_79_cast_fp16 = add(x = out_11_cast_fp16, y = input_69_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor input_81_cast_fp16 = relu(x = input_79_cast_fp16)[name = tensor("input_81_cast_fp16")]; tensor input_83_pad_type_0 = const()[name = tensor("input_83_pad_type_0"), val = tensor("custom")]; tensor input_83_pad_0 = const()[name = tensor("input_83_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_83_strides_0 = const()[name = tensor("input_83_strides_0"), val = tensor([1, 1])]; @@ -145,8 +145,8 @@ program(1.0) tensor input_83_groups_0 = const()[name = tensor("input_83_groups_0"), val = tensor(1)]; tensor const_33_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(309760))), name = tensor("const_33_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_34_to_fp16 = const()[name = tensor("const_34_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310336)))]; - tensor input_85_cast_fp16 = conv(bias = const_34_to_fp16, dilations = input_83_dilations_0, groups = input_83_groups_0, pad = input_83_pad_0, pad_type = input_83_pad_type_0, strides = input_83_strides_0, weight = const_33_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor("input_85_cast_fp16")]; - tensor input_87_cast_fp16 = relu(x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor input_85_cast_fp16 = conv(bias = const_34_to_fp16, dilations = input_83_dilations_0, groups = input_83_groups_0, pad = input_83_pad_0, pad_type = input_83_pad_type_0, strides = input_83_strides_0, weight = const_33_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor input_87_cast_fp16 = relu(x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; tensor input_89_pad_type_0 = const()[name = tensor("input_89_pad_type_0"), val = tensor("custom")]; tensor input_89_pad_0 = const()[name = tensor("input_89_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_89_strides_0 = const()[name = tensor("input_89_strides_0"), val = tensor([1, 1])]; @@ -154,9 +154,9 @@ program(1.0) tensor input_89_groups_0 = const()[name = tensor("input_89_groups_0"), val = tensor(1)]; tensor const_35_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347456))), name = tensor("const_35_to_fp16_palettized"), shape = tensor([64, 64, 3, 3])]; tensor const_36_to_fp16 = const()[name = tensor("const_36_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348032)))]; - tensor out_13_cast_fp16 = conv(bias = const_36_to_fp16, dilations = input_89_dilations_0, groups = input_89_groups_0, pad = input_89_pad_0, pad_type = input_89_pad_type_0, strides = input_89_strides_0, weight = const_35_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor("out_13_cast_fp16")]; - tensor input_91_cast_fp16 = add(x = out_13_cast_fp16, y = input_81_cast_fp16)[name = tensor("input_91_cast_fp16")]; - tensor input_93_cast_fp16 = relu(x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor out_13_cast_fp16 = conv(bias = const_36_to_fp16, dilations = input_89_dilations_0, groups = input_89_groups_0, pad = input_89_pad_0, pad_type = input_89_pad_type_0, strides = input_89_strides_0, weight = const_35_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor("out_13_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = out_13_cast_fp16, y = input_81_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor input_93_cast_fp16 = relu(x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; tensor input_95_pad_type_0 = const()[name = tensor("input_95_pad_type_0"), val = tensor("custom")]; tensor input_95_pad_0 = const()[name = tensor("input_95_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_95_strides_0 = const()[name = tensor("input_95_strides_0"), val = tensor([2, 2])]; @@ -164,8 +164,8 @@ program(1.0) tensor input_95_groups_0 = const()[name = tensor("input_95_groups_0"), val = tensor(1)]; tensor const_37_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(348224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422016))), name = tensor("const_37_to_fp16_palettized"), shape = tensor([128, 64, 3, 3])]; tensor const_38_to_fp16 = const()[name = tensor("const_38_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422592)))]; - tensor input_97_cast_fp16 = conv(bias = const_38_to_fp16, dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = const_37_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("input_97_cast_fp16")]; - tensor input_99_cast_fp16 = relu(x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor input_97_cast_fp16 = conv(bias = const_38_to_fp16, dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = const_37_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor input_99_cast_fp16 = relu(x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; tensor input_101_pad_type_0 = const()[name = tensor("input_101_pad_type_0"), val = tensor("custom")]; tensor input_101_pad_0 = const()[name = tensor("input_101_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_101_strides_0 = const()[name = tensor("input_101_strides_0"), val = tensor([1, 1])]; @@ -173,7 +173,7 @@ program(1.0) tensor input_101_groups_0 = const()[name = tensor("input_101_groups_0"), val = tensor(1)]; tensor const_39_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570432))), name = tensor("const_39_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_40_to_fp16 = const()[name = tensor("const_40_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571008)))]; - tensor out_15_cast_fp16 = conv(bias = const_40_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_39_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor("out_15_cast_fp16")]; + tensor out_15_cast_fp16 = conv(bias = const_40_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_39_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor("out_15_cast_fp16")]; tensor input_103_pad_type_0 = const()[name = tensor("input_103_pad_type_0"), val = tensor("valid")]; tensor input_103_strides_0 = const()[name = tensor("input_103_strides_0"), val = tensor([2, 2])]; tensor input_103_pad_0 = const()[name = tensor("input_103_pad_0"), val = tensor([0, 0, 0, 0])]; @@ -181,9 +181,9 @@ program(1.0) tensor input_103_groups_0 = const()[name = tensor("input_103_groups_0"), val = tensor(1)]; tensor const_41_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(579584))), name = tensor("const_41_to_fp16_palettized"), shape = tensor([128, 64, 1, 1])]; tensor const_42_to_fp16 = const()[name = tensor("const_42_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580160)))]; - tensor var_367_cast_fp16 = conv(bias = const_42_to_fp16, dilations = input_103_dilations_0, groups = input_103_groups_0, pad = input_103_pad_0, pad_type = input_103_pad_type_0, strides = input_103_strides_0, weight = const_41_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("op_367_cast_fp16")]; - tensor input_105_cast_fp16 = add(x = out_15_cast_fp16, y = var_367_cast_fp16)[name = tensor("input_105_cast_fp16")]; - tensor input_107_cast_fp16 = relu(x = input_105_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor var_367_cast_fp16 = conv(bias = const_42_to_fp16, dilations = input_103_dilations_0, groups = input_103_groups_0, pad = input_103_pad_0, pad_type = input_103_pad_type_0, strides = input_103_strides_0, weight = const_41_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("op_367_cast_fp16")]; + tensor input_105_cast_fp16 = add(x = out_15_cast_fp16, y = var_367_cast_fp16)[name = tensor("input_105_cast_fp16")]; + tensor input_107_cast_fp16 = relu(x = input_105_cast_fp16)[name = tensor("input_107_cast_fp16")]; tensor input_109_pad_type_0 = const()[name = tensor("input_109_pad_type_0"), val = tensor("custom")]; tensor input_109_pad_0 = const()[name = tensor("input_109_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_109_strides_0 = const()[name = tensor("input_109_strides_0"), val = tensor([1, 1])]; @@ -191,8 +191,8 @@ program(1.0) tensor input_109_groups_0 = const()[name = tensor("input_109_groups_0"), val = tensor(1)]; tensor const_43_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728000))), name = tensor("const_43_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_44_to_fp16 = const()[name = tensor("const_44_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728576)))]; - tensor input_111_cast_fp16 = conv(bias = const_44_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = const_43_to_fp16_palettized, x = input_107_cast_fp16)[name = tensor("input_111_cast_fp16")]; - tensor input_113_cast_fp16 = relu(x = input_111_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor input_111_cast_fp16 = conv(bias = const_44_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = const_43_to_fp16_palettized, x = input_107_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor input_113_cast_fp16 = relu(x = input_111_cast_fp16)[name = tensor("input_113_cast_fp16")]; tensor input_115_pad_type_0 = const()[name = tensor("input_115_pad_type_0"), val = tensor("custom")]; tensor input_115_pad_0 = const()[name = tensor("input_115_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_115_strides_0 = const()[name = tensor("input_115_strides_0"), val = tensor([1, 1])]; @@ -200,9 +200,9 @@ program(1.0) tensor input_115_groups_0 = const()[name = tensor("input_115_groups_0"), val = tensor(1)]; tensor const_45_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(728896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(876416))), name = tensor("const_45_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_46_to_fp16 = const()[name = tensor("const_46_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(876992)))]; - tensor out_17_cast_fp16 = conv(bias = const_46_to_fp16, dilations = input_115_dilations_0, groups = input_115_groups_0, pad = input_115_pad_0, pad_type = input_115_pad_type_0, strides = input_115_strides_0, weight = const_45_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("out_17_cast_fp16")]; - tensor input_117_cast_fp16 = add(x = out_17_cast_fp16, y = input_107_cast_fp16)[name = tensor("input_117_cast_fp16")]; - tensor input_119_cast_fp16 = relu(x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor out_17_cast_fp16 = conv(bias = const_46_to_fp16, dilations = input_115_dilations_0, groups = input_115_groups_0, pad = input_115_pad_0, pad_type = input_115_pad_type_0, strides = input_115_strides_0, weight = const_45_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("out_17_cast_fp16")]; + tensor input_117_cast_fp16 = add(x = out_17_cast_fp16, y = input_107_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor input_119_cast_fp16 = relu(x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; tensor input_121_pad_type_0 = const()[name = tensor("input_121_pad_type_0"), val = tensor("custom")]; tensor input_121_pad_0 = const()[name = tensor("input_121_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_121_strides_0 = const()[name = tensor("input_121_strides_0"), val = tensor([1, 1])]; @@ -210,8 +210,8 @@ program(1.0) tensor input_121_groups_0 = const()[name = tensor("input_121_groups_0"), val = tensor(1)]; tensor const_47_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(877312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024832))), name = tensor("const_47_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_48_to_fp16 = const()[name = tensor("const_48_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1025408)))]; - tensor input_123_cast_fp16 = conv(bias = const_48_to_fp16, dilations = input_121_dilations_0, groups = input_121_groups_0, pad = input_121_pad_0, pad_type = input_121_pad_type_0, strides = input_121_strides_0, weight = const_47_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor("input_123_cast_fp16")]; - tensor input_125_cast_fp16 = relu(x = input_123_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor input_123_cast_fp16 = conv(bias = const_48_to_fp16, dilations = input_121_dilations_0, groups = input_121_groups_0, pad = input_121_pad_0, pad_type = input_121_pad_type_0, strides = input_121_strides_0, weight = const_47_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor input_125_cast_fp16 = relu(x = input_123_cast_fp16)[name = tensor("input_125_cast_fp16")]; tensor input_127_pad_type_0 = const()[name = tensor("input_127_pad_type_0"), val = tensor("custom")]; tensor input_127_pad_0 = const()[name = tensor("input_127_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_127_strides_0 = const()[name = tensor("input_127_strides_0"), val = tensor([1, 1])]; @@ -219,9 +219,9 @@ program(1.0) tensor input_127_groups_0 = const()[name = tensor("input_127_groups_0"), val = tensor(1)]; tensor const_49_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1025728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1173248))), name = tensor("const_49_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_50_to_fp16 = const()[name = tensor("const_50_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1173824)))]; - tensor out_19_cast_fp16 = conv(bias = const_50_to_fp16, dilations = input_127_dilations_0, groups = input_127_groups_0, pad = input_127_pad_0, pad_type = input_127_pad_type_0, strides = input_127_strides_0, weight = const_49_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("out_19_cast_fp16")]; - tensor input_129_cast_fp16 = add(x = out_19_cast_fp16, y = input_119_cast_fp16)[name = tensor("input_129_cast_fp16")]; - tensor input_131_cast_fp16 = relu(x = input_129_cast_fp16)[name = tensor("input_131_cast_fp16")]; + tensor out_19_cast_fp16 = conv(bias = const_50_to_fp16, dilations = input_127_dilations_0, groups = input_127_groups_0, pad = input_127_pad_0, pad_type = input_127_pad_type_0, strides = input_127_strides_0, weight = const_49_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("out_19_cast_fp16")]; + tensor input_129_cast_fp16 = add(x = out_19_cast_fp16, y = input_119_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor input_131_cast_fp16 = relu(x = input_129_cast_fp16)[name = tensor("input_131_cast_fp16")]; tensor input_133_pad_type_0 = const()[name = tensor("input_133_pad_type_0"), val = tensor("custom")]; tensor input_133_pad_0 = const()[name = tensor("input_133_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_133_strides_0 = const()[name = tensor("input_133_strides_0"), val = tensor([1, 1])]; @@ -229,8 +229,8 @@ program(1.0) tensor input_133_groups_0 = const()[name = tensor("input_133_groups_0"), val = tensor(1)]; tensor const_51_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1174144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1321664))), name = tensor("const_51_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_52_to_fp16 = const()[name = tensor("const_52_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322240)))]; - tensor input_135_cast_fp16 = conv(bias = const_52_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = const_51_to_fp16_palettized, x = input_131_cast_fp16)[name = tensor("input_135_cast_fp16")]; - tensor input_137_cast_fp16 = relu(x = input_135_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor input_135_cast_fp16 = conv(bias = const_52_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = const_51_to_fp16_palettized, x = input_131_cast_fp16)[name = tensor("input_135_cast_fp16")]; + tensor input_137_cast_fp16 = relu(x = input_135_cast_fp16)[name = tensor("input_137_cast_fp16")]; tensor input_139_pad_type_0 = const()[name = tensor("input_139_pad_type_0"), val = tensor("custom")]; tensor input_139_pad_0 = const()[name = tensor("input_139_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_139_strides_0 = const()[name = tensor("input_139_strides_0"), val = tensor([1, 1])]; @@ -238,9 +238,9 @@ program(1.0) tensor input_139_groups_0 = const()[name = tensor("input_139_groups_0"), val = tensor(1)]; tensor const_53_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1470080))), name = tensor("const_53_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_54_to_fp16 = const()[name = tensor("const_54_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1470656)))]; - tensor out_21_cast_fp16 = conv(bias = const_54_to_fp16, dilations = input_139_dilations_0, groups = input_139_groups_0, pad = input_139_pad_0, pad_type = input_139_pad_type_0, strides = input_139_strides_0, weight = const_53_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("out_21_cast_fp16")]; - tensor input_141_cast_fp16 = add(x = out_21_cast_fp16, y = input_131_cast_fp16)[name = tensor("input_141_cast_fp16")]; - tensor input_143_cast_fp16 = relu(x = input_141_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor out_21_cast_fp16 = conv(bias = const_54_to_fp16, dilations = input_139_dilations_0, groups = input_139_groups_0, pad = input_139_pad_0, pad_type = input_139_pad_type_0, strides = input_139_strides_0, weight = const_53_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("out_21_cast_fp16")]; + tensor input_141_cast_fp16 = add(x = out_21_cast_fp16, y = input_131_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor input_143_cast_fp16 = relu(x = input_141_cast_fp16)[name = tensor("input_143_cast_fp16")]; tensor input_145_pad_type_0 = const()[name = tensor("input_145_pad_type_0"), val = tensor("custom")]; tensor input_145_pad_0 = const()[name = tensor("input_145_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_145_strides_0 = const()[name = tensor("input_145_strides_0"), val = tensor([1, 1])]; @@ -248,8 +248,8 @@ program(1.0) tensor input_145_groups_0 = const()[name = tensor("input_145_groups_0"), val = tensor(1)]; tensor const_55_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1470976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1618496))), name = tensor("const_55_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_56_to_fp16 = const()[name = tensor("const_56_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1619072)))]; - tensor input_147_cast_fp16 = conv(bias = const_56_to_fp16, dilations = input_145_dilations_0, groups = input_145_groups_0, pad = input_145_pad_0, pad_type = input_145_pad_type_0, strides = input_145_strides_0, weight = const_55_to_fp16_palettized, x = input_143_cast_fp16)[name = tensor("input_147_cast_fp16")]; - tensor input_149_cast_fp16 = relu(x = input_147_cast_fp16)[name = tensor("input_149_cast_fp16")]; + tensor input_147_cast_fp16 = conv(bias = const_56_to_fp16, dilations = input_145_dilations_0, groups = input_145_groups_0, pad = input_145_pad_0, pad_type = input_145_pad_type_0, strides = input_145_strides_0, weight = const_55_to_fp16_palettized, x = input_143_cast_fp16)[name = tensor("input_147_cast_fp16")]; + tensor input_149_cast_fp16 = relu(x = input_147_cast_fp16)[name = tensor("input_149_cast_fp16")]; tensor input_151_pad_type_0 = const()[name = tensor("input_151_pad_type_0"), val = tensor("custom")]; tensor input_151_pad_0 = const()[name = tensor("input_151_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_151_strides_0 = const()[name = tensor("input_151_strides_0"), val = tensor([1, 1])]; @@ -257,9 +257,9 @@ program(1.0) tensor input_151_groups_0 = const()[name = tensor("input_151_groups_0"), val = tensor(1)]; tensor const_57_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1619392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1766912))), name = tensor("const_57_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_58_to_fp16 = const()[name = tensor("const_58_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1767488)))]; - tensor out_23_cast_fp16 = conv(bias = const_58_to_fp16, dilations = input_151_dilations_0, groups = input_151_groups_0, pad = input_151_pad_0, pad_type = input_151_pad_type_0, strides = input_151_strides_0, weight = const_57_to_fp16_palettized, x = input_149_cast_fp16)[name = tensor("out_23_cast_fp16")]; - tensor input_153_cast_fp16 = add(x = out_23_cast_fp16, y = input_143_cast_fp16)[name = tensor("input_153_cast_fp16")]; - tensor input_155_cast_fp16 = relu(x = input_153_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor out_23_cast_fp16 = conv(bias = const_58_to_fp16, dilations = input_151_dilations_0, groups = input_151_groups_0, pad = input_151_pad_0, pad_type = input_151_pad_type_0, strides = input_151_strides_0, weight = const_57_to_fp16_palettized, x = input_149_cast_fp16)[name = tensor("out_23_cast_fp16")]; + tensor input_153_cast_fp16 = add(x = out_23_cast_fp16, y = input_143_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor input_155_cast_fp16 = relu(x = input_153_cast_fp16)[name = tensor("input_155_cast_fp16")]; tensor input_157_pad_type_0 = const()[name = tensor("input_157_pad_type_0"), val = tensor("custom")]; tensor input_157_pad_0 = const()[name = tensor("input_157_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_157_strides_0 = const()[name = tensor("input_157_strides_0"), val = tensor([1, 1])]; @@ -267,8 +267,8 @@ program(1.0) tensor input_157_groups_0 = const()[name = tensor("input_157_groups_0"), val = tensor(1)]; tensor const_59_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1767808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1915328))), name = tensor("const_59_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_60_to_fp16 = const()[name = tensor("const_60_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1915904)))]; - tensor input_159_cast_fp16 = conv(bias = const_60_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = const_59_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor("input_159_cast_fp16")]; - tensor input_161_cast_fp16 = relu(x = input_159_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor input_159_cast_fp16 = conv(bias = const_60_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = const_59_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor("input_159_cast_fp16")]; + tensor input_161_cast_fp16 = relu(x = input_159_cast_fp16)[name = tensor("input_161_cast_fp16")]; tensor input_163_pad_type_0 = const()[name = tensor("input_163_pad_type_0"), val = tensor("custom")]; tensor input_163_pad_0 = const()[name = tensor("input_163_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_163_strides_0 = const()[name = tensor("input_163_strides_0"), val = tensor([1, 1])]; @@ -276,9 +276,9 @@ program(1.0) tensor input_163_groups_0 = const()[name = tensor("input_163_groups_0"), val = tensor(1)]; tensor const_61_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1916224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2063744))), name = tensor("const_61_to_fp16_palettized"), shape = tensor([128, 128, 3, 3])]; tensor const_62_to_fp16 = const()[name = tensor("const_62_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2064320)))]; - tensor out_25_cast_fp16 = conv(bias = const_62_to_fp16, dilations = input_163_dilations_0, groups = input_163_groups_0, pad = input_163_pad_0, pad_type = input_163_pad_type_0, strides = input_163_strides_0, weight = const_61_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor("out_25_cast_fp16")]; - tensor input_165_cast_fp16 = add(x = out_25_cast_fp16, y = input_155_cast_fp16)[name = tensor("input_165_cast_fp16")]; - tensor input_167_cast_fp16 = relu(x = input_165_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor out_25_cast_fp16 = conv(bias = const_62_to_fp16, dilations = input_163_dilations_0, groups = input_163_groups_0, pad = input_163_pad_0, pad_type = input_163_pad_type_0, strides = input_163_strides_0, weight = const_61_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor("out_25_cast_fp16")]; + tensor input_165_cast_fp16 = add(x = out_25_cast_fp16, y = input_155_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor input_167_cast_fp16 = relu(x = input_165_cast_fp16)[name = tensor("input_167_cast_fp16")]; tensor input_169_pad_type_0 = const()[name = tensor("input_169_pad_type_0"), val = tensor("custom")]; tensor input_169_pad_0 = const()[name = tensor("input_169_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_169_strides_0 = const()[name = tensor("input_169_strides_0"), val = tensor([2, 2])]; @@ -286,8 +286,8 @@ program(1.0) tensor input_169_groups_0 = const()[name = tensor("input_169_groups_0"), val = tensor(1)]; tensor const_63_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2064640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2359616))), name = tensor("const_63_to_fp16_palettized"), shape = tensor([256, 128, 3, 3])]; tensor const_64_to_fp16 = const()[name = tensor("const_64_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2360192)))]; - tensor input_171_cast_fp16 = conv(bias = const_64_to_fp16, dilations = input_169_dilations_0, groups = input_169_groups_0, pad = input_169_pad_0, pad_type = input_169_pad_type_0, strides = input_169_strides_0, weight = const_63_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("input_171_cast_fp16")]; - tensor input_173_cast_fp16 = relu(x = input_171_cast_fp16)[name = tensor("input_173_cast_fp16")]; + tensor input_171_cast_fp16 = conv(bias = const_64_to_fp16, dilations = input_169_dilations_0, groups = input_169_groups_0, pad = input_169_pad_0, pad_type = input_169_pad_type_0, strides = input_169_strides_0, weight = const_63_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("input_171_cast_fp16")]; + tensor input_173_cast_fp16 = relu(x = input_171_cast_fp16)[name = tensor("input_173_cast_fp16")]; tensor input_175_pad_type_0 = const()[name = tensor("input_175_pad_type_0"), val = tensor("custom")]; tensor input_175_pad_0 = const()[name = tensor("input_175_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_175_strides_0 = const()[name = tensor("input_175_strides_0"), val = tensor([1, 1])]; @@ -295,7 +295,7 @@ program(1.0) tensor input_175_groups_0 = const()[name = tensor("input_175_groups_0"), val = tensor(1)]; tensor const_65_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2360768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2950656))), name = tensor("const_65_to_fp16_palettized"), shape = tensor([256, 256, 3, 3])]; tensor const_66_to_fp16 = const()[name = tensor("const_66_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2951232)))]; - tensor out_27_cast_fp16 = conv(bias = const_66_to_fp16, dilations = input_175_dilations_0, groups = input_175_groups_0, pad = input_175_pad_0, pad_type = input_175_pad_type_0, strides = input_175_strides_0, weight = const_65_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("out_27_cast_fp16")]; + tensor out_27_cast_fp16 = conv(bias = const_66_to_fp16, dilations = input_175_dilations_0, groups = input_175_groups_0, pad = input_175_pad_0, pad_type = input_175_pad_type_0, strides = input_175_strides_0, weight = const_65_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("out_27_cast_fp16")]; tensor input_177_pad_type_0 = const()[name = tensor("input_177_pad_type_0"), val = tensor("valid")]; tensor input_177_strides_0 = const()[name = tensor("input_177_strides_0"), val = tensor([2, 2])]; tensor input_177_pad_0 = const()[name = tensor("input_177_pad_0"), val = tensor([0, 0, 0, 0])]; @@ -303,9 +303,9 @@ program(1.0) tensor input_177_groups_0 = const()[name = tensor("input_177_groups_0"), val = tensor(1)]; tensor const_67_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2951808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2984640))), name = tensor("const_67_to_fp16_palettized"), shape = tensor([256, 128, 1, 1])]; tensor const_68_to_fp16 = const()[name = tensor("const_68_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2985216)))]; - tensor var_558_cast_fp16 = conv(bias = const_68_to_fp16, dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = const_67_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("op_558_cast_fp16")]; - tensor input_179_cast_fp16 = add(x = out_27_cast_fp16, y = var_558_cast_fp16)[name = tensor("input_179_cast_fp16")]; - tensor input_181_cast_fp16 = relu(x = input_179_cast_fp16)[name = tensor("input_181_cast_fp16")]; + tensor var_558_cast_fp16 = conv(bias = const_68_to_fp16, dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = const_67_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("op_558_cast_fp16")]; + tensor input_179_cast_fp16 = add(x = out_27_cast_fp16, y = var_558_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor input_181_cast_fp16 = relu(x = input_179_cast_fp16)[name = tensor("input_181_cast_fp16")]; tensor input_183_pad_type_0 = const()[name = tensor("input_183_pad_type_0"), val = tensor("custom")]; tensor input_183_pad_0 = const()[name = tensor("input_183_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_183_strides_0 = const()[name = tensor("input_183_strides_0"), val = tensor([1, 1])]; @@ -313,8 +313,8 @@ program(1.0) tensor input_183_groups_0 = const()[name = tensor("input_183_groups_0"), val = tensor(1)]; tensor const_69_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2985792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3575680))), name = tensor("const_69_to_fp16_palettized"), shape = tensor([256, 256, 3, 3])]; tensor const_70_to_fp16 = const()[name = tensor("const_70_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3576256)))]; - tensor input_185_cast_fp16 = conv(bias = const_70_to_fp16, dilations = input_183_dilations_0, groups = input_183_groups_0, pad = input_183_pad_0, pad_type = input_183_pad_type_0, strides = input_183_strides_0, weight = const_69_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor("input_185_cast_fp16")]; - tensor input_187_cast_fp16 = relu(x = input_185_cast_fp16)[name = tensor("input_187_cast_fp16")]; + tensor input_185_cast_fp16 = conv(bias = const_70_to_fp16, dilations = input_183_dilations_0, groups = input_183_groups_0, pad = input_183_pad_0, pad_type = input_183_pad_type_0, strides = input_183_strides_0, weight = const_69_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor input_187_cast_fp16 = relu(x = input_185_cast_fp16)[name = tensor("input_187_cast_fp16")]; tensor input_189_pad_type_0 = const()[name = tensor("input_189_pad_type_0"), val = tensor("custom")]; tensor input_189_pad_0 = const()[name = tensor("input_189_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_189_strides_0 = const()[name = tensor("input_189_strides_0"), val = tensor([1, 1])]; @@ -322,9 +322,9 @@ program(1.0) tensor input_189_groups_0 = const()[name = tensor("input_189_groups_0"), val = tensor(1)]; tensor const_71_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3576832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4166720))), name = tensor("const_71_to_fp16_palettized"), shape = tensor([256, 256, 3, 3])]; tensor const_72_to_fp16 = const()[name = tensor("const_72_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4167296)))]; - tensor out_29_cast_fp16 = conv(bias = const_72_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = const_71_to_fp16_palettized, x = input_187_cast_fp16)[name = tensor("out_29_cast_fp16")]; - tensor input_191_cast_fp16 = add(x = out_29_cast_fp16, y = input_181_cast_fp16)[name = tensor("input_191_cast_fp16")]; - tensor input_193_cast_fp16 = relu(x = input_191_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor out_29_cast_fp16 = conv(bias = const_72_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = const_71_to_fp16_palettized, x = input_187_cast_fp16)[name = tensor("out_29_cast_fp16")]; + tensor input_191_cast_fp16 = add(x = out_29_cast_fp16, y = input_181_cast_fp16)[name = tensor("input_191_cast_fp16")]; + tensor input_193_cast_fp16 = relu(x = input_191_cast_fp16)[name = tensor("input_193_cast_fp16")]; tensor input_195_pad_type_0 = const()[name = tensor("input_195_pad_type_0"), val = tensor("custom")]; tensor input_195_pad_0 = const()[name = tensor("input_195_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_195_strides_0 = const()[name = tensor("input_195_strides_0"), val = tensor([1, 1])]; @@ -332,8 +332,8 @@ program(1.0) tensor input_195_groups_0 = const()[name = tensor("input_195_groups_0"), val = tensor(1)]; tensor const_73_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4167872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4757760))), name = tensor("const_73_to_fp16_palettized"), shape = tensor([256, 256, 3, 3])]; tensor const_74_to_fp16 = const()[name = tensor("const_74_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4758336)))]; - tensor input_197_cast_fp16 = conv(bias = const_74_to_fp16, dilations = input_195_dilations_0, groups = input_195_groups_0, pad = input_195_pad_0, pad_type = input_195_pad_type_0, strides = input_195_strides_0, weight = const_73_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("input_197_cast_fp16")]; - tensor input_199_cast_fp16 = relu(x = input_197_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor input_197_cast_fp16 = conv(bias = const_74_to_fp16, dilations = input_195_dilations_0, groups = input_195_groups_0, pad = input_195_pad_0, pad_type = input_195_pad_type_0, strides = input_195_strides_0, weight = const_73_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("input_197_cast_fp16")]; + tensor input_199_cast_fp16 = relu(x = input_197_cast_fp16)[name = tensor("input_199_cast_fp16")]; tensor input_201_pad_type_0 = const()[name = tensor("input_201_pad_type_0"), val = tensor("custom")]; tensor input_201_pad_0 = const()[name = tensor("input_201_pad_0"), val = tensor([1, 1, 1, 1])]; tensor input_201_strides_0 = const()[name = tensor("input_201_strides_0"), val = tensor([1, 1])]; @@ -341,44 +341,44 @@ program(1.0) tensor input_201_groups_0 = const()[name = tensor("input_201_groups_0"), val = tensor(1)]; tensor const_75_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4758912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5348800))), name = tensor("const_75_to_fp16_palettized"), shape = tensor([256, 256, 3, 3])]; tensor const_76_to_fp16 = const()[name = tensor("const_76_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5349376)))]; - tensor out_cast_fp16 = conv(bias = const_76_to_fp16, dilations = input_201_dilations_0, groups = input_201_groups_0, pad = input_201_pad_0, pad_type = input_201_pad_type_0, strides = input_201_strides_0, weight = const_75_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor("out_cast_fp16")]; - tensor input_203_cast_fp16 = add(x = out_cast_fp16, y = input_193_cast_fp16)[name = tensor("input_203_cast_fp16")]; - tensor x_cast_fp16 = relu(x = input_203_cast_fp16)[name = tensor("x_cast_fp16")]; - tensor var_637 = const()[name = tensor("op_637"), val = tensor([1, 2560, 125])]; - tensor sequences_cast_fp16 = reshape(shape = var_637, x = x_cast_fp16)[name = tensor("sequences_cast_fp16")]; + tensor out_cast_fp16 = conv(bias = const_76_to_fp16, dilations = input_201_dilations_0, groups = input_201_groups_0, pad = input_201_pad_0, pad_type = input_201_pad_type_0, strides = input_201_strides_0, weight = const_75_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor("out_cast_fp16")]; + tensor input_203_cast_fp16 = add(x = out_cast_fp16, y = input_193_cast_fp16)[name = tensor("input_203_cast_fp16")]; + tensor x_cast_fp16 = relu(x = input_203_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor var_637 = const()[name = tensor("op_637"), val = tensor([1, 2560, 375])]; + tensor sequences_cast_fp16 = reshape(shape = var_637, x = x_cast_fp16)[name = tensor("sequences_cast_fp16")]; tensor expand_dims_0_axes_0 = const()[name = tensor("expand_dims_0_axes_0"), val = tensor([3])]; - tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = speaker_masks)[name = tensor("expand_dims_0_cast_fp16")]; - tensor upsample_nearest_neighbor_0_scale_factor_height_0 = const()[name = tensor("upsample_nearest_neighbor_0_scale_factor_height_0"), val = tensor(0x1.b2a2a4p-3)]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = speaker_masks)[name = tensor("expand_dims_0_cast_fp16")]; + tensor upsample_nearest_neighbor_0_scale_factor_height_0 = const()[name = tensor("upsample_nearest_neighbor_0_scale_factor_height_0"), val = tensor(0x1.b2a296p-3)]; tensor upsample_nearest_neighbor_0_scale_factor_width_0 = const()[name = tensor("upsample_nearest_neighbor_0_scale_factor_width_0"), val = tensor(0x1p+0)]; - tensor upsample_nearest_neighbor_0_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = upsample_nearest_neighbor_0_scale_factor_height_0, scale_factor_width = upsample_nearest_neighbor_0_scale_factor_width_0, x = expand_dims_0_cast_fp16)[name = tensor("upsample_nearest_neighbor_0_cast_fp16")]; + tensor upsample_nearest_neighbor_0_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = upsample_nearest_neighbor_0_scale_factor_height_0, scale_factor_width = upsample_nearest_neighbor_0_scale_factor_width_0, x = expand_dims_0_cast_fp16)[name = tensor("upsample_nearest_neighbor_0_cast_fp16")]; tensor weights_1_axes_0 = const()[name = tensor("weights_1_axes_0"), val = tensor([3])]; - tensor weights_1_cast_fp16 = squeeze(axes = weights_1_axes_0, x = upsample_nearest_neighbor_0_cast_fp16)[name = tensor("weights_1_cast_fp16")]; + tensor weights_1_cast_fp16 = squeeze(axes = weights_1_axes_0, x = upsample_nearest_neighbor_0_cast_fp16)[name = tensor("weights_1_cast_fp16")]; tensor var_643_begin_0 = const()[name = tensor("op_643_begin_0"), val = tensor([0, 0, 0])]; - tensor var_643_end_0 = const()[name = tensor("op_643_end_0"), val = tensor([1, 1, 125])]; + tensor var_643_end_0 = const()[name = tensor("op_643_end_0"), val = tensor([1, 1, 375])]; tensor var_643_end_mask_0 = const()[name = tensor("op_643_end_mask_0"), val = tensor([true, false, true])]; tensor var_643_squeeze_mask_0 = const()[name = tensor("op_643_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_643_cast_fp16 = slice_by_index(begin = var_643_begin_0, end = var_643_end_0, end_mask = var_643_end_mask_0, squeeze_mask = var_643_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_643_cast_fp16")]; + tensor var_643_cast_fp16 = slice_by_index(begin = var_643_begin_0, end = var_643_end_0, end_mask = var_643_end_mask_0, squeeze_mask = var_643_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_643_cast_fp16")]; tensor weights_5_axes_0 = const()[name = tensor("weights_5_axes_0"), val = tensor([1])]; - tensor weights_5_cast_fp16 = expand_dims(axes = weights_5_axes_0, x = var_643_cast_fp16)[name = tensor("weights_5_cast_fp16")]; + tensor weights_5_cast_fp16 = expand_dims(axes = weights_5_axes_0, x = var_643_cast_fp16)[name = tensor("weights_5_cast_fp16")]; tensor var_647_axes_0 = const()[name = tensor("op_647_axes_0"), val = tensor([2])]; tensor var_647_keep_dims_0 = const()[name = tensor("op_647_keep_dims_0"), val = tensor(false)]; tensor var_647_cast_fp16 = reduce_sum(axes = var_647_axes_0, keep_dims = var_647_keep_dims_0, x = weights_5_cast_fp16)[name = tensor("op_647_cast_fp16")]; tensor var_648_to_fp16 = const()[name = tensor("op_648_to_fp16"), val = tensor(0x1p-24)]; tensor v1_1_cast_fp16 = add(x = var_647_cast_fp16, y = var_648_to_fp16)[name = tensor("v1_1_cast_fp16")]; - tensor var_650_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_5_cast_fp16)[name = tensor("op_650_cast_fp16")]; + tensor var_650_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_5_cast_fp16)[name = tensor("op_650_cast_fp16")]; tensor var_652_axes_0 = const()[name = tensor("op_652_axes_0"), val = tensor([2])]; tensor var_652_keep_dims_0 = const()[name = tensor("op_652_keep_dims_0"), val = tensor(false)]; tensor var_652_cast_fp16 = reduce_sum(axes = var_652_axes_0, keep_dims = var_652_keep_dims_0, x = var_650_cast_fp16)[name = tensor("op_652_cast_fp16")]; tensor mean_1_cast_fp16 = real_div(x = var_652_cast_fp16, y = v1_1_cast_fp16)[name = tensor("mean_1_cast_fp16")]; tensor var_654_axes_0 = const()[name = tensor("op_654_axes_0"), val = tensor([2])]; tensor var_654_cast_fp16 = expand_dims(axes = var_654_axes_0, x = mean_1_cast_fp16)[name = tensor("op_654_cast_fp16")]; - tensor var_655_cast_fp16 = sub(x = sequences_cast_fp16, y = var_654_cast_fp16)[name = tensor("op_655_cast_fp16")]; - tensor dx2_1_cast_fp16 = mul(x = var_655_cast_fp16, y = var_655_cast_fp16)[name = tensor("dx2_1_cast_fp16")]; - tensor var_657_cast_fp16 = mul(x = weights_5_cast_fp16, y = weights_5_cast_fp16)[name = tensor("op_657_cast_fp16")]; + tensor var_655_cast_fp16 = sub(x = sequences_cast_fp16, y = var_654_cast_fp16)[name = tensor("op_655_cast_fp16")]; + tensor dx2_1_cast_fp16 = mul(x = var_655_cast_fp16, y = var_655_cast_fp16)[name = tensor("dx2_1_cast_fp16")]; + tensor var_657_cast_fp16 = mul(x = weights_5_cast_fp16, y = weights_5_cast_fp16)[name = tensor("op_657_cast_fp16")]; tensor v2_1_axes_0 = const()[name = tensor("v2_1_axes_0"), val = tensor([2])]; tensor v2_1_keep_dims_0 = const()[name = tensor("v2_1_keep_dims_0"), val = tensor(false)]; tensor v2_1_cast_fp16 = reduce_sum(axes = v2_1_axes_0, keep_dims = v2_1_keep_dims_0, x = var_657_cast_fp16)[name = tensor("v2_1_cast_fp16")]; - tensor var_660_cast_fp16 = mul(x = dx2_1_cast_fp16, y = weights_5_cast_fp16)[name = tensor("op_660_cast_fp16")]; + tensor var_660_cast_fp16 = mul(x = dx2_1_cast_fp16, y = weights_5_cast_fp16)[name = tensor("op_660_cast_fp16")]; tensor var_662_axes_0 = const()[name = tensor("op_662_axes_0"), val = tensor([2])]; tensor var_662_keep_dims_0 = const()[name = tensor("op_662_keep_dims_0"), val = tensor(false)]; tensor var_662_cast_fp16 = reduce_sum(axes = var_662_axes_0, keep_dims = var_662_keep_dims_0, x = var_660_cast_fp16)[name = tensor("op_662_cast_fp16")]; @@ -391,31 +391,31 @@ program(1.0) tensor var_670_interleave_0 = const()[name = tensor("op_670_interleave_0"), val = tensor(false)]; tensor var_670_cast_fp16 = concat(axis = var_72, interleave = var_670_interleave_0, values = (mean_1_cast_fp16, std_1_cast_fp16))[name = tensor("op_670_cast_fp16")]; tensor var_672_begin_0 = const()[name = tensor("op_672_begin_0"), val = tensor([0, 1, 0])]; - tensor var_672_end_0 = const()[name = tensor("op_672_end_0"), val = tensor([1, 2, 125])]; + tensor var_672_end_0 = const()[name = tensor("op_672_end_0"), val = tensor([1, 2, 375])]; tensor var_672_end_mask_0 = const()[name = tensor("op_672_end_mask_0"), val = tensor([true, false, true])]; tensor var_672_squeeze_mask_0 = const()[name = tensor("op_672_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_672_cast_fp16 = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, squeeze_mask = var_672_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_672_cast_fp16")]; + tensor var_672_cast_fp16 = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, squeeze_mask = var_672_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_672_cast_fp16")]; tensor weights_9_axes_0 = const()[name = tensor("weights_9_axes_0"), val = tensor([1])]; - tensor weights_9_cast_fp16 = expand_dims(axes = weights_9_axes_0, x = var_672_cast_fp16)[name = tensor("weights_9_cast_fp16")]; + tensor weights_9_cast_fp16 = expand_dims(axes = weights_9_axes_0, x = var_672_cast_fp16)[name = tensor("weights_9_cast_fp16")]; tensor var_676_axes_0 = const()[name = tensor("op_676_axes_0"), val = tensor([2])]; tensor var_676_keep_dims_0 = const()[name = tensor("op_676_keep_dims_0"), val = tensor(false)]; tensor var_676_cast_fp16 = reduce_sum(axes = var_676_axes_0, keep_dims = var_676_keep_dims_0, x = weights_9_cast_fp16)[name = tensor("op_676_cast_fp16")]; tensor var_677_to_fp16 = const()[name = tensor("op_677_to_fp16"), val = tensor(0x1p-24)]; tensor v1_3_cast_fp16 = add(x = var_676_cast_fp16, y = var_677_to_fp16)[name = tensor("v1_3_cast_fp16")]; - tensor var_679_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_9_cast_fp16)[name = tensor("op_679_cast_fp16")]; + tensor var_679_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_9_cast_fp16)[name = tensor("op_679_cast_fp16")]; tensor var_681_axes_0 = const()[name = tensor("op_681_axes_0"), val = tensor([2])]; tensor var_681_keep_dims_0 = const()[name = tensor("op_681_keep_dims_0"), val = tensor(false)]; tensor var_681_cast_fp16 = reduce_sum(axes = var_681_axes_0, keep_dims = var_681_keep_dims_0, x = var_679_cast_fp16)[name = tensor("op_681_cast_fp16")]; tensor mean_3_cast_fp16 = real_div(x = var_681_cast_fp16, y = v1_3_cast_fp16)[name = tensor("mean_3_cast_fp16")]; tensor var_683_axes_0 = const()[name = tensor("op_683_axes_0"), val = tensor([2])]; tensor var_683_cast_fp16 = expand_dims(axes = var_683_axes_0, x = mean_3_cast_fp16)[name = tensor("op_683_cast_fp16")]; - tensor var_684_cast_fp16 = sub(x = sequences_cast_fp16, y = var_683_cast_fp16)[name = tensor("op_684_cast_fp16")]; - tensor dx2_3_cast_fp16 = mul(x = var_684_cast_fp16, y = var_684_cast_fp16)[name = tensor("dx2_3_cast_fp16")]; - tensor var_686_cast_fp16 = mul(x = weights_9_cast_fp16, y = weights_9_cast_fp16)[name = tensor("op_686_cast_fp16")]; + tensor var_684_cast_fp16 = sub(x = sequences_cast_fp16, y = var_683_cast_fp16)[name = tensor("op_684_cast_fp16")]; + tensor dx2_3_cast_fp16 = mul(x = var_684_cast_fp16, y = var_684_cast_fp16)[name = tensor("dx2_3_cast_fp16")]; + tensor var_686_cast_fp16 = mul(x = weights_9_cast_fp16, y = weights_9_cast_fp16)[name = tensor("op_686_cast_fp16")]; tensor v2_3_axes_0 = const()[name = tensor("v2_3_axes_0"), val = tensor([2])]; tensor v2_3_keep_dims_0 = const()[name = tensor("v2_3_keep_dims_0"), val = tensor(false)]; tensor v2_3_cast_fp16 = reduce_sum(axes = v2_3_axes_0, keep_dims = v2_3_keep_dims_0, x = var_686_cast_fp16)[name = tensor("v2_3_cast_fp16")]; - tensor var_689_cast_fp16 = mul(x = dx2_3_cast_fp16, y = weights_9_cast_fp16)[name = tensor("op_689_cast_fp16")]; + tensor var_689_cast_fp16 = mul(x = dx2_3_cast_fp16, y = weights_9_cast_fp16)[name = tensor("op_689_cast_fp16")]; tensor var_691_axes_0 = const()[name = tensor("op_691_axes_0"), val = tensor([2])]; tensor var_691_keep_dims_0 = const()[name = tensor("op_691_keep_dims_0"), val = tensor(false)]; tensor var_691_cast_fp16 = reduce_sum(axes = var_691_axes_0, keep_dims = var_691_keep_dims_0, x = var_689_cast_fp16)[name = tensor("op_691_cast_fp16")]; @@ -428,31 +428,31 @@ program(1.0) tensor var_699_interleave_0 = const()[name = tensor("op_699_interleave_0"), val = tensor(false)]; tensor var_699_cast_fp16 = concat(axis = var_72, interleave = var_699_interleave_0, values = (mean_3_cast_fp16, std_3_cast_fp16))[name = tensor("op_699_cast_fp16")]; tensor var_701_begin_0 = const()[name = tensor("op_701_begin_0"), val = tensor([0, 2, 0])]; - tensor var_701_end_0 = const()[name = tensor("op_701_end_0"), val = tensor([1, 3, 125])]; + tensor var_701_end_0 = const()[name = tensor("op_701_end_0"), val = tensor([1, 3, 375])]; tensor var_701_end_mask_0 = const()[name = tensor("op_701_end_mask_0"), val = tensor([true, false, true])]; tensor var_701_squeeze_mask_0 = const()[name = tensor("op_701_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_701_cast_fp16 = slice_by_index(begin = var_701_begin_0, end = var_701_end_0, end_mask = var_701_end_mask_0, squeeze_mask = var_701_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_701_cast_fp16")]; + tensor var_701_cast_fp16 = slice_by_index(begin = var_701_begin_0, end = var_701_end_0, end_mask = var_701_end_mask_0, squeeze_mask = var_701_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_701_cast_fp16")]; tensor weights_13_axes_0 = const()[name = tensor("weights_13_axes_0"), val = tensor([1])]; - tensor weights_13_cast_fp16 = expand_dims(axes = weights_13_axes_0, x = var_701_cast_fp16)[name = tensor("weights_13_cast_fp16")]; + tensor weights_13_cast_fp16 = expand_dims(axes = weights_13_axes_0, x = var_701_cast_fp16)[name = tensor("weights_13_cast_fp16")]; tensor var_705_axes_0 = const()[name = tensor("op_705_axes_0"), val = tensor([2])]; tensor var_705_keep_dims_0 = const()[name = tensor("op_705_keep_dims_0"), val = tensor(false)]; tensor var_705_cast_fp16 = reduce_sum(axes = var_705_axes_0, keep_dims = var_705_keep_dims_0, x = weights_13_cast_fp16)[name = tensor("op_705_cast_fp16")]; tensor var_706_to_fp16 = const()[name = tensor("op_706_to_fp16"), val = tensor(0x1p-24)]; tensor v1_5_cast_fp16 = add(x = var_705_cast_fp16, y = var_706_to_fp16)[name = tensor("v1_5_cast_fp16")]; - tensor var_708_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_13_cast_fp16)[name = tensor("op_708_cast_fp16")]; + tensor var_708_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_13_cast_fp16)[name = tensor("op_708_cast_fp16")]; tensor var_710_axes_0 = const()[name = tensor("op_710_axes_0"), val = tensor([2])]; tensor var_710_keep_dims_0 = const()[name = tensor("op_710_keep_dims_0"), val = tensor(false)]; tensor var_710_cast_fp16 = reduce_sum(axes = var_710_axes_0, keep_dims = var_710_keep_dims_0, x = var_708_cast_fp16)[name = tensor("op_710_cast_fp16")]; tensor mean_5_cast_fp16 = real_div(x = var_710_cast_fp16, y = v1_5_cast_fp16)[name = tensor("mean_5_cast_fp16")]; tensor var_712_axes_0 = const()[name = tensor("op_712_axes_0"), val = tensor([2])]; tensor var_712_cast_fp16 = expand_dims(axes = var_712_axes_0, x = mean_5_cast_fp16)[name = tensor("op_712_cast_fp16")]; - tensor var_713_cast_fp16 = sub(x = sequences_cast_fp16, y = var_712_cast_fp16)[name = tensor("op_713_cast_fp16")]; - tensor dx2_5_cast_fp16 = mul(x = var_713_cast_fp16, y = var_713_cast_fp16)[name = tensor("dx2_5_cast_fp16")]; - tensor var_715_cast_fp16 = mul(x = weights_13_cast_fp16, y = weights_13_cast_fp16)[name = tensor("op_715_cast_fp16")]; + tensor var_713_cast_fp16 = sub(x = sequences_cast_fp16, y = var_712_cast_fp16)[name = tensor("op_713_cast_fp16")]; + tensor dx2_5_cast_fp16 = mul(x = var_713_cast_fp16, y = var_713_cast_fp16)[name = tensor("dx2_5_cast_fp16")]; + tensor var_715_cast_fp16 = mul(x = weights_13_cast_fp16, y = weights_13_cast_fp16)[name = tensor("op_715_cast_fp16")]; tensor v2_5_axes_0 = const()[name = tensor("v2_5_axes_0"), val = tensor([2])]; tensor v2_5_keep_dims_0 = const()[name = tensor("v2_5_keep_dims_0"), val = tensor(false)]; tensor v2_5_cast_fp16 = reduce_sum(axes = v2_5_axes_0, keep_dims = v2_5_keep_dims_0, x = var_715_cast_fp16)[name = tensor("v2_5_cast_fp16")]; - tensor var_718_cast_fp16 = mul(x = dx2_5_cast_fp16, y = weights_13_cast_fp16)[name = tensor("op_718_cast_fp16")]; + tensor var_718_cast_fp16 = mul(x = dx2_5_cast_fp16, y = weights_13_cast_fp16)[name = tensor("op_718_cast_fp16")]; tensor var_720_axes_0 = const()[name = tensor("op_720_axes_0"), val = tensor([2])]; tensor var_720_keep_dims_0 = const()[name = tensor("op_720_keep_dims_0"), val = tensor(false)]; tensor var_720_cast_fp16 = reduce_sum(axes = var_720_axes_0, keep_dims = var_720_keep_dims_0, x = var_718_cast_fp16)[name = tensor("op_720_cast_fp16")]; @@ -465,31 +465,31 @@ program(1.0) tensor var_728_interleave_0 = const()[name = tensor("op_728_interleave_0"), val = tensor(false)]; tensor var_728_cast_fp16 = concat(axis = var_72, interleave = var_728_interleave_0, values = (mean_5_cast_fp16, std_5_cast_fp16))[name = tensor("op_728_cast_fp16")]; tensor var_730_begin_0 = const()[name = tensor("op_730_begin_0"), val = tensor([0, 3, 0])]; - tensor var_730_end_0 = const()[name = tensor("op_730_end_0"), val = tensor([1, 4, 125])]; + tensor var_730_end_0 = const()[name = tensor("op_730_end_0"), val = tensor([1, 4, 375])]; tensor var_730_end_mask_0 = const()[name = tensor("op_730_end_mask_0"), val = tensor([true, false, true])]; tensor var_730_squeeze_mask_0 = const()[name = tensor("op_730_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_730_cast_fp16 = slice_by_index(begin = var_730_begin_0, end = var_730_end_0, end_mask = var_730_end_mask_0, squeeze_mask = var_730_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_730_cast_fp16")]; + tensor var_730_cast_fp16 = slice_by_index(begin = var_730_begin_0, end = var_730_end_0, end_mask = var_730_end_mask_0, squeeze_mask = var_730_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_730_cast_fp16")]; tensor weights_17_axes_0 = const()[name = tensor("weights_17_axes_0"), val = tensor([1])]; - tensor weights_17_cast_fp16 = expand_dims(axes = weights_17_axes_0, x = var_730_cast_fp16)[name = tensor("weights_17_cast_fp16")]; + tensor weights_17_cast_fp16 = expand_dims(axes = weights_17_axes_0, x = var_730_cast_fp16)[name = tensor("weights_17_cast_fp16")]; tensor var_734_axes_0 = const()[name = tensor("op_734_axes_0"), val = tensor([2])]; tensor var_734_keep_dims_0 = const()[name = tensor("op_734_keep_dims_0"), val = tensor(false)]; tensor var_734_cast_fp16 = reduce_sum(axes = var_734_axes_0, keep_dims = var_734_keep_dims_0, x = weights_17_cast_fp16)[name = tensor("op_734_cast_fp16")]; tensor var_735_to_fp16 = const()[name = tensor("op_735_to_fp16"), val = tensor(0x1p-24)]; tensor v1_7_cast_fp16 = add(x = var_734_cast_fp16, y = var_735_to_fp16)[name = tensor("v1_7_cast_fp16")]; - tensor var_737_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_17_cast_fp16)[name = tensor("op_737_cast_fp16")]; + tensor var_737_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_17_cast_fp16)[name = tensor("op_737_cast_fp16")]; tensor var_739_axes_0 = const()[name = tensor("op_739_axes_0"), val = tensor([2])]; tensor var_739_keep_dims_0 = const()[name = tensor("op_739_keep_dims_0"), val = tensor(false)]; tensor var_739_cast_fp16 = reduce_sum(axes = var_739_axes_0, keep_dims = var_739_keep_dims_0, x = var_737_cast_fp16)[name = tensor("op_739_cast_fp16")]; tensor mean_7_cast_fp16 = real_div(x = var_739_cast_fp16, y = v1_7_cast_fp16)[name = tensor("mean_7_cast_fp16")]; tensor var_741_axes_0 = const()[name = tensor("op_741_axes_0"), val = tensor([2])]; tensor var_741_cast_fp16 = expand_dims(axes = var_741_axes_0, x = mean_7_cast_fp16)[name = tensor("op_741_cast_fp16")]; - tensor var_742_cast_fp16 = sub(x = sequences_cast_fp16, y = var_741_cast_fp16)[name = tensor("op_742_cast_fp16")]; - tensor dx2_7_cast_fp16 = mul(x = var_742_cast_fp16, y = var_742_cast_fp16)[name = tensor("dx2_7_cast_fp16")]; - tensor var_744_cast_fp16 = mul(x = weights_17_cast_fp16, y = weights_17_cast_fp16)[name = tensor("op_744_cast_fp16")]; + tensor var_742_cast_fp16 = sub(x = sequences_cast_fp16, y = var_741_cast_fp16)[name = tensor("op_742_cast_fp16")]; + tensor dx2_7_cast_fp16 = mul(x = var_742_cast_fp16, y = var_742_cast_fp16)[name = tensor("dx2_7_cast_fp16")]; + tensor var_744_cast_fp16 = mul(x = weights_17_cast_fp16, y = weights_17_cast_fp16)[name = tensor("op_744_cast_fp16")]; tensor v2_7_axes_0 = const()[name = tensor("v2_7_axes_0"), val = tensor([2])]; tensor v2_7_keep_dims_0 = const()[name = tensor("v2_7_keep_dims_0"), val = tensor(false)]; tensor v2_7_cast_fp16 = reduce_sum(axes = v2_7_axes_0, keep_dims = v2_7_keep_dims_0, x = var_744_cast_fp16)[name = tensor("v2_7_cast_fp16")]; - tensor var_747_cast_fp16 = mul(x = dx2_7_cast_fp16, y = weights_17_cast_fp16)[name = tensor("op_747_cast_fp16")]; + tensor var_747_cast_fp16 = mul(x = dx2_7_cast_fp16, y = weights_17_cast_fp16)[name = tensor("op_747_cast_fp16")]; tensor var_749_axes_0 = const()[name = tensor("op_749_axes_0"), val = tensor([2])]; tensor var_749_keep_dims_0 = const()[name = tensor("op_749_keep_dims_0"), val = tensor(false)]; tensor var_749_cast_fp16 = reduce_sum(axes = var_749_axes_0, keep_dims = var_749_keep_dims_0, x = var_747_cast_fp16)[name = tensor("op_749_cast_fp16")]; @@ -502,31 +502,31 @@ program(1.0) tensor var_757_interleave_0 = const()[name = tensor("op_757_interleave_0"), val = tensor(false)]; tensor var_757_cast_fp16 = concat(axis = var_72, interleave = var_757_interleave_0, values = (mean_7_cast_fp16, std_7_cast_fp16))[name = tensor("op_757_cast_fp16")]; tensor var_759_begin_0 = const()[name = tensor("op_759_begin_0"), val = tensor([0, 4, 0])]; - tensor var_759_end_0 = const()[name = tensor("op_759_end_0"), val = tensor([1, 5, 125])]; + tensor var_759_end_0 = const()[name = tensor("op_759_end_0"), val = tensor([1, 5, 375])]; tensor var_759_end_mask_0 = const()[name = tensor("op_759_end_mask_0"), val = tensor([true, false, true])]; tensor var_759_squeeze_mask_0 = const()[name = tensor("op_759_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_759_cast_fp16 = slice_by_index(begin = var_759_begin_0, end = var_759_end_0, end_mask = var_759_end_mask_0, squeeze_mask = var_759_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_759_cast_fp16")]; + tensor var_759_cast_fp16 = slice_by_index(begin = var_759_begin_0, end = var_759_end_0, end_mask = var_759_end_mask_0, squeeze_mask = var_759_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_759_cast_fp16")]; tensor weights_21_axes_0 = const()[name = tensor("weights_21_axes_0"), val = tensor([1])]; - tensor weights_21_cast_fp16 = expand_dims(axes = weights_21_axes_0, x = var_759_cast_fp16)[name = tensor("weights_21_cast_fp16")]; + tensor weights_21_cast_fp16 = expand_dims(axes = weights_21_axes_0, x = var_759_cast_fp16)[name = tensor("weights_21_cast_fp16")]; tensor var_763_axes_0 = const()[name = tensor("op_763_axes_0"), val = tensor([2])]; tensor var_763_keep_dims_0 = const()[name = tensor("op_763_keep_dims_0"), val = tensor(false)]; tensor var_763_cast_fp16 = reduce_sum(axes = var_763_axes_0, keep_dims = var_763_keep_dims_0, x = weights_21_cast_fp16)[name = tensor("op_763_cast_fp16")]; tensor var_764_to_fp16 = const()[name = tensor("op_764_to_fp16"), val = tensor(0x1p-24)]; tensor v1_9_cast_fp16 = add(x = var_763_cast_fp16, y = var_764_to_fp16)[name = tensor("v1_9_cast_fp16")]; - tensor var_766_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_21_cast_fp16)[name = tensor("op_766_cast_fp16")]; + tensor var_766_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_21_cast_fp16)[name = tensor("op_766_cast_fp16")]; tensor var_768_axes_0 = const()[name = tensor("op_768_axes_0"), val = tensor([2])]; tensor var_768_keep_dims_0 = const()[name = tensor("op_768_keep_dims_0"), val = tensor(false)]; tensor var_768_cast_fp16 = reduce_sum(axes = var_768_axes_0, keep_dims = var_768_keep_dims_0, x = var_766_cast_fp16)[name = tensor("op_768_cast_fp16")]; tensor mean_9_cast_fp16 = real_div(x = var_768_cast_fp16, y = v1_9_cast_fp16)[name = tensor("mean_9_cast_fp16")]; tensor var_770_axes_0 = const()[name = tensor("op_770_axes_0"), val = tensor([2])]; tensor var_770_cast_fp16 = expand_dims(axes = var_770_axes_0, x = mean_9_cast_fp16)[name = tensor("op_770_cast_fp16")]; - tensor var_771_cast_fp16 = sub(x = sequences_cast_fp16, y = var_770_cast_fp16)[name = tensor("op_771_cast_fp16")]; - tensor dx2_9_cast_fp16 = mul(x = var_771_cast_fp16, y = var_771_cast_fp16)[name = tensor("dx2_9_cast_fp16")]; - tensor var_773_cast_fp16 = mul(x = weights_21_cast_fp16, y = weights_21_cast_fp16)[name = tensor("op_773_cast_fp16")]; + tensor var_771_cast_fp16 = sub(x = sequences_cast_fp16, y = var_770_cast_fp16)[name = tensor("op_771_cast_fp16")]; + tensor dx2_9_cast_fp16 = mul(x = var_771_cast_fp16, y = var_771_cast_fp16)[name = tensor("dx2_9_cast_fp16")]; + tensor var_773_cast_fp16 = mul(x = weights_21_cast_fp16, y = weights_21_cast_fp16)[name = tensor("op_773_cast_fp16")]; tensor v2_9_axes_0 = const()[name = tensor("v2_9_axes_0"), val = tensor([2])]; tensor v2_9_keep_dims_0 = const()[name = tensor("v2_9_keep_dims_0"), val = tensor(false)]; tensor v2_9_cast_fp16 = reduce_sum(axes = v2_9_axes_0, keep_dims = v2_9_keep_dims_0, x = var_773_cast_fp16)[name = tensor("v2_9_cast_fp16")]; - tensor var_776_cast_fp16 = mul(x = dx2_9_cast_fp16, y = weights_21_cast_fp16)[name = tensor("op_776_cast_fp16")]; + tensor var_776_cast_fp16 = mul(x = dx2_9_cast_fp16, y = weights_21_cast_fp16)[name = tensor("op_776_cast_fp16")]; tensor var_778_axes_0 = const()[name = tensor("op_778_axes_0"), val = tensor([2])]; tensor var_778_keep_dims_0 = const()[name = tensor("op_778_keep_dims_0"), val = tensor(false)]; tensor var_778_cast_fp16 = reduce_sum(axes = var_778_axes_0, keep_dims = var_778_keep_dims_0, x = var_776_cast_fp16)[name = tensor("op_778_cast_fp16")]; @@ -539,31 +539,31 @@ program(1.0) tensor var_786_interleave_0 = const()[name = tensor("op_786_interleave_0"), val = tensor(false)]; tensor var_786_cast_fp16 = concat(axis = var_72, interleave = var_786_interleave_0, values = (mean_9_cast_fp16, std_9_cast_fp16))[name = tensor("op_786_cast_fp16")]; tensor var_788_begin_0 = const()[name = tensor("op_788_begin_0"), val = tensor([0, 5, 0])]; - tensor var_788_end_0 = const()[name = tensor("op_788_end_0"), val = tensor([1, 6, 125])]; + tensor var_788_end_0 = const()[name = tensor("op_788_end_0"), val = tensor([1, 6, 375])]; tensor var_788_end_mask_0 = const()[name = tensor("op_788_end_mask_0"), val = tensor([true, false, true])]; tensor var_788_squeeze_mask_0 = const()[name = tensor("op_788_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_788_cast_fp16 = slice_by_index(begin = var_788_begin_0, end = var_788_end_0, end_mask = var_788_end_mask_0, squeeze_mask = var_788_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_788_cast_fp16")]; + tensor var_788_cast_fp16 = slice_by_index(begin = var_788_begin_0, end = var_788_end_0, end_mask = var_788_end_mask_0, squeeze_mask = var_788_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_788_cast_fp16")]; tensor weights_25_axes_0 = const()[name = tensor("weights_25_axes_0"), val = tensor([1])]; - tensor weights_25_cast_fp16 = expand_dims(axes = weights_25_axes_0, x = var_788_cast_fp16)[name = tensor("weights_25_cast_fp16")]; + tensor weights_25_cast_fp16 = expand_dims(axes = weights_25_axes_0, x = var_788_cast_fp16)[name = tensor("weights_25_cast_fp16")]; tensor var_792_axes_0 = const()[name = tensor("op_792_axes_0"), val = tensor([2])]; tensor var_792_keep_dims_0 = const()[name = tensor("op_792_keep_dims_0"), val = tensor(false)]; tensor var_792_cast_fp16 = reduce_sum(axes = var_792_axes_0, keep_dims = var_792_keep_dims_0, x = weights_25_cast_fp16)[name = tensor("op_792_cast_fp16")]; tensor var_793_to_fp16 = const()[name = tensor("op_793_to_fp16"), val = tensor(0x1p-24)]; tensor v1_11_cast_fp16 = add(x = var_792_cast_fp16, y = var_793_to_fp16)[name = tensor("v1_11_cast_fp16")]; - tensor var_795_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_25_cast_fp16)[name = tensor("op_795_cast_fp16")]; + tensor var_795_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_25_cast_fp16)[name = tensor("op_795_cast_fp16")]; tensor var_797_axes_0 = const()[name = tensor("op_797_axes_0"), val = tensor([2])]; tensor var_797_keep_dims_0 = const()[name = tensor("op_797_keep_dims_0"), val = tensor(false)]; tensor var_797_cast_fp16 = reduce_sum(axes = var_797_axes_0, keep_dims = var_797_keep_dims_0, x = var_795_cast_fp16)[name = tensor("op_797_cast_fp16")]; tensor mean_11_cast_fp16 = real_div(x = var_797_cast_fp16, y = v1_11_cast_fp16)[name = tensor("mean_11_cast_fp16")]; tensor var_799_axes_0 = const()[name = tensor("op_799_axes_0"), val = tensor([2])]; tensor var_799_cast_fp16 = expand_dims(axes = var_799_axes_0, x = mean_11_cast_fp16)[name = tensor("op_799_cast_fp16")]; - tensor var_800_cast_fp16 = sub(x = sequences_cast_fp16, y = var_799_cast_fp16)[name = tensor("op_800_cast_fp16")]; - tensor dx2_11_cast_fp16 = mul(x = var_800_cast_fp16, y = var_800_cast_fp16)[name = tensor("dx2_11_cast_fp16")]; - tensor var_802_cast_fp16 = mul(x = weights_25_cast_fp16, y = weights_25_cast_fp16)[name = tensor("op_802_cast_fp16")]; + tensor var_800_cast_fp16 = sub(x = sequences_cast_fp16, y = var_799_cast_fp16)[name = tensor("op_800_cast_fp16")]; + tensor dx2_11_cast_fp16 = mul(x = var_800_cast_fp16, y = var_800_cast_fp16)[name = tensor("dx2_11_cast_fp16")]; + tensor var_802_cast_fp16 = mul(x = weights_25_cast_fp16, y = weights_25_cast_fp16)[name = tensor("op_802_cast_fp16")]; tensor v2_11_axes_0 = const()[name = tensor("v2_11_axes_0"), val = tensor([2])]; tensor v2_11_keep_dims_0 = const()[name = tensor("v2_11_keep_dims_0"), val = tensor(false)]; tensor v2_11_cast_fp16 = reduce_sum(axes = v2_11_axes_0, keep_dims = v2_11_keep_dims_0, x = var_802_cast_fp16)[name = tensor("v2_11_cast_fp16")]; - tensor var_805_cast_fp16 = mul(x = dx2_11_cast_fp16, y = weights_25_cast_fp16)[name = tensor("op_805_cast_fp16")]; + tensor var_805_cast_fp16 = mul(x = dx2_11_cast_fp16, y = weights_25_cast_fp16)[name = tensor("op_805_cast_fp16")]; tensor var_807_axes_0 = const()[name = tensor("op_807_axes_0"), val = tensor([2])]; tensor var_807_keep_dims_0 = const()[name = tensor("op_807_keep_dims_0"), val = tensor(false)]; tensor var_807_cast_fp16 = reduce_sum(axes = var_807_axes_0, keep_dims = var_807_keep_dims_0, x = var_805_cast_fp16)[name = tensor("op_807_cast_fp16")]; @@ -576,31 +576,31 @@ program(1.0) tensor var_815_interleave_0 = const()[name = tensor("op_815_interleave_0"), val = tensor(false)]; tensor var_815_cast_fp16 = concat(axis = var_72, interleave = var_815_interleave_0, values = (mean_11_cast_fp16, std_11_cast_fp16))[name = tensor("op_815_cast_fp16")]; tensor var_817_begin_0 = const()[name = tensor("op_817_begin_0"), val = tensor([0, 6, 0])]; - tensor var_817_end_0 = const()[name = tensor("op_817_end_0"), val = tensor([1, 7, 125])]; + tensor var_817_end_0 = const()[name = tensor("op_817_end_0"), val = tensor([1, 7, 375])]; tensor var_817_end_mask_0 = const()[name = tensor("op_817_end_mask_0"), val = tensor([true, false, true])]; tensor var_817_squeeze_mask_0 = const()[name = tensor("op_817_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_817_cast_fp16 = slice_by_index(begin = var_817_begin_0, end = var_817_end_0, end_mask = var_817_end_mask_0, squeeze_mask = var_817_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_817_cast_fp16")]; + tensor var_817_cast_fp16 = slice_by_index(begin = var_817_begin_0, end = var_817_end_0, end_mask = var_817_end_mask_0, squeeze_mask = var_817_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_817_cast_fp16")]; tensor weights_29_axes_0 = const()[name = tensor("weights_29_axes_0"), val = tensor([1])]; - tensor weights_29_cast_fp16 = expand_dims(axes = weights_29_axes_0, x = var_817_cast_fp16)[name = tensor("weights_29_cast_fp16")]; + tensor weights_29_cast_fp16 = expand_dims(axes = weights_29_axes_0, x = var_817_cast_fp16)[name = tensor("weights_29_cast_fp16")]; tensor var_821_axes_0 = const()[name = tensor("op_821_axes_0"), val = tensor([2])]; tensor var_821_keep_dims_0 = const()[name = tensor("op_821_keep_dims_0"), val = tensor(false)]; tensor var_821_cast_fp16 = reduce_sum(axes = var_821_axes_0, keep_dims = var_821_keep_dims_0, x = weights_29_cast_fp16)[name = tensor("op_821_cast_fp16")]; tensor var_822_to_fp16 = const()[name = tensor("op_822_to_fp16"), val = tensor(0x1p-24)]; tensor v1_13_cast_fp16 = add(x = var_821_cast_fp16, y = var_822_to_fp16)[name = tensor("v1_13_cast_fp16")]; - tensor var_824_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_29_cast_fp16)[name = tensor("op_824_cast_fp16")]; + tensor var_824_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_29_cast_fp16)[name = tensor("op_824_cast_fp16")]; tensor var_826_axes_0 = const()[name = tensor("op_826_axes_0"), val = tensor([2])]; tensor var_826_keep_dims_0 = const()[name = tensor("op_826_keep_dims_0"), val = tensor(false)]; tensor var_826_cast_fp16 = reduce_sum(axes = var_826_axes_0, keep_dims = var_826_keep_dims_0, x = var_824_cast_fp16)[name = tensor("op_826_cast_fp16")]; tensor mean_13_cast_fp16 = real_div(x = var_826_cast_fp16, y = v1_13_cast_fp16)[name = tensor("mean_13_cast_fp16")]; tensor var_828_axes_0 = const()[name = tensor("op_828_axes_0"), val = tensor([2])]; tensor var_828_cast_fp16 = expand_dims(axes = var_828_axes_0, x = mean_13_cast_fp16)[name = tensor("op_828_cast_fp16")]; - tensor var_829_cast_fp16 = sub(x = sequences_cast_fp16, y = var_828_cast_fp16)[name = tensor("op_829_cast_fp16")]; - tensor dx2_13_cast_fp16 = mul(x = var_829_cast_fp16, y = var_829_cast_fp16)[name = tensor("dx2_13_cast_fp16")]; - tensor var_831_cast_fp16 = mul(x = weights_29_cast_fp16, y = weights_29_cast_fp16)[name = tensor("op_831_cast_fp16")]; + tensor var_829_cast_fp16 = sub(x = sequences_cast_fp16, y = var_828_cast_fp16)[name = tensor("op_829_cast_fp16")]; + tensor dx2_13_cast_fp16 = mul(x = var_829_cast_fp16, y = var_829_cast_fp16)[name = tensor("dx2_13_cast_fp16")]; + tensor var_831_cast_fp16 = mul(x = weights_29_cast_fp16, y = weights_29_cast_fp16)[name = tensor("op_831_cast_fp16")]; tensor v2_13_axes_0 = const()[name = tensor("v2_13_axes_0"), val = tensor([2])]; tensor v2_13_keep_dims_0 = const()[name = tensor("v2_13_keep_dims_0"), val = tensor(false)]; tensor v2_13_cast_fp16 = reduce_sum(axes = v2_13_axes_0, keep_dims = v2_13_keep_dims_0, x = var_831_cast_fp16)[name = tensor("v2_13_cast_fp16")]; - tensor var_834_cast_fp16 = mul(x = dx2_13_cast_fp16, y = weights_29_cast_fp16)[name = tensor("op_834_cast_fp16")]; + tensor var_834_cast_fp16 = mul(x = dx2_13_cast_fp16, y = weights_29_cast_fp16)[name = tensor("op_834_cast_fp16")]; tensor var_836_axes_0 = const()[name = tensor("op_836_axes_0"), val = tensor([2])]; tensor var_836_keep_dims_0 = const()[name = tensor("op_836_keep_dims_0"), val = tensor(false)]; tensor var_836_cast_fp16 = reduce_sum(axes = var_836_axes_0, keep_dims = var_836_keep_dims_0, x = var_834_cast_fp16)[name = tensor("op_836_cast_fp16")]; @@ -613,31 +613,31 @@ program(1.0) tensor var_844_interleave_0 = const()[name = tensor("op_844_interleave_0"), val = tensor(false)]; tensor var_844_cast_fp16 = concat(axis = var_72, interleave = var_844_interleave_0, values = (mean_13_cast_fp16, std_13_cast_fp16))[name = tensor("op_844_cast_fp16")]; tensor var_846_begin_0 = const()[name = tensor("op_846_begin_0"), val = tensor([0, 7, 0])]; - tensor var_846_end_0 = const()[name = tensor("op_846_end_0"), val = tensor([1, 8, 125])]; + tensor var_846_end_0 = const()[name = tensor("op_846_end_0"), val = tensor([1, 8, 375])]; tensor var_846_end_mask_0 = const()[name = tensor("op_846_end_mask_0"), val = tensor([true, false, true])]; tensor var_846_squeeze_mask_0 = const()[name = tensor("op_846_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_846_cast_fp16 = slice_by_index(begin = var_846_begin_0, end = var_846_end_0, end_mask = var_846_end_mask_0, squeeze_mask = var_846_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_846_cast_fp16")]; + tensor var_846_cast_fp16 = slice_by_index(begin = var_846_begin_0, end = var_846_end_0, end_mask = var_846_end_mask_0, squeeze_mask = var_846_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_846_cast_fp16")]; tensor weights_33_axes_0 = const()[name = tensor("weights_33_axes_0"), val = tensor([1])]; - tensor weights_33_cast_fp16 = expand_dims(axes = weights_33_axes_0, x = var_846_cast_fp16)[name = tensor("weights_33_cast_fp16")]; + tensor weights_33_cast_fp16 = expand_dims(axes = weights_33_axes_0, x = var_846_cast_fp16)[name = tensor("weights_33_cast_fp16")]; tensor var_850_axes_0 = const()[name = tensor("op_850_axes_0"), val = tensor([2])]; tensor var_850_keep_dims_0 = const()[name = tensor("op_850_keep_dims_0"), val = tensor(false)]; tensor var_850_cast_fp16 = reduce_sum(axes = var_850_axes_0, keep_dims = var_850_keep_dims_0, x = weights_33_cast_fp16)[name = tensor("op_850_cast_fp16")]; tensor var_851_to_fp16 = const()[name = tensor("op_851_to_fp16"), val = tensor(0x1p-24)]; tensor v1_15_cast_fp16 = add(x = var_850_cast_fp16, y = var_851_to_fp16)[name = tensor("v1_15_cast_fp16")]; - tensor var_853_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_33_cast_fp16)[name = tensor("op_853_cast_fp16")]; + tensor var_853_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_33_cast_fp16)[name = tensor("op_853_cast_fp16")]; tensor var_855_axes_0 = const()[name = tensor("op_855_axes_0"), val = tensor([2])]; tensor var_855_keep_dims_0 = const()[name = tensor("op_855_keep_dims_0"), val = tensor(false)]; tensor var_855_cast_fp16 = reduce_sum(axes = var_855_axes_0, keep_dims = var_855_keep_dims_0, x = var_853_cast_fp16)[name = tensor("op_855_cast_fp16")]; tensor mean_15_cast_fp16 = real_div(x = var_855_cast_fp16, y = v1_15_cast_fp16)[name = tensor("mean_15_cast_fp16")]; tensor var_857_axes_0 = const()[name = tensor("op_857_axes_0"), val = tensor([2])]; tensor var_857_cast_fp16 = expand_dims(axes = var_857_axes_0, x = mean_15_cast_fp16)[name = tensor("op_857_cast_fp16")]; - tensor var_858_cast_fp16 = sub(x = sequences_cast_fp16, y = var_857_cast_fp16)[name = tensor("op_858_cast_fp16")]; - tensor dx2_15_cast_fp16 = mul(x = var_858_cast_fp16, y = var_858_cast_fp16)[name = tensor("dx2_15_cast_fp16")]; - tensor var_860_cast_fp16 = mul(x = weights_33_cast_fp16, y = weights_33_cast_fp16)[name = tensor("op_860_cast_fp16")]; + tensor var_858_cast_fp16 = sub(x = sequences_cast_fp16, y = var_857_cast_fp16)[name = tensor("op_858_cast_fp16")]; + tensor dx2_15_cast_fp16 = mul(x = var_858_cast_fp16, y = var_858_cast_fp16)[name = tensor("dx2_15_cast_fp16")]; + tensor var_860_cast_fp16 = mul(x = weights_33_cast_fp16, y = weights_33_cast_fp16)[name = tensor("op_860_cast_fp16")]; tensor v2_15_axes_0 = const()[name = tensor("v2_15_axes_0"), val = tensor([2])]; tensor v2_15_keep_dims_0 = const()[name = tensor("v2_15_keep_dims_0"), val = tensor(false)]; tensor v2_15_cast_fp16 = reduce_sum(axes = v2_15_axes_0, keep_dims = v2_15_keep_dims_0, x = var_860_cast_fp16)[name = tensor("v2_15_cast_fp16")]; - tensor var_863_cast_fp16 = mul(x = dx2_15_cast_fp16, y = weights_33_cast_fp16)[name = tensor("op_863_cast_fp16")]; + tensor var_863_cast_fp16 = mul(x = dx2_15_cast_fp16, y = weights_33_cast_fp16)[name = tensor("op_863_cast_fp16")]; tensor var_865_axes_0 = const()[name = tensor("op_865_axes_0"), val = tensor([2])]; tensor var_865_keep_dims_0 = const()[name = tensor("op_865_keep_dims_0"), val = tensor(false)]; tensor var_865_cast_fp16 = reduce_sum(axes = var_865_axes_0, keep_dims = var_865_keep_dims_0, x = var_863_cast_fp16)[name = tensor("op_865_cast_fp16")]; @@ -650,31 +650,31 @@ program(1.0) tensor var_873_interleave_0 = const()[name = tensor("op_873_interleave_0"), val = tensor(false)]; tensor var_873_cast_fp16 = concat(axis = var_72, interleave = var_873_interleave_0, values = (mean_15_cast_fp16, std_15_cast_fp16))[name = tensor("op_873_cast_fp16")]; tensor var_875_begin_0 = const()[name = tensor("op_875_begin_0"), val = tensor([0, 8, 0])]; - tensor var_875_end_0 = const()[name = tensor("op_875_end_0"), val = tensor([1, 9, 125])]; + tensor var_875_end_0 = const()[name = tensor("op_875_end_0"), val = tensor([1, 9, 375])]; tensor var_875_end_mask_0 = const()[name = tensor("op_875_end_mask_0"), val = tensor([true, false, true])]; tensor var_875_squeeze_mask_0 = const()[name = tensor("op_875_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_875_cast_fp16 = slice_by_index(begin = var_875_begin_0, end = var_875_end_0, end_mask = var_875_end_mask_0, squeeze_mask = var_875_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_875_cast_fp16")]; + tensor var_875_cast_fp16 = slice_by_index(begin = var_875_begin_0, end = var_875_end_0, end_mask = var_875_end_mask_0, squeeze_mask = var_875_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_875_cast_fp16")]; tensor weights_37_axes_0 = const()[name = tensor("weights_37_axes_0"), val = tensor([1])]; - tensor weights_37_cast_fp16 = expand_dims(axes = weights_37_axes_0, x = var_875_cast_fp16)[name = tensor("weights_37_cast_fp16")]; + tensor weights_37_cast_fp16 = expand_dims(axes = weights_37_axes_0, x = var_875_cast_fp16)[name = tensor("weights_37_cast_fp16")]; tensor var_879_axes_0 = const()[name = tensor("op_879_axes_0"), val = tensor([2])]; tensor var_879_keep_dims_0 = const()[name = tensor("op_879_keep_dims_0"), val = tensor(false)]; tensor var_879_cast_fp16 = reduce_sum(axes = var_879_axes_0, keep_dims = var_879_keep_dims_0, x = weights_37_cast_fp16)[name = tensor("op_879_cast_fp16")]; tensor var_880_to_fp16 = const()[name = tensor("op_880_to_fp16"), val = tensor(0x1p-24)]; tensor v1_17_cast_fp16 = add(x = var_879_cast_fp16, y = var_880_to_fp16)[name = tensor("v1_17_cast_fp16")]; - tensor var_882_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_37_cast_fp16)[name = tensor("op_882_cast_fp16")]; + tensor var_882_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_37_cast_fp16)[name = tensor("op_882_cast_fp16")]; tensor var_884_axes_0 = const()[name = tensor("op_884_axes_0"), val = tensor([2])]; tensor var_884_keep_dims_0 = const()[name = tensor("op_884_keep_dims_0"), val = tensor(false)]; tensor var_884_cast_fp16 = reduce_sum(axes = var_884_axes_0, keep_dims = var_884_keep_dims_0, x = var_882_cast_fp16)[name = tensor("op_884_cast_fp16")]; tensor mean_17_cast_fp16 = real_div(x = var_884_cast_fp16, y = v1_17_cast_fp16)[name = tensor("mean_17_cast_fp16")]; tensor var_886_axes_0 = const()[name = tensor("op_886_axes_0"), val = tensor([2])]; tensor var_886_cast_fp16 = expand_dims(axes = var_886_axes_0, x = mean_17_cast_fp16)[name = tensor("op_886_cast_fp16")]; - tensor var_887_cast_fp16 = sub(x = sequences_cast_fp16, y = var_886_cast_fp16)[name = tensor("op_887_cast_fp16")]; - tensor dx2_17_cast_fp16 = mul(x = var_887_cast_fp16, y = var_887_cast_fp16)[name = tensor("dx2_17_cast_fp16")]; - tensor var_889_cast_fp16 = mul(x = weights_37_cast_fp16, y = weights_37_cast_fp16)[name = tensor("op_889_cast_fp16")]; + tensor var_887_cast_fp16 = sub(x = sequences_cast_fp16, y = var_886_cast_fp16)[name = tensor("op_887_cast_fp16")]; + tensor dx2_17_cast_fp16 = mul(x = var_887_cast_fp16, y = var_887_cast_fp16)[name = tensor("dx2_17_cast_fp16")]; + tensor var_889_cast_fp16 = mul(x = weights_37_cast_fp16, y = weights_37_cast_fp16)[name = tensor("op_889_cast_fp16")]; tensor v2_17_axes_0 = const()[name = tensor("v2_17_axes_0"), val = tensor([2])]; tensor v2_17_keep_dims_0 = const()[name = tensor("v2_17_keep_dims_0"), val = tensor(false)]; tensor v2_17_cast_fp16 = reduce_sum(axes = v2_17_axes_0, keep_dims = v2_17_keep_dims_0, x = var_889_cast_fp16)[name = tensor("v2_17_cast_fp16")]; - tensor var_892_cast_fp16 = mul(x = dx2_17_cast_fp16, y = weights_37_cast_fp16)[name = tensor("op_892_cast_fp16")]; + tensor var_892_cast_fp16 = mul(x = dx2_17_cast_fp16, y = weights_37_cast_fp16)[name = tensor("op_892_cast_fp16")]; tensor var_894_axes_0 = const()[name = tensor("op_894_axes_0"), val = tensor([2])]; tensor var_894_keep_dims_0 = const()[name = tensor("op_894_keep_dims_0"), val = tensor(false)]; tensor var_894_cast_fp16 = reduce_sum(axes = var_894_axes_0, keep_dims = var_894_keep_dims_0, x = var_892_cast_fp16)[name = tensor("op_894_cast_fp16")]; @@ -687,31 +687,31 @@ program(1.0) tensor var_902_interleave_0 = const()[name = tensor("op_902_interleave_0"), val = tensor(false)]; tensor var_902_cast_fp16 = concat(axis = var_72, interleave = var_902_interleave_0, values = (mean_17_cast_fp16, std_17_cast_fp16))[name = tensor("op_902_cast_fp16")]; tensor var_904_begin_0 = const()[name = tensor("op_904_begin_0"), val = tensor([0, 9, 0])]; - tensor var_904_end_0 = const()[name = tensor("op_904_end_0"), val = tensor([1, 10, 125])]; + tensor var_904_end_0 = const()[name = tensor("op_904_end_0"), val = tensor([1, 10, 375])]; tensor var_904_end_mask_0 = const()[name = tensor("op_904_end_mask_0"), val = tensor([true, false, true])]; tensor var_904_squeeze_mask_0 = const()[name = tensor("op_904_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_904_cast_fp16 = slice_by_index(begin = var_904_begin_0, end = var_904_end_0, end_mask = var_904_end_mask_0, squeeze_mask = var_904_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_904_cast_fp16")]; + tensor var_904_cast_fp16 = slice_by_index(begin = var_904_begin_0, end = var_904_end_0, end_mask = var_904_end_mask_0, squeeze_mask = var_904_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_904_cast_fp16")]; tensor weights_41_axes_0 = const()[name = tensor("weights_41_axes_0"), val = tensor([1])]; - tensor weights_41_cast_fp16 = expand_dims(axes = weights_41_axes_0, x = var_904_cast_fp16)[name = tensor("weights_41_cast_fp16")]; + tensor weights_41_cast_fp16 = expand_dims(axes = weights_41_axes_0, x = var_904_cast_fp16)[name = tensor("weights_41_cast_fp16")]; tensor var_908_axes_0 = const()[name = tensor("op_908_axes_0"), val = tensor([2])]; tensor var_908_keep_dims_0 = const()[name = tensor("op_908_keep_dims_0"), val = tensor(false)]; tensor var_908_cast_fp16 = reduce_sum(axes = var_908_axes_0, keep_dims = var_908_keep_dims_0, x = weights_41_cast_fp16)[name = tensor("op_908_cast_fp16")]; tensor var_909_to_fp16 = const()[name = tensor("op_909_to_fp16"), val = tensor(0x1p-24)]; tensor v1_19_cast_fp16 = add(x = var_908_cast_fp16, y = var_909_to_fp16)[name = tensor("v1_19_cast_fp16")]; - tensor var_911_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_41_cast_fp16)[name = tensor("op_911_cast_fp16")]; + tensor var_911_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_41_cast_fp16)[name = tensor("op_911_cast_fp16")]; tensor var_913_axes_0 = const()[name = tensor("op_913_axes_0"), val = tensor([2])]; tensor var_913_keep_dims_0 = const()[name = tensor("op_913_keep_dims_0"), val = tensor(false)]; tensor var_913_cast_fp16 = reduce_sum(axes = var_913_axes_0, keep_dims = var_913_keep_dims_0, x = var_911_cast_fp16)[name = tensor("op_913_cast_fp16")]; tensor mean_19_cast_fp16 = real_div(x = var_913_cast_fp16, y = v1_19_cast_fp16)[name = tensor("mean_19_cast_fp16")]; tensor var_915_axes_0 = const()[name = tensor("op_915_axes_0"), val = tensor([2])]; tensor var_915_cast_fp16 = expand_dims(axes = var_915_axes_0, x = mean_19_cast_fp16)[name = tensor("op_915_cast_fp16")]; - tensor var_916_cast_fp16 = sub(x = sequences_cast_fp16, y = var_915_cast_fp16)[name = tensor("op_916_cast_fp16")]; - tensor dx2_19_cast_fp16 = mul(x = var_916_cast_fp16, y = var_916_cast_fp16)[name = tensor("dx2_19_cast_fp16")]; - tensor var_918_cast_fp16 = mul(x = weights_41_cast_fp16, y = weights_41_cast_fp16)[name = tensor("op_918_cast_fp16")]; + tensor var_916_cast_fp16 = sub(x = sequences_cast_fp16, y = var_915_cast_fp16)[name = tensor("op_916_cast_fp16")]; + tensor dx2_19_cast_fp16 = mul(x = var_916_cast_fp16, y = var_916_cast_fp16)[name = tensor("dx2_19_cast_fp16")]; + tensor var_918_cast_fp16 = mul(x = weights_41_cast_fp16, y = weights_41_cast_fp16)[name = tensor("op_918_cast_fp16")]; tensor v2_19_axes_0 = const()[name = tensor("v2_19_axes_0"), val = tensor([2])]; tensor v2_19_keep_dims_0 = const()[name = tensor("v2_19_keep_dims_0"), val = tensor(false)]; tensor v2_19_cast_fp16 = reduce_sum(axes = v2_19_axes_0, keep_dims = v2_19_keep_dims_0, x = var_918_cast_fp16)[name = tensor("v2_19_cast_fp16")]; - tensor var_921_cast_fp16 = mul(x = dx2_19_cast_fp16, y = weights_41_cast_fp16)[name = tensor("op_921_cast_fp16")]; + tensor var_921_cast_fp16 = mul(x = dx2_19_cast_fp16, y = weights_41_cast_fp16)[name = tensor("op_921_cast_fp16")]; tensor var_923_axes_0 = const()[name = tensor("op_923_axes_0"), val = tensor([2])]; tensor var_923_keep_dims_0 = const()[name = tensor("op_923_keep_dims_0"), val = tensor(false)]; tensor var_923_cast_fp16 = reduce_sum(axes = var_923_axes_0, keep_dims = var_923_keep_dims_0, x = var_921_cast_fp16)[name = tensor("op_923_cast_fp16")]; @@ -724,31 +724,31 @@ program(1.0) tensor var_931_interleave_0 = const()[name = tensor("op_931_interleave_0"), val = tensor(false)]; tensor var_931_cast_fp16 = concat(axis = var_72, interleave = var_931_interleave_0, values = (mean_19_cast_fp16, std_19_cast_fp16))[name = tensor("op_931_cast_fp16")]; tensor var_933_begin_0 = const()[name = tensor("op_933_begin_0"), val = tensor([0, 10, 0])]; - tensor var_933_end_0 = const()[name = tensor("op_933_end_0"), val = tensor([1, 11, 125])]; + tensor var_933_end_0 = const()[name = tensor("op_933_end_0"), val = tensor([1, 11, 375])]; tensor var_933_end_mask_0 = const()[name = tensor("op_933_end_mask_0"), val = tensor([true, false, true])]; tensor var_933_squeeze_mask_0 = const()[name = tensor("op_933_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_933_cast_fp16 = slice_by_index(begin = var_933_begin_0, end = var_933_end_0, end_mask = var_933_end_mask_0, squeeze_mask = var_933_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_933_cast_fp16")]; + tensor var_933_cast_fp16 = slice_by_index(begin = var_933_begin_0, end = var_933_end_0, end_mask = var_933_end_mask_0, squeeze_mask = var_933_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_933_cast_fp16")]; tensor weights_45_axes_0 = const()[name = tensor("weights_45_axes_0"), val = tensor([1])]; - tensor weights_45_cast_fp16 = expand_dims(axes = weights_45_axes_0, x = var_933_cast_fp16)[name = tensor("weights_45_cast_fp16")]; + tensor weights_45_cast_fp16 = expand_dims(axes = weights_45_axes_0, x = var_933_cast_fp16)[name = tensor("weights_45_cast_fp16")]; tensor var_937_axes_0 = const()[name = tensor("op_937_axes_0"), val = tensor([2])]; tensor var_937_keep_dims_0 = const()[name = tensor("op_937_keep_dims_0"), val = tensor(false)]; tensor var_937_cast_fp16 = reduce_sum(axes = var_937_axes_0, keep_dims = var_937_keep_dims_0, x = weights_45_cast_fp16)[name = tensor("op_937_cast_fp16")]; tensor var_938_to_fp16 = const()[name = tensor("op_938_to_fp16"), val = tensor(0x1p-24)]; tensor v1_21_cast_fp16 = add(x = var_937_cast_fp16, y = var_938_to_fp16)[name = tensor("v1_21_cast_fp16")]; - tensor var_940_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_45_cast_fp16)[name = tensor("op_940_cast_fp16")]; + tensor var_940_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_45_cast_fp16)[name = tensor("op_940_cast_fp16")]; tensor var_942_axes_0 = const()[name = tensor("op_942_axes_0"), val = tensor([2])]; tensor var_942_keep_dims_0 = const()[name = tensor("op_942_keep_dims_0"), val = tensor(false)]; tensor var_942_cast_fp16 = reduce_sum(axes = var_942_axes_0, keep_dims = var_942_keep_dims_0, x = var_940_cast_fp16)[name = tensor("op_942_cast_fp16")]; tensor mean_21_cast_fp16 = real_div(x = var_942_cast_fp16, y = v1_21_cast_fp16)[name = tensor("mean_21_cast_fp16")]; tensor var_944_axes_0 = const()[name = tensor("op_944_axes_0"), val = tensor([2])]; tensor var_944_cast_fp16 = expand_dims(axes = var_944_axes_0, x = mean_21_cast_fp16)[name = tensor("op_944_cast_fp16")]; - tensor var_945_cast_fp16 = sub(x = sequences_cast_fp16, y = var_944_cast_fp16)[name = tensor("op_945_cast_fp16")]; - tensor dx2_21_cast_fp16 = mul(x = var_945_cast_fp16, y = var_945_cast_fp16)[name = tensor("dx2_21_cast_fp16")]; - tensor var_947_cast_fp16 = mul(x = weights_45_cast_fp16, y = weights_45_cast_fp16)[name = tensor("op_947_cast_fp16")]; + tensor var_945_cast_fp16 = sub(x = sequences_cast_fp16, y = var_944_cast_fp16)[name = tensor("op_945_cast_fp16")]; + tensor dx2_21_cast_fp16 = mul(x = var_945_cast_fp16, y = var_945_cast_fp16)[name = tensor("dx2_21_cast_fp16")]; + tensor var_947_cast_fp16 = mul(x = weights_45_cast_fp16, y = weights_45_cast_fp16)[name = tensor("op_947_cast_fp16")]; tensor v2_21_axes_0 = const()[name = tensor("v2_21_axes_0"), val = tensor([2])]; tensor v2_21_keep_dims_0 = const()[name = tensor("v2_21_keep_dims_0"), val = tensor(false)]; tensor v2_21_cast_fp16 = reduce_sum(axes = v2_21_axes_0, keep_dims = v2_21_keep_dims_0, x = var_947_cast_fp16)[name = tensor("v2_21_cast_fp16")]; - tensor var_950_cast_fp16 = mul(x = dx2_21_cast_fp16, y = weights_45_cast_fp16)[name = tensor("op_950_cast_fp16")]; + tensor var_950_cast_fp16 = mul(x = dx2_21_cast_fp16, y = weights_45_cast_fp16)[name = tensor("op_950_cast_fp16")]; tensor var_952_axes_0 = const()[name = tensor("op_952_axes_0"), val = tensor([2])]; tensor var_952_keep_dims_0 = const()[name = tensor("op_952_keep_dims_0"), val = tensor(false)]; tensor var_952_cast_fp16 = reduce_sum(axes = var_952_axes_0, keep_dims = var_952_keep_dims_0, x = var_950_cast_fp16)[name = tensor("op_952_cast_fp16")]; @@ -761,31 +761,31 @@ program(1.0) tensor var_960_interleave_0 = const()[name = tensor("op_960_interleave_0"), val = tensor(false)]; tensor var_960_cast_fp16 = concat(axis = var_72, interleave = var_960_interleave_0, values = (mean_21_cast_fp16, std_21_cast_fp16))[name = tensor("op_960_cast_fp16")]; tensor var_962_begin_0 = const()[name = tensor("op_962_begin_0"), val = tensor([0, 11, 0])]; - tensor var_962_end_0 = const()[name = tensor("op_962_end_0"), val = tensor([1, 12, 125])]; + tensor var_962_end_0 = const()[name = tensor("op_962_end_0"), val = tensor([1, 12, 375])]; tensor var_962_end_mask_0 = const()[name = tensor("op_962_end_mask_0"), val = tensor([true, false, true])]; tensor var_962_squeeze_mask_0 = const()[name = tensor("op_962_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, squeeze_mask = var_962_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_962_cast_fp16")]; + tensor var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, squeeze_mask = var_962_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_962_cast_fp16")]; tensor weights_49_axes_0 = const()[name = tensor("weights_49_axes_0"), val = tensor([1])]; - tensor weights_49_cast_fp16 = expand_dims(axes = weights_49_axes_0, x = var_962_cast_fp16)[name = tensor("weights_49_cast_fp16")]; + tensor weights_49_cast_fp16 = expand_dims(axes = weights_49_axes_0, x = var_962_cast_fp16)[name = tensor("weights_49_cast_fp16")]; tensor var_966_axes_0 = const()[name = tensor("op_966_axes_0"), val = tensor([2])]; tensor var_966_keep_dims_0 = const()[name = tensor("op_966_keep_dims_0"), val = tensor(false)]; tensor var_966_cast_fp16 = reduce_sum(axes = var_966_axes_0, keep_dims = var_966_keep_dims_0, x = weights_49_cast_fp16)[name = tensor("op_966_cast_fp16")]; tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(0x1p-24)]; tensor v1_23_cast_fp16 = add(x = var_966_cast_fp16, y = var_967_to_fp16)[name = tensor("v1_23_cast_fp16")]; - tensor var_969_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_49_cast_fp16)[name = tensor("op_969_cast_fp16")]; + tensor var_969_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_49_cast_fp16)[name = tensor("op_969_cast_fp16")]; tensor var_971_axes_0 = const()[name = tensor("op_971_axes_0"), val = tensor([2])]; tensor var_971_keep_dims_0 = const()[name = tensor("op_971_keep_dims_0"), val = tensor(false)]; tensor var_971_cast_fp16 = reduce_sum(axes = var_971_axes_0, keep_dims = var_971_keep_dims_0, x = var_969_cast_fp16)[name = tensor("op_971_cast_fp16")]; tensor mean_23_cast_fp16 = real_div(x = var_971_cast_fp16, y = v1_23_cast_fp16)[name = tensor("mean_23_cast_fp16")]; tensor var_973_axes_0 = const()[name = tensor("op_973_axes_0"), val = tensor([2])]; tensor var_973_cast_fp16 = expand_dims(axes = var_973_axes_0, x = mean_23_cast_fp16)[name = tensor("op_973_cast_fp16")]; - tensor var_974_cast_fp16 = sub(x = sequences_cast_fp16, y = var_973_cast_fp16)[name = tensor("op_974_cast_fp16")]; - tensor dx2_23_cast_fp16 = mul(x = var_974_cast_fp16, y = var_974_cast_fp16)[name = tensor("dx2_23_cast_fp16")]; - tensor var_976_cast_fp16 = mul(x = weights_49_cast_fp16, y = weights_49_cast_fp16)[name = tensor("op_976_cast_fp16")]; + tensor var_974_cast_fp16 = sub(x = sequences_cast_fp16, y = var_973_cast_fp16)[name = tensor("op_974_cast_fp16")]; + tensor dx2_23_cast_fp16 = mul(x = var_974_cast_fp16, y = var_974_cast_fp16)[name = tensor("dx2_23_cast_fp16")]; + tensor var_976_cast_fp16 = mul(x = weights_49_cast_fp16, y = weights_49_cast_fp16)[name = tensor("op_976_cast_fp16")]; tensor v2_23_axes_0 = const()[name = tensor("v2_23_axes_0"), val = tensor([2])]; tensor v2_23_keep_dims_0 = const()[name = tensor("v2_23_keep_dims_0"), val = tensor(false)]; tensor v2_23_cast_fp16 = reduce_sum(axes = v2_23_axes_0, keep_dims = v2_23_keep_dims_0, x = var_976_cast_fp16)[name = tensor("v2_23_cast_fp16")]; - tensor var_979_cast_fp16 = mul(x = dx2_23_cast_fp16, y = weights_49_cast_fp16)[name = tensor("op_979_cast_fp16")]; + tensor var_979_cast_fp16 = mul(x = dx2_23_cast_fp16, y = weights_49_cast_fp16)[name = tensor("op_979_cast_fp16")]; tensor var_981_axes_0 = const()[name = tensor("op_981_axes_0"), val = tensor([2])]; tensor var_981_keep_dims_0 = const()[name = tensor("op_981_keep_dims_0"), val = tensor(false)]; tensor var_981_cast_fp16 = reduce_sum(axes = var_981_axes_0, keep_dims = var_981_keep_dims_0, x = var_979_cast_fp16)[name = tensor("op_981_cast_fp16")]; @@ -798,31 +798,31 @@ program(1.0) tensor var_989_interleave_0 = const()[name = tensor("op_989_interleave_0"), val = tensor(false)]; tensor var_989_cast_fp16 = concat(axis = var_72, interleave = var_989_interleave_0, values = (mean_23_cast_fp16, std_23_cast_fp16))[name = tensor("op_989_cast_fp16")]; tensor var_991_begin_0 = const()[name = tensor("op_991_begin_0"), val = tensor([0, 12, 0])]; - tensor var_991_end_0 = const()[name = tensor("op_991_end_0"), val = tensor([1, 13, 125])]; + tensor var_991_end_0 = const()[name = tensor("op_991_end_0"), val = tensor([1, 13, 375])]; tensor var_991_end_mask_0 = const()[name = tensor("op_991_end_mask_0"), val = tensor([true, false, true])]; tensor var_991_squeeze_mask_0 = const()[name = tensor("op_991_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, squeeze_mask = var_991_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_991_cast_fp16")]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, squeeze_mask = var_991_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_991_cast_fp16")]; tensor weights_53_axes_0 = const()[name = tensor("weights_53_axes_0"), val = tensor([1])]; - tensor weights_53_cast_fp16 = expand_dims(axes = weights_53_axes_0, x = var_991_cast_fp16)[name = tensor("weights_53_cast_fp16")]; + tensor weights_53_cast_fp16 = expand_dims(axes = weights_53_axes_0, x = var_991_cast_fp16)[name = tensor("weights_53_cast_fp16")]; tensor var_995_axes_0 = const()[name = tensor("op_995_axes_0"), val = tensor([2])]; tensor var_995_keep_dims_0 = const()[name = tensor("op_995_keep_dims_0"), val = tensor(false)]; tensor var_995_cast_fp16 = reduce_sum(axes = var_995_axes_0, keep_dims = var_995_keep_dims_0, x = weights_53_cast_fp16)[name = tensor("op_995_cast_fp16")]; tensor var_996_to_fp16 = const()[name = tensor("op_996_to_fp16"), val = tensor(0x1p-24)]; tensor v1_25_cast_fp16 = add(x = var_995_cast_fp16, y = var_996_to_fp16)[name = tensor("v1_25_cast_fp16")]; - tensor var_998_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_53_cast_fp16)[name = tensor("op_998_cast_fp16")]; + tensor var_998_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_53_cast_fp16)[name = tensor("op_998_cast_fp16")]; tensor var_1000_axes_0 = const()[name = tensor("op_1000_axes_0"), val = tensor([2])]; tensor var_1000_keep_dims_0 = const()[name = tensor("op_1000_keep_dims_0"), val = tensor(false)]; tensor var_1000_cast_fp16 = reduce_sum(axes = var_1000_axes_0, keep_dims = var_1000_keep_dims_0, x = var_998_cast_fp16)[name = tensor("op_1000_cast_fp16")]; tensor mean_25_cast_fp16 = real_div(x = var_1000_cast_fp16, y = v1_25_cast_fp16)[name = tensor("mean_25_cast_fp16")]; tensor var_1002_axes_0 = const()[name = tensor("op_1002_axes_0"), val = tensor([2])]; tensor var_1002_cast_fp16 = expand_dims(axes = var_1002_axes_0, x = mean_25_cast_fp16)[name = tensor("op_1002_cast_fp16")]; - tensor var_1003_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1002_cast_fp16)[name = tensor("op_1003_cast_fp16")]; - tensor dx2_25_cast_fp16 = mul(x = var_1003_cast_fp16, y = var_1003_cast_fp16)[name = tensor("dx2_25_cast_fp16")]; - tensor var_1005_cast_fp16 = mul(x = weights_53_cast_fp16, y = weights_53_cast_fp16)[name = tensor("op_1005_cast_fp16")]; + tensor var_1003_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1002_cast_fp16)[name = tensor("op_1003_cast_fp16")]; + tensor dx2_25_cast_fp16 = mul(x = var_1003_cast_fp16, y = var_1003_cast_fp16)[name = tensor("dx2_25_cast_fp16")]; + tensor var_1005_cast_fp16 = mul(x = weights_53_cast_fp16, y = weights_53_cast_fp16)[name = tensor("op_1005_cast_fp16")]; tensor v2_25_axes_0 = const()[name = tensor("v2_25_axes_0"), val = tensor([2])]; tensor v2_25_keep_dims_0 = const()[name = tensor("v2_25_keep_dims_0"), val = tensor(false)]; tensor v2_25_cast_fp16 = reduce_sum(axes = v2_25_axes_0, keep_dims = v2_25_keep_dims_0, x = var_1005_cast_fp16)[name = tensor("v2_25_cast_fp16")]; - tensor var_1008_cast_fp16 = mul(x = dx2_25_cast_fp16, y = weights_53_cast_fp16)[name = tensor("op_1008_cast_fp16")]; + tensor var_1008_cast_fp16 = mul(x = dx2_25_cast_fp16, y = weights_53_cast_fp16)[name = tensor("op_1008_cast_fp16")]; tensor var_1010_axes_0 = const()[name = tensor("op_1010_axes_0"), val = tensor([2])]; tensor var_1010_keep_dims_0 = const()[name = tensor("op_1010_keep_dims_0"), val = tensor(false)]; tensor var_1010_cast_fp16 = reduce_sum(axes = var_1010_axes_0, keep_dims = var_1010_keep_dims_0, x = var_1008_cast_fp16)[name = tensor("op_1010_cast_fp16")]; @@ -835,31 +835,31 @@ program(1.0) tensor var_1018_interleave_0 = const()[name = tensor("op_1018_interleave_0"), val = tensor(false)]; tensor var_1018_cast_fp16 = concat(axis = var_72, interleave = var_1018_interleave_0, values = (mean_25_cast_fp16, std_25_cast_fp16))[name = tensor("op_1018_cast_fp16")]; tensor var_1020_begin_0 = const()[name = tensor("op_1020_begin_0"), val = tensor([0, 13, 0])]; - tensor var_1020_end_0 = const()[name = tensor("op_1020_end_0"), val = tensor([1, 14, 125])]; + tensor var_1020_end_0 = const()[name = tensor("op_1020_end_0"), val = tensor([1, 14, 375])]; tensor var_1020_end_mask_0 = const()[name = tensor("op_1020_end_mask_0"), val = tensor([true, false, true])]; tensor var_1020_squeeze_mask_0 = const()[name = tensor("op_1020_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1020_cast_fp16 = slice_by_index(begin = var_1020_begin_0, end = var_1020_end_0, end_mask = var_1020_end_mask_0, squeeze_mask = var_1020_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1020_cast_fp16")]; + tensor var_1020_cast_fp16 = slice_by_index(begin = var_1020_begin_0, end = var_1020_end_0, end_mask = var_1020_end_mask_0, squeeze_mask = var_1020_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1020_cast_fp16")]; tensor weights_57_axes_0 = const()[name = tensor("weights_57_axes_0"), val = tensor([1])]; - tensor weights_57_cast_fp16 = expand_dims(axes = weights_57_axes_0, x = var_1020_cast_fp16)[name = tensor("weights_57_cast_fp16")]; + tensor weights_57_cast_fp16 = expand_dims(axes = weights_57_axes_0, x = var_1020_cast_fp16)[name = tensor("weights_57_cast_fp16")]; tensor var_1024_axes_0 = const()[name = tensor("op_1024_axes_0"), val = tensor([2])]; tensor var_1024_keep_dims_0 = const()[name = tensor("op_1024_keep_dims_0"), val = tensor(false)]; tensor var_1024_cast_fp16 = reduce_sum(axes = var_1024_axes_0, keep_dims = var_1024_keep_dims_0, x = weights_57_cast_fp16)[name = tensor("op_1024_cast_fp16")]; tensor var_1025_to_fp16 = const()[name = tensor("op_1025_to_fp16"), val = tensor(0x1p-24)]; tensor v1_27_cast_fp16 = add(x = var_1024_cast_fp16, y = var_1025_to_fp16)[name = tensor("v1_27_cast_fp16")]; - tensor var_1027_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_57_cast_fp16)[name = tensor("op_1027_cast_fp16")]; + tensor var_1027_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_57_cast_fp16)[name = tensor("op_1027_cast_fp16")]; tensor var_1029_axes_0 = const()[name = tensor("op_1029_axes_0"), val = tensor([2])]; tensor var_1029_keep_dims_0 = const()[name = tensor("op_1029_keep_dims_0"), val = tensor(false)]; tensor var_1029_cast_fp16 = reduce_sum(axes = var_1029_axes_0, keep_dims = var_1029_keep_dims_0, x = var_1027_cast_fp16)[name = tensor("op_1029_cast_fp16")]; tensor mean_27_cast_fp16 = real_div(x = var_1029_cast_fp16, y = v1_27_cast_fp16)[name = tensor("mean_27_cast_fp16")]; tensor var_1031_axes_0 = const()[name = tensor("op_1031_axes_0"), val = tensor([2])]; tensor var_1031_cast_fp16 = expand_dims(axes = var_1031_axes_0, x = mean_27_cast_fp16)[name = tensor("op_1031_cast_fp16")]; - tensor var_1032_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1031_cast_fp16)[name = tensor("op_1032_cast_fp16")]; - tensor dx2_27_cast_fp16 = mul(x = var_1032_cast_fp16, y = var_1032_cast_fp16)[name = tensor("dx2_27_cast_fp16")]; - tensor var_1034_cast_fp16 = mul(x = weights_57_cast_fp16, y = weights_57_cast_fp16)[name = tensor("op_1034_cast_fp16")]; + tensor var_1032_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1031_cast_fp16)[name = tensor("op_1032_cast_fp16")]; + tensor dx2_27_cast_fp16 = mul(x = var_1032_cast_fp16, y = var_1032_cast_fp16)[name = tensor("dx2_27_cast_fp16")]; + tensor var_1034_cast_fp16 = mul(x = weights_57_cast_fp16, y = weights_57_cast_fp16)[name = tensor("op_1034_cast_fp16")]; tensor v2_27_axes_0 = const()[name = tensor("v2_27_axes_0"), val = tensor([2])]; tensor v2_27_keep_dims_0 = const()[name = tensor("v2_27_keep_dims_0"), val = tensor(false)]; tensor v2_27_cast_fp16 = reduce_sum(axes = v2_27_axes_0, keep_dims = v2_27_keep_dims_0, x = var_1034_cast_fp16)[name = tensor("v2_27_cast_fp16")]; - tensor var_1037_cast_fp16 = mul(x = dx2_27_cast_fp16, y = weights_57_cast_fp16)[name = tensor("op_1037_cast_fp16")]; + tensor var_1037_cast_fp16 = mul(x = dx2_27_cast_fp16, y = weights_57_cast_fp16)[name = tensor("op_1037_cast_fp16")]; tensor var_1039_axes_0 = const()[name = tensor("op_1039_axes_0"), val = tensor([2])]; tensor var_1039_keep_dims_0 = const()[name = tensor("op_1039_keep_dims_0"), val = tensor(false)]; tensor var_1039_cast_fp16 = reduce_sum(axes = var_1039_axes_0, keep_dims = var_1039_keep_dims_0, x = var_1037_cast_fp16)[name = tensor("op_1039_cast_fp16")]; @@ -872,31 +872,31 @@ program(1.0) tensor var_1047_interleave_0 = const()[name = tensor("op_1047_interleave_0"), val = tensor(false)]; tensor var_1047_cast_fp16 = concat(axis = var_72, interleave = var_1047_interleave_0, values = (mean_27_cast_fp16, std_27_cast_fp16))[name = tensor("op_1047_cast_fp16")]; tensor var_1049_begin_0 = const()[name = tensor("op_1049_begin_0"), val = tensor([0, 14, 0])]; - tensor var_1049_end_0 = const()[name = tensor("op_1049_end_0"), val = tensor([1, 15, 125])]; + tensor var_1049_end_0 = const()[name = tensor("op_1049_end_0"), val = tensor([1, 15, 375])]; tensor var_1049_end_mask_0 = const()[name = tensor("op_1049_end_mask_0"), val = tensor([true, false, true])]; tensor var_1049_squeeze_mask_0 = const()[name = tensor("op_1049_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1049_cast_fp16 = slice_by_index(begin = var_1049_begin_0, end = var_1049_end_0, end_mask = var_1049_end_mask_0, squeeze_mask = var_1049_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1049_cast_fp16")]; + tensor var_1049_cast_fp16 = slice_by_index(begin = var_1049_begin_0, end = var_1049_end_0, end_mask = var_1049_end_mask_0, squeeze_mask = var_1049_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1049_cast_fp16")]; tensor weights_61_axes_0 = const()[name = tensor("weights_61_axes_0"), val = tensor([1])]; - tensor weights_61_cast_fp16 = expand_dims(axes = weights_61_axes_0, x = var_1049_cast_fp16)[name = tensor("weights_61_cast_fp16")]; + tensor weights_61_cast_fp16 = expand_dims(axes = weights_61_axes_0, x = var_1049_cast_fp16)[name = tensor("weights_61_cast_fp16")]; tensor var_1053_axes_0 = const()[name = tensor("op_1053_axes_0"), val = tensor([2])]; tensor var_1053_keep_dims_0 = const()[name = tensor("op_1053_keep_dims_0"), val = tensor(false)]; tensor var_1053_cast_fp16 = reduce_sum(axes = var_1053_axes_0, keep_dims = var_1053_keep_dims_0, x = weights_61_cast_fp16)[name = tensor("op_1053_cast_fp16")]; tensor var_1054_to_fp16 = const()[name = tensor("op_1054_to_fp16"), val = tensor(0x1p-24)]; tensor v1_29_cast_fp16 = add(x = var_1053_cast_fp16, y = var_1054_to_fp16)[name = tensor("v1_29_cast_fp16")]; - tensor var_1056_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_61_cast_fp16)[name = tensor("op_1056_cast_fp16")]; + tensor var_1056_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_61_cast_fp16)[name = tensor("op_1056_cast_fp16")]; tensor var_1058_axes_0 = const()[name = tensor("op_1058_axes_0"), val = tensor([2])]; tensor var_1058_keep_dims_0 = const()[name = tensor("op_1058_keep_dims_0"), val = tensor(false)]; tensor var_1058_cast_fp16 = reduce_sum(axes = var_1058_axes_0, keep_dims = var_1058_keep_dims_0, x = var_1056_cast_fp16)[name = tensor("op_1058_cast_fp16")]; tensor mean_29_cast_fp16 = real_div(x = var_1058_cast_fp16, y = v1_29_cast_fp16)[name = tensor("mean_29_cast_fp16")]; tensor var_1060_axes_0 = const()[name = tensor("op_1060_axes_0"), val = tensor([2])]; tensor var_1060_cast_fp16 = expand_dims(axes = var_1060_axes_0, x = mean_29_cast_fp16)[name = tensor("op_1060_cast_fp16")]; - tensor var_1061_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1060_cast_fp16)[name = tensor("op_1061_cast_fp16")]; - tensor dx2_29_cast_fp16 = mul(x = var_1061_cast_fp16, y = var_1061_cast_fp16)[name = tensor("dx2_29_cast_fp16")]; - tensor var_1063_cast_fp16 = mul(x = weights_61_cast_fp16, y = weights_61_cast_fp16)[name = tensor("op_1063_cast_fp16")]; + tensor var_1061_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1060_cast_fp16)[name = tensor("op_1061_cast_fp16")]; + tensor dx2_29_cast_fp16 = mul(x = var_1061_cast_fp16, y = var_1061_cast_fp16)[name = tensor("dx2_29_cast_fp16")]; + tensor var_1063_cast_fp16 = mul(x = weights_61_cast_fp16, y = weights_61_cast_fp16)[name = tensor("op_1063_cast_fp16")]; tensor v2_29_axes_0 = const()[name = tensor("v2_29_axes_0"), val = tensor([2])]; tensor v2_29_keep_dims_0 = const()[name = tensor("v2_29_keep_dims_0"), val = tensor(false)]; tensor v2_29_cast_fp16 = reduce_sum(axes = v2_29_axes_0, keep_dims = v2_29_keep_dims_0, x = var_1063_cast_fp16)[name = tensor("v2_29_cast_fp16")]; - tensor var_1066_cast_fp16 = mul(x = dx2_29_cast_fp16, y = weights_61_cast_fp16)[name = tensor("op_1066_cast_fp16")]; + tensor var_1066_cast_fp16 = mul(x = dx2_29_cast_fp16, y = weights_61_cast_fp16)[name = tensor("op_1066_cast_fp16")]; tensor var_1068_axes_0 = const()[name = tensor("op_1068_axes_0"), val = tensor([2])]; tensor var_1068_keep_dims_0 = const()[name = tensor("op_1068_keep_dims_0"), val = tensor(false)]; tensor var_1068_cast_fp16 = reduce_sum(axes = var_1068_axes_0, keep_dims = var_1068_keep_dims_0, x = var_1066_cast_fp16)[name = tensor("op_1068_cast_fp16")]; @@ -909,31 +909,31 @@ program(1.0) tensor var_1076_interleave_0 = const()[name = tensor("op_1076_interleave_0"), val = tensor(false)]; tensor var_1076_cast_fp16 = concat(axis = var_72, interleave = var_1076_interleave_0, values = (mean_29_cast_fp16, std_29_cast_fp16))[name = tensor("op_1076_cast_fp16")]; tensor var_1078_begin_0 = const()[name = tensor("op_1078_begin_0"), val = tensor([0, 15, 0])]; - tensor var_1078_end_0 = const()[name = tensor("op_1078_end_0"), val = tensor([1, 16, 125])]; + tensor var_1078_end_0 = const()[name = tensor("op_1078_end_0"), val = tensor([1, 16, 375])]; tensor var_1078_end_mask_0 = const()[name = tensor("op_1078_end_mask_0"), val = tensor([true, false, true])]; tensor var_1078_squeeze_mask_0 = const()[name = tensor("op_1078_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1078_cast_fp16 = slice_by_index(begin = var_1078_begin_0, end = var_1078_end_0, end_mask = var_1078_end_mask_0, squeeze_mask = var_1078_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1078_cast_fp16")]; + tensor var_1078_cast_fp16 = slice_by_index(begin = var_1078_begin_0, end = var_1078_end_0, end_mask = var_1078_end_mask_0, squeeze_mask = var_1078_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1078_cast_fp16")]; tensor weights_65_axes_0 = const()[name = tensor("weights_65_axes_0"), val = tensor([1])]; - tensor weights_65_cast_fp16 = expand_dims(axes = weights_65_axes_0, x = var_1078_cast_fp16)[name = tensor("weights_65_cast_fp16")]; + tensor weights_65_cast_fp16 = expand_dims(axes = weights_65_axes_0, x = var_1078_cast_fp16)[name = tensor("weights_65_cast_fp16")]; tensor var_1082_axes_0 = const()[name = tensor("op_1082_axes_0"), val = tensor([2])]; tensor var_1082_keep_dims_0 = const()[name = tensor("op_1082_keep_dims_0"), val = tensor(false)]; tensor var_1082_cast_fp16 = reduce_sum(axes = var_1082_axes_0, keep_dims = var_1082_keep_dims_0, x = weights_65_cast_fp16)[name = tensor("op_1082_cast_fp16")]; tensor var_1083_to_fp16 = const()[name = tensor("op_1083_to_fp16"), val = tensor(0x1p-24)]; tensor v1_31_cast_fp16 = add(x = var_1082_cast_fp16, y = var_1083_to_fp16)[name = tensor("v1_31_cast_fp16")]; - tensor var_1085_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_65_cast_fp16)[name = tensor("op_1085_cast_fp16")]; + tensor var_1085_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_65_cast_fp16)[name = tensor("op_1085_cast_fp16")]; tensor var_1087_axes_0 = const()[name = tensor("op_1087_axes_0"), val = tensor([2])]; tensor var_1087_keep_dims_0 = const()[name = tensor("op_1087_keep_dims_0"), val = tensor(false)]; tensor var_1087_cast_fp16 = reduce_sum(axes = var_1087_axes_0, keep_dims = var_1087_keep_dims_0, x = var_1085_cast_fp16)[name = tensor("op_1087_cast_fp16")]; tensor mean_31_cast_fp16 = real_div(x = var_1087_cast_fp16, y = v1_31_cast_fp16)[name = tensor("mean_31_cast_fp16")]; tensor var_1089_axes_0 = const()[name = tensor("op_1089_axes_0"), val = tensor([2])]; tensor var_1089_cast_fp16 = expand_dims(axes = var_1089_axes_0, x = mean_31_cast_fp16)[name = tensor("op_1089_cast_fp16")]; - tensor var_1090_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1089_cast_fp16)[name = tensor("op_1090_cast_fp16")]; - tensor dx2_31_cast_fp16 = mul(x = var_1090_cast_fp16, y = var_1090_cast_fp16)[name = tensor("dx2_31_cast_fp16")]; - tensor var_1092_cast_fp16 = mul(x = weights_65_cast_fp16, y = weights_65_cast_fp16)[name = tensor("op_1092_cast_fp16")]; + tensor var_1090_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1089_cast_fp16)[name = tensor("op_1090_cast_fp16")]; + tensor dx2_31_cast_fp16 = mul(x = var_1090_cast_fp16, y = var_1090_cast_fp16)[name = tensor("dx2_31_cast_fp16")]; + tensor var_1092_cast_fp16 = mul(x = weights_65_cast_fp16, y = weights_65_cast_fp16)[name = tensor("op_1092_cast_fp16")]; tensor v2_31_axes_0 = const()[name = tensor("v2_31_axes_0"), val = tensor([2])]; tensor v2_31_keep_dims_0 = const()[name = tensor("v2_31_keep_dims_0"), val = tensor(false)]; tensor v2_31_cast_fp16 = reduce_sum(axes = v2_31_axes_0, keep_dims = v2_31_keep_dims_0, x = var_1092_cast_fp16)[name = tensor("v2_31_cast_fp16")]; - tensor var_1095_cast_fp16 = mul(x = dx2_31_cast_fp16, y = weights_65_cast_fp16)[name = tensor("op_1095_cast_fp16")]; + tensor var_1095_cast_fp16 = mul(x = dx2_31_cast_fp16, y = weights_65_cast_fp16)[name = tensor("op_1095_cast_fp16")]; tensor var_1097_axes_0 = const()[name = tensor("op_1097_axes_0"), val = tensor([2])]; tensor var_1097_keep_dims_0 = const()[name = tensor("op_1097_keep_dims_0"), val = tensor(false)]; tensor var_1097_cast_fp16 = reduce_sum(axes = var_1097_axes_0, keep_dims = var_1097_keep_dims_0, x = var_1095_cast_fp16)[name = tensor("op_1097_cast_fp16")]; @@ -946,31 +946,31 @@ program(1.0) tensor var_1105_interleave_0 = const()[name = tensor("op_1105_interleave_0"), val = tensor(false)]; tensor var_1105_cast_fp16 = concat(axis = var_72, interleave = var_1105_interleave_0, values = (mean_31_cast_fp16, std_31_cast_fp16))[name = tensor("op_1105_cast_fp16")]; tensor var_1107_begin_0 = const()[name = tensor("op_1107_begin_0"), val = tensor([0, 16, 0])]; - tensor var_1107_end_0 = const()[name = tensor("op_1107_end_0"), val = tensor([1, 17, 125])]; + tensor var_1107_end_0 = const()[name = tensor("op_1107_end_0"), val = tensor([1, 17, 375])]; tensor var_1107_end_mask_0 = const()[name = tensor("op_1107_end_mask_0"), val = tensor([true, false, true])]; tensor var_1107_squeeze_mask_0 = const()[name = tensor("op_1107_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1107_cast_fp16 = slice_by_index(begin = var_1107_begin_0, end = var_1107_end_0, end_mask = var_1107_end_mask_0, squeeze_mask = var_1107_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1107_cast_fp16")]; + tensor var_1107_cast_fp16 = slice_by_index(begin = var_1107_begin_0, end = var_1107_end_0, end_mask = var_1107_end_mask_0, squeeze_mask = var_1107_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1107_cast_fp16")]; tensor weights_69_axes_0 = const()[name = tensor("weights_69_axes_0"), val = tensor([1])]; - tensor weights_69_cast_fp16 = expand_dims(axes = weights_69_axes_0, x = var_1107_cast_fp16)[name = tensor("weights_69_cast_fp16")]; + tensor weights_69_cast_fp16 = expand_dims(axes = weights_69_axes_0, x = var_1107_cast_fp16)[name = tensor("weights_69_cast_fp16")]; tensor var_1111_axes_0 = const()[name = tensor("op_1111_axes_0"), val = tensor([2])]; tensor var_1111_keep_dims_0 = const()[name = tensor("op_1111_keep_dims_0"), val = tensor(false)]; tensor var_1111_cast_fp16 = reduce_sum(axes = var_1111_axes_0, keep_dims = var_1111_keep_dims_0, x = weights_69_cast_fp16)[name = tensor("op_1111_cast_fp16")]; tensor var_1112_to_fp16 = const()[name = tensor("op_1112_to_fp16"), val = tensor(0x1p-24)]; tensor v1_33_cast_fp16 = add(x = var_1111_cast_fp16, y = var_1112_to_fp16)[name = tensor("v1_33_cast_fp16")]; - tensor var_1114_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_69_cast_fp16)[name = tensor("op_1114_cast_fp16")]; + tensor var_1114_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_69_cast_fp16)[name = tensor("op_1114_cast_fp16")]; tensor var_1116_axes_0 = const()[name = tensor("op_1116_axes_0"), val = tensor([2])]; tensor var_1116_keep_dims_0 = const()[name = tensor("op_1116_keep_dims_0"), val = tensor(false)]; tensor var_1116_cast_fp16 = reduce_sum(axes = var_1116_axes_0, keep_dims = var_1116_keep_dims_0, x = var_1114_cast_fp16)[name = tensor("op_1116_cast_fp16")]; tensor mean_33_cast_fp16 = real_div(x = var_1116_cast_fp16, y = v1_33_cast_fp16)[name = tensor("mean_33_cast_fp16")]; tensor var_1118_axes_0 = const()[name = tensor("op_1118_axes_0"), val = tensor([2])]; tensor var_1118_cast_fp16 = expand_dims(axes = var_1118_axes_0, x = mean_33_cast_fp16)[name = tensor("op_1118_cast_fp16")]; - tensor var_1119_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1118_cast_fp16)[name = tensor("op_1119_cast_fp16")]; - tensor dx2_33_cast_fp16 = mul(x = var_1119_cast_fp16, y = var_1119_cast_fp16)[name = tensor("dx2_33_cast_fp16")]; - tensor var_1121_cast_fp16 = mul(x = weights_69_cast_fp16, y = weights_69_cast_fp16)[name = tensor("op_1121_cast_fp16")]; + tensor var_1119_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1118_cast_fp16)[name = tensor("op_1119_cast_fp16")]; + tensor dx2_33_cast_fp16 = mul(x = var_1119_cast_fp16, y = var_1119_cast_fp16)[name = tensor("dx2_33_cast_fp16")]; + tensor var_1121_cast_fp16 = mul(x = weights_69_cast_fp16, y = weights_69_cast_fp16)[name = tensor("op_1121_cast_fp16")]; tensor v2_33_axes_0 = const()[name = tensor("v2_33_axes_0"), val = tensor([2])]; tensor v2_33_keep_dims_0 = const()[name = tensor("v2_33_keep_dims_0"), val = tensor(false)]; tensor v2_33_cast_fp16 = reduce_sum(axes = v2_33_axes_0, keep_dims = v2_33_keep_dims_0, x = var_1121_cast_fp16)[name = tensor("v2_33_cast_fp16")]; - tensor var_1124_cast_fp16 = mul(x = dx2_33_cast_fp16, y = weights_69_cast_fp16)[name = tensor("op_1124_cast_fp16")]; + tensor var_1124_cast_fp16 = mul(x = dx2_33_cast_fp16, y = weights_69_cast_fp16)[name = tensor("op_1124_cast_fp16")]; tensor var_1126_axes_0 = const()[name = tensor("op_1126_axes_0"), val = tensor([2])]; tensor var_1126_keep_dims_0 = const()[name = tensor("op_1126_keep_dims_0"), val = tensor(false)]; tensor var_1126_cast_fp16 = reduce_sum(axes = var_1126_axes_0, keep_dims = var_1126_keep_dims_0, x = var_1124_cast_fp16)[name = tensor("op_1126_cast_fp16")]; @@ -983,31 +983,31 @@ program(1.0) tensor var_1134_interleave_0 = const()[name = tensor("op_1134_interleave_0"), val = tensor(false)]; tensor var_1134_cast_fp16 = concat(axis = var_72, interleave = var_1134_interleave_0, values = (mean_33_cast_fp16, std_33_cast_fp16))[name = tensor("op_1134_cast_fp16")]; tensor var_1136_begin_0 = const()[name = tensor("op_1136_begin_0"), val = tensor([0, 17, 0])]; - tensor var_1136_end_0 = const()[name = tensor("op_1136_end_0"), val = tensor([1, 18, 125])]; + tensor var_1136_end_0 = const()[name = tensor("op_1136_end_0"), val = tensor([1, 18, 375])]; tensor var_1136_end_mask_0 = const()[name = tensor("op_1136_end_mask_0"), val = tensor([true, false, true])]; tensor var_1136_squeeze_mask_0 = const()[name = tensor("op_1136_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1136_cast_fp16 = slice_by_index(begin = var_1136_begin_0, end = var_1136_end_0, end_mask = var_1136_end_mask_0, squeeze_mask = var_1136_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1136_cast_fp16")]; + tensor var_1136_cast_fp16 = slice_by_index(begin = var_1136_begin_0, end = var_1136_end_0, end_mask = var_1136_end_mask_0, squeeze_mask = var_1136_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1136_cast_fp16")]; tensor weights_73_axes_0 = const()[name = tensor("weights_73_axes_0"), val = tensor([1])]; - tensor weights_73_cast_fp16 = expand_dims(axes = weights_73_axes_0, x = var_1136_cast_fp16)[name = tensor("weights_73_cast_fp16")]; + tensor weights_73_cast_fp16 = expand_dims(axes = weights_73_axes_0, x = var_1136_cast_fp16)[name = tensor("weights_73_cast_fp16")]; tensor var_1140_axes_0 = const()[name = tensor("op_1140_axes_0"), val = tensor([2])]; tensor var_1140_keep_dims_0 = const()[name = tensor("op_1140_keep_dims_0"), val = tensor(false)]; tensor var_1140_cast_fp16 = reduce_sum(axes = var_1140_axes_0, keep_dims = var_1140_keep_dims_0, x = weights_73_cast_fp16)[name = tensor("op_1140_cast_fp16")]; tensor var_1141_to_fp16 = const()[name = tensor("op_1141_to_fp16"), val = tensor(0x1p-24)]; tensor v1_35_cast_fp16 = add(x = var_1140_cast_fp16, y = var_1141_to_fp16)[name = tensor("v1_35_cast_fp16")]; - tensor var_1143_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_73_cast_fp16)[name = tensor("op_1143_cast_fp16")]; + tensor var_1143_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_73_cast_fp16)[name = tensor("op_1143_cast_fp16")]; tensor var_1145_axes_0 = const()[name = tensor("op_1145_axes_0"), val = tensor([2])]; tensor var_1145_keep_dims_0 = const()[name = tensor("op_1145_keep_dims_0"), val = tensor(false)]; tensor var_1145_cast_fp16 = reduce_sum(axes = var_1145_axes_0, keep_dims = var_1145_keep_dims_0, x = var_1143_cast_fp16)[name = tensor("op_1145_cast_fp16")]; tensor mean_35_cast_fp16 = real_div(x = var_1145_cast_fp16, y = v1_35_cast_fp16)[name = tensor("mean_35_cast_fp16")]; tensor var_1147_axes_0 = const()[name = tensor("op_1147_axes_0"), val = tensor([2])]; tensor var_1147_cast_fp16 = expand_dims(axes = var_1147_axes_0, x = mean_35_cast_fp16)[name = tensor("op_1147_cast_fp16")]; - tensor var_1148_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1147_cast_fp16)[name = tensor("op_1148_cast_fp16")]; - tensor dx2_35_cast_fp16 = mul(x = var_1148_cast_fp16, y = var_1148_cast_fp16)[name = tensor("dx2_35_cast_fp16")]; - tensor var_1150_cast_fp16 = mul(x = weights_73_cast_fp16, y = weights_73_cast_fp16)[name = tensor("op_1150_cast_fp16")]; + tensor var_1148_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1147_cast_fp16)[name = tensor("op_1148_cast_fp16")]; + tensor dx2_35_cast_fp16 = mul(x = var_1148_cast_fp16, y = var_1148_cast_fp16)[name = tensor("dx2_35_cast_fp16")]; + tensor var_1150_cast_fp16 = mul(x = weights_73_cast_fp16, y = weights_73_cast_fp16)[name = tensor("op_1150_cast_fp16")]; tensor v2_35_axes_0 = const()[name = tensor("v2_35_axes_0"), val = tensor([2])]; tensor v2_35_keep_dims_0 = const()[name = tensor("v2_35_keep_dims_0"), val = tensor(false)]; tensor v2_35_cast_fp16 = reduce_sum(axes = v2_35_axes_0, keep_dims = v2_35_keep_dims_0, x = var_1150_cast_fp16)[name = tensor("v2_35_cast_fp16")]; - tensor var_1153_cast_fp16 = mul(x = dx2_35_cast_fp16, y = weights_73_cast_fp16)[name = tensor("op_1153_cast_fp16")]; + tensor var_1153_cast_fp16 = mul(x = dx2_35_cast_fp16, y = weights_73_cast_fp16)[name = tensor("op_1153_cast_fp16")]; tensor var_1155_axes_0 = const()[name = tensor("op_1155_axes_0"), val = tensor([2])]; tensor var_1155_keep_dims_0 = const()[name = tensor("op_1155_keep_dims_0"), val = tensor(false)]; tensor var_1155_cast_fp16 = reduce_sum(axes = var_1155_axes_0, keep_dims = var_1155_keep_dims_0, x = var_1153_cast_fp16)[name = tensor("op_1155_cast_fp16")]; @@ -1020,31 +1020,31 @@ program(1.0) tensor var_1163_interleave_0 = const()[name = tensor("op_1163_interleave_0"), val = tensor(false)]; tensor var_1163_cast_fp16 = concat(axis = var_72, interleave = var_1163_interleave_0, values = (mean_35_cast_fp16, std_35_cast_fp16))[name = tensor("op_1163_cast_fp16")]; tensor var_1165_begin_0 = const()[name = tensor("op_1165_begin_0"), val = tensor([0, 18, 0])]; - tensor var_1165_end_0 = const()[name = tensor("op_1165_end_0"), val = tensor([1, 19, 125])]; + tensor var_1165_end_0 = const()[name = tensor("op_1165_end_0"), val = tensor([1, 19, 375])]; tensor var_1165_end_mask_0 = const()[name = tensor("op_1165_end_mask_0"), val = tensor([true, false, true])]; tensor var_1165_squeeze_mask_0 = const()[name = tensor("op_1165_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1165_cast_fp16 = slice_by_index(begin = var_1165_begin_0, end = var_1165_end_0, end_mask = var_1165_end_mask_0, squeeze_mask = var_1165_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1165_cast_fp16")]; + tensor var_1165_cast_fp16 = slice_by_index(begin = var_1165_begin_0, end = var_1165_end_0, end_mask = var_1165_end_mask_0, squeeze_mask = var_1165_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1165_cast_fp16")]; tensor weights_77_axes_0 = const()[name = tensor("weights_77_axes_0"), val = tensor([1])]; - tensor weights_77_cast_fp16 = expand_dims(axes = weights_77_axes_0, x = var_1165_cast_fp16)[name = tensor("weights_77_cast_fp16")]; + tensor weights_77_cast_fp16 = expand_dims(axes = weights_77_axes_0, x = var_1165_cast_fp16)[name = tensor("weights_77_cast_fp16")]; tensor var_1169_axes_0 = const()[name = tensor("op_1169_axes_0"), val = tensor([2])]; tensor var_1169_keep_dims_0 = const()[name = tensor("op_1169_keep_dims_0"), val = tensor(false)]; tensor var_1169_cast_fp16 = reduce_sum(axes = var_1169_axes_0, keep_dims = var_1169_keep_dims_0, x = weights_77_cast_fp16)[name = tensor("op_1169_cast_fp16")]; tensor var_1170_to_fp16 = const()[name = tensor("op_1170_to_fp16"), val = tensor(0x1p-24)]; tensor v1_37_cast_fp16 = add(x = var_1169_cast_fp16, y = var_1170_to_fp16)[name = tensor("v1_37_cast_fp16")]; - tensor var_1172_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_77_cast_fp16)[name = tensor("op_1172_cast_fp16")]; + tensor var_1172_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_77_cast_fp16)[name = tensor("op_1172_cast_fp16")]; tensor var_1174_axes_0 = const()[name = tensor("op_1174_axes_0"), val = tensor([2])]; tensor var_1174_keep_dims_0 = const()[name = tensor("op_1174_keep_dims_0"), val = tensor(false)]; tensor var_1174_cast_fp16 = reduce_sum(axes = var_1174_axes_0, keep_dims = var_1174_keep_dims_0, x = var_1172_cast_fp16)[name = tensor("op_1174_cast_fp16")]; tensor mean_37_cast_fp16 = real_div(x = var_1174_cast_fp16, y = v1_37_cast_fp16)[name = tensor("mean_37_cast_fp16")]; tensor var_1176_axes_0 = const()[name = tensor("op_1176_axes_0"), val = tensor([2])]; tensor var_1176_cast_fp16 = expand_dims(axes = var_1176_axes_0, x = mean_37_cast_fp16)[name = tensor("op_1176_cast_fp16")]; - tensor var_1177_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1176_cast_fp16)[name = tensor("op_1177_cast_fp16")]; - tensor dx2_37_cast_fp16 = mul(x = var_1177_cast_fp16, y = var_1177_cast_fp16)[name = tensor("dx2_37_cast_fp16")]; - tensor var_1179_cast_fp16 = mul(x = weights_77_cast_fp16, y = weights_77_cast_fp16)[name = tensor("op_1179_cast_fp16")]; + tensor var_1177_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1176_cast_fp16)[name = tensor("op_1177_cast_fp16")]; + tensor dx2_37_cast_fp16 = mul(x = var_1177_cast_fp16, y = var_1177_cast_fp16)[name = tensor("dx2_37_cast_fp16")]; + tensor var_1179_cast_fp16 = mul(x = weights_77_cast_fp16, y = weights_77_cast_fp16)[name = tensor("op_1179_cast_fp16")]; tensor v2_37_axes_0 = const()[name = tensor("v2_37_axes_0"), val = tensor([2])]; tensor v2_37_keep_dims_0 = const()[name = tensor("v2_37_keep_dims_0"), val = tensor(false)]; tensor v2_37_cast_fp16 = reduce_sum(axes = v2_37_axes_0, keep_dims = v2_37_keep_dims_0, x = var_1179_cast_fp16)[name = tensor("v2_37_cast_fp16")]; - tensor var_1182_cast_fp16 = mul(x = dx2_37_cast_fp16, y = weights_77_cast_fp16)[name = tensor("op_1182_cast_fp16")]; + tensor var_1182_cast_fp16 = mul(x = dx2_37_cast_fp16, y = weights_77_cast_fp16)[name = tensor("op_1182_cast_fp16")]; tensor var_1184_axes_0 = const()[name = tensor("op_1184_axes_0"), val = tensor([2])]; tensor var_1184_keep_dims_0 = const()[name = tensor("op_1184_keep_dims_0"), val = tensor(false)]; tensor var_1184_cast_fp16 = reduce_sum(axes = var_1184_axes_0, keep_dims = var_1184_keep_dims_0, x = var_1182_cast_fp16)[name = tensor("op_1184_cast_fp16")]; @@ -1057,31 +1057,31 @@ program(1.0) tensor var_1192_interleave_0 = const()[name = tensor("op_1192_interleave_0"), val = tensor(false)]; tensor var_1192_cast_fp16 = concat(axis = var_72, interleave = var_1192_interleave_0, values = (mean_37_cast_fp16, std_37_cast_fp16))[name = tensor("op_1192_cast_fp16")]; tensor var_1194_begin_0 = const()[name = tensor("op_1194_begin_0"), val = tensor([0, 19, 0])]; - tensor var_1194_end_0 = const()[name = tensor("op_1194_end_0"), val = tensor([1, 20, 125])]; + tensor var_1194_end_0 = const()[name = tensor("op_1194_end_0"), val = tensor([1, 20, 375])]; tensor var_1194_end_mask_0 = const()[name = tensor("op_1194_end_mask_0"), val = tensor([true, false, true])]; tensor var_1194_squeeze_mask_0 = const()[name = tensor("op_1194_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, squeeze_mask = var_1194_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1194_cast_fp16")]; + tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, squeeze_mask = var_1194_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1194_cast_fp16")]; tensor weights_81_axes_0 = const()[name = tensor("weights_81_axes_0"), val = tensor([1])]; - tensor weights_81_cast_fp16 = expand_dims(axes = weights_81_axes_0, x = var_1194_cast_fp16)[name = tensor("weights_81_cast_fp16")]; + tensor weights_81_cast_fp16 = expand_dims(axes = weights_81_axes_0, x = var_1194_cast_fp16)[name = tensor("weights_81_cast_fp16")]; tensor var_1198_axes_0 = const()[name = tensor("op_1198_axes_0"), val = tensor([2])]; tensor var_1198_keep_dims_0 = const()[name = tensor("op_1198_keep_dims_0"), val = tensor(false)]; tensor var_1198_cast_fp16 = reduce_sum(axes = var_1198_axes_0, keep_dims = var_1198_keep_dims_0, x = weights_81_cast_fp16)[name = tensor("op_1198_cast_fp16")]; tensor var_1199_to_fp16 = const()[name = tensor("op_1199_to_fp16"), val = tensor(0x1p-24)]; tensor v1_39_cast_fp16 = add(x = var_1198_cast_fp16, y = var_1199_to_fp16)[name = tensor("v1_39_cast_fp16")]; - tensor var_1201_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_81_cast_fp16)[name = tensor("op_1201_cast_fp16")]; + tensor var_1201_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_81_cast_fp16)[name = tensor("op_1201_cast_fp16")]; tensor var_1203_axes_0 = const()[name = tensor("op_1203_axes_0"), val = tensor([2])]; tensor var_1203_keep_dims_0 = const()[name = tensor("op_1203_keep_dims_0"), val = tensor(false)]; tensor var_1203_cast_fp16 = reduce_sum(axes = var_1203_axes_0, keep_dims = var_1203_keep_dims_0, x = var_1201_cast_fp16)[name = tensor("op_1203_cast_fp16")]; tensor mean_39_cast_fp16 = real_div(x = var_1203_cast_fp16, y = v1_39_cast_fp16)[name = tensor("mean_39_cast_fp16")]; tensor var_1205_axes_0 = const()[name = tensor("op_1205_axes_0"), val = tensor([2])]; tensor var_1205_cast_fp16 = expand_dims(axes = var_1205_axes_0, x = mean_39_cast_fp16)[name = tensor("op_1205_cast_fp16")]; - tensor var_1206_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1205_cast_fp16)[name = tensor("op_1206_cast_fp16")]; - tensor dx2_39_cast_fp16 = mul(x = var_1206_cast_fp16, y = var_1206_cast_fp16)[name = tensor("dx2_39_cast_fp16")]; - tensor var_1208_cast_fp16 = mul(x = weights_81_cast_fp16, y = weights_81_cast_fp16)[name = tensor("op_1208_cast_fp16")]; + tensor var_1206_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1205_cast_fp16)[name = tensor("op_1206_cast_fp16")]; + tensor dx2_39_cast_fp16 = mul(x = var_1206_cast_fp16, y = var_1206_cast_fp16)[name = tensor("dx2_39_cast_fp16")]; + tensor var_1208_cast_fp16 = mul(x = weights_81_cast_fp16, y = weights_81_cast_fp16)[name = tensor("op_1208_cast_fp16")]; tensor v2_39_axes_0 = const()[name = tensor("v2_39_axes_0"), val = tensor([2])]; tensor v2_39_keep_dims_0 = const()[name = tensor("v2_39_keep_dims_0"), val = tensor(false)]; tensor v2_39_cast_fp16 = reduce_sum(axes = v2_39_axes_0, keep_dims = v2_39_keep_dims_0, x = var_1208_cast_fp16)[name = tensor("v2_39_cast_fp16")]; - tensor var_1211_cast_fp16 = mul(x = dx2_39_cast_fp16, y = weights_81_cast_fp16)[name = tensor("op_1211_cast_fp16")]; + tensor var_1211_cast_fp16 = mul(x = dx2_39_cast_fp16, y = weights_81_cast_fp16)[name = tensor("op_1211_cast_fp16")]; tensor var_1213_axes_0 = const()[name = tensor("op_1213_axes_0"), val = tensor([2])]; tensor var_1213_keep_dims_0 = const()[name = tensor("op_1213_keep_dims_0"), val = tensor(false)]; tensor var_1213_cast_fp16 = reduce_sum(axes = var_1213_axes_0, keep_dims = var_1213_keep_dims_0, x = var_1211_cast_fp16)[name = tensor("op_1213_cast_fp16")]; @@ -1094,31 +1094,31 @@ program(1.0) tensor var_1221_interleave_0 = const()[name = tensor("op_1221_interleave_0"), val = tensor(false)]; tensor var_1221_cast_fp16 = concat(axis = var_72, interleave = var_1221_interleave_0, values = (mean_39_cast_fp16, std_39_cast_fp16))[name = tensor("op_1221_cast_fp16")]; tensor var_1223_begin_0 = const()[name = tensor("op_1223_begin_0"), val = tensor([0, 20, 0])]; - tensor var_1223_end_0 = const()[name = tensor("op_1223_end_0"), val = tensor([1, 21, 125])]; + tensor var_1223_end_0 = const()[name = tensor("op_1223_end_0"), val = tensor([1, 21, 375])]; tensor var_1223_end_mask_0 = const()[name = tensor("op_1223_end_mask_0"), val = tensor([true, false, true])]; tensor var_1223_squeeze_mask_0 = const()[name = tensor("op_1223_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1223_cast_fp16 = slice_by_index(begin = var_1223_begin_0, end = var_1223_end_0, end_mask = var_1223_end_mask_0, squeeze_mask = var_1223_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1223_cast_fp16")]; + tensor var_1223_cast_fp16 = slice_by_index(begin = var_1223_begin_0, end = var_1223_end_0, end_mask = var_1223_end_mask_0, squeeze_mask = var_1223_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1223_cast_fp16")]; tensor weights_85_axes_0 = const()[name = tensor("weights_85_axes_0"), val = tensor([1])]; - tensor weights_85_cast_fp16 = expand_dims(axes = weights_85_axes_0, x = var_1223_cast_fp16)[name = tensor("weights_85_cast_fp16")]; + tensor weights_85_cast_fp16 = expand_dims(axes = weights_85_axes_0, x = var_1223_cast_fp16)[name = tensor("weights_85_cast_fp16")]; tensor var_1227_axes_0 = const()[name = tensor("op_1227_axes_0"), val = tensor([2])]; tensor var_1227_keep_dims_0 = const()[name = tensor("op_1227_keep_dims_0"), val = tensor(false)]; tensor var_1227_cast_fp16 = reduce_sum(axes = var_1227_axes_0, keep_dims = var_1227_keep_dims_0, x = weights_85_cast_fp16)[name = tensor("op_1227_cast_fp16")]; tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(0x1p-24)]; tensor v1_41_cast_fp16 = add(x = var_1227_cast_fp16, y = var_1228_to_fp16)[name = tensor("v1_41_cast_fp16")]; - tensor var_1230_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_85_cast_fp16)[name = tensor("op_1230_cast_fp16")]; + tensor var_1230_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_85_cast_fp16)[name = tensor("op_1230_cast_fp16")]; tensor var_1232_axes_0 = const()[name = tensor("op_1232_axes_0"), val = tensor([2])]; tensor var_1232_keep_dims_0 = const()[name = tensor("op_1232_keep_dims_0"), val = tensor(false)]; tensor var_1232_cast_fp16 = reduce_sum(axes = var_1232_axes_0, keep_dims = var_1232_keep_dims_0, x = var_1230_cast_fp16)[name = tensor("op_1232_cast_fp16")]; tensor mean_41_cast_fp16 = real_div(x = var_1232_cast_fp16, y = v1_41_cast_fp16)[name = tensor("mean_41_cast_fp16")]; tensor var_1234_axes_0 = const()[name = tensor("op_1234_axes_0"), val = tensor([2])]; tensor var_1234_cast_fp16 = expand_dims(axes = var_1234_axes_0, x = mean_41_cast_fp16)[name = tensor("op_1234_cast_fp16")]; - tensor var_1235_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1234_cast_fp16)[name = tensor("op_1235_cast_fp16")]; - tensor dx2_41_cast_fp16 = mul(x = var_1235_cast_fp16, y = var_1235_cast_fp16)[name = tensor("dx2_41_cast_fp16")]; - tensor var_1237_cast_fp16 = mul(x = weights_85_cast_fp16, y = weights_85_cast_fp16)[name = tensor("op_1237_cast_fp16")]; + tensor var_1235_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1234_cast_fp16)[name = tensor("op_1235_cast_fp16")]; + tensor dx2_41_cast_fp16 = mul(x = var_1235_cast_fp16, y = var_1235_cast_fp16)[name = tensor("dx2_41_cast_fp16")]; + tensor var_1237_cast_fp16 = mul(x = weights_85_cast_fp16, y = weights_85_cast_fp16)[name = tensor("op_1237_cast_fp16")]; tensor v2_41_axes_0 = const()[name = tensor("v2_41_axes_0"), val = tensor([2])]; tensor v2_41_keep_dims_0 = const()[name = tensor("v2_41_keep_dims_0"), val = tensor(false)]; tensor v2_41_cast_fp16 = reduce_sum(axes = v2_41_axes_0, keep_dims = v2_41_keep_dims_0, x = var_1237_cast_fp16)[name = tensor("v2_41_cast_fp16")]; - tensor var_1240_cast_fp16 = mul(x = dx2_41_cast_fp16, y = weights_85_cast_fp16)[name = tensor("op_1240_cast_fp16")]; + tensor var_1240_cast_fp16 = mul(x = dx2_41_cast_fp16, y = weights_85_cast_fp16)[name = tensor("op_1240_cast_fp16")]; tensor var_1242_axes_0 = const()[name = tensor("op_1242_axes_0"), val = tensor([2])]; tensor var_1242_keep_dims_0 = const()[name = tensor("op_1242_keep_dims_0"), val = tensor(false)]; tensor var_1242_cast_fp16 = reduce_sum(axes = var_1242_axes_0, keep_dims = var_1242_keep_dims_0, x = var_1240_cast_fp16)[name = tensor("op_1242_cast_fp16")]; @@ -1131,31 +1131,31 @@ program(1.0) tensor var_1250_interleave_0 = const()[name = tensor("op_1250_interleave_0"), val = tensor(false)]; tensor var_1250_cast_fp16 = concat(axis = var_72, interleave = var_1250_interleave_0, values = (mean_41_cast_fp16, std_41_cast_fp16))[name = tensor("op_1250_cast_fp16")]; tensor var_1252_begin_0 = const()[name = tensor("op_1252_begin_0"), val = tensor([0, 21, 0])]; - tensor var_1252_end_0 = const()[name = tensor("op_1252_end_0"), val = tensor([1, 22, 125])]; + tensor var_1252_end_0 = const()[name = tensor("op_1252_end_0"), val = tensor([1, 22, 375])]; tensor var_1252_end_mask_0 = const()[name = tensor("op_1252_end_mask_0"), val = tensor([true, false, true])]; tensor var_1252_squeeze_mask_0 = const()[name = tensor("op_1252_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1252_cast_fp16 = slice_by_index(begin = var_1252_begin_0, end = var_1252_end_0, end_mask = var_1252_end_mask_0, squeeze_mask = var_1252_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1252_cast_fp16")]; + tensor var_1252_cast_fp16 = slice_by_index(begin = var_1252_begin_0, end = var_1252_end_0, end_mask = var_1252_end_mask_0, squeeze_mask = var_1252_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1252_cast_fp16")]; tensor weights_89_axes_0 = const()[name = tensor("weights_89_axes_0"), val = tensor([1])]; - tensor weights_89_cast_fp16 = expand_dims(axes = weights_89_axes_0, x = var_1252_cast_fp16)[name = tensor("weights_89_cast_fp16")]; + tensor weights_89_cast_fp16 = expand_dims(axes = weights_89_axes_0, x = var_1252_cast_fp16)[name = tensor("weights_89_cast_fp16")]; tensor var_1256_axes_0 = const()[name = tensor("op_1256_axes_0"), val = tensor([2])]; tensor var_1256_keep_dims_0 = const()[name = tensor("op_1256_keep_dims_0"), val = tensor(false)]; tensor var_1256_cast_fp16 = reduce_sum(axes = var_1256_axes_0, keep_dims = var_1256_keep_dims_0, x = weights_89_cast_fp16)[name = tensor("op_1256_cast_fp16")]; tensor var_1257_to_fp16 = const()[name = tensor("op_1257_to_fp16"), val = tensor(0x1p-24)]; tensor v1_43_cast_fp16 = add(x = var_1256_cast_fp16, y = var_1257_to_fp16)[name = tensor("v1_43_cast_fp16")]; - tensor var_1259_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_89_cast_fp16)[name = tensor("op_1259_cast_fp16")]; + tensor var_1259_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_89_cast_fp16)[name = tensor("op_1259_cast_fp16")]; tensor var_1261_axes_0 = const()[name = tensor("op_1261_axes_0"), val = tensor([2])]; tensor var_1261_keep_dims_0 = const()[name = tensor("op_1261_keep_dims_0"), val = tensor(false)]; tensor var_1261_cast_fp16 = reduce_sum(axes = var_1261_axes_0, keep_dims = var_1261_keep_dims_0, x = var_1259_cast_fp16)[name = tensor("op_1261_cast_fp16")]; tensor mean_43_cast_fp16 = real_div(x = var_1261_cast_fp16, y = v1_43_cast_fp16)[name = tensor("mean_43_cast_fp16")]; tensor var_1263_axes_0 = const()[name = tensor("op_1263_axes_0"), val = tensor([2])]; tensor var_1263_cast_fp16 = expand_dims(axes = var_1263_axes_0, x = mean_43_cast_fp16)[name = tensor("op_1263_cast_fp16")]; - tensor var_1264_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1263_cast_fp16)[name = tensor("op_1264_cast_fp16")]; - tensor dx2_43_cast_fp16 = mul(x = var_1264_cast_fp16, y = var_1264_cast_fp16)[name = tensor("dx2_43_cast_fp16")]; - tensor var_1266_cast_fp16 = mul(x = weights_89_cast_fp16, y = weights_89_cast_fp16)[name = tensor("op_1266_cast_fp16")]; + tensor var_1264_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1263_cast_fp16)[name = tensor("op_1264_cast_fp16")]; + tensor dx2_43_cast_fp16 = mul(x = var_1264_cast_fp16, y = var_1264_cast_fp16)[name = tensor("dx2_43_cast_fp16")]; + tensor var_1266_cast_fp16 = mul(x = weights_89_cast_fp16, y = weights_89_cast_fp16)[name = tensor("op_1266_cast_fp16")]; tensor v2_43_axes_0 = const()[name = tensor("v2_43_axes_0"), val = tensor([2])]; tensor v2_43_keep_dims_0 = const()[name = tensor("v2_43_keep_dims_0"), val = tensor(false)]; tensor v2_43_cast_fp16 = reduce_sum(axes = v2_43_axes_0, keep_dims = v2_43_keep_dims_0, x = var_1266_cast_fp16)[name = tensor("v2_43_cast_fp16")]; - tensor var_1269_cast_fp16 = mul(x = dx2_43_cast_fp16, y = weights_89_cast_fp16)[name = tensor("op_1269_cast_fp16")]; + tensor var_1269_cast_fp16 = mul(x = dx2_43_cast_fp16, y = weights_89_cast_fp16)[name = tensor("op_1269_cast_fp16")]; tensor var_1271_axes_0 = const()[name = tensor("op_1271_axes_0"), val = tensor([2])]; tensor var_1271_keep_dims_0 = const()[name = tensor("op_1271_keep_dims_0"), val = tensor(false)]; tensor var_1271_cast_fp16 = reduce_sum(axes = var_1271_axes_0, keep_dims = var_1271_keep_dims_0, x = var_1269_cast_fp16)[name = tensor("op_1271_cast_fp16")]; @@ -1168,31 +1168,31 @@ program(1.0) tensor var_1279_interleave_0 = const()[name = tensor("op_1279_interleave_0"), val = tensor(false)]; tensor var_1279_cast_fp16 = concat(axis = var_72, interleave = var_1279_interleave_0, values = (mean_43_cast_fp16, std_43_cast_fp16))[name = tensor("op_1279_cast_fp16")]; tensor var_1281_begin_0 = const()[name = tensor("op_1281_begin_0"), val = tensor([0, 22, 0])]; - tensor var_1281_end_0 = const()[name = tensor("op_1281_end_0"), val = tensor([1, 23, 125])]; + tensor var_1281_end_0 = const()[name = tensor("op_1281_end_0"), val = tensor([1, 23, 375])]; tensor var_1281_end_mask_0 = const()[name = tensor("op_1281_end_mask_0"), val = tensor([true, false, true])]; tensor var_1281_squeeze_mask_0 = const()[name = tensor("op_1281_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1281_cast_fp16 = slice_by_index(begin = var_1281_begin_0, end = var_1281_end_0, end_mask = var_1281_end_mask_0, squeeze_mask = var_1281_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1281_cast_fp16")]; + tensor var_1281_cast_fp16 = slice_by_index(begin = var_1281_begin_0, end = var_1281_end_0, end_mask = var_1281_end_mask_0, squeeze_mask = var_1281_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1281_cast_fp16")]; tensor weights_93_axes_0 = const()[name = tensor("weights_93_axes_0"), val = tensor([1])]; - tensor weights_93_cast_fp16 = expand_dims(axes = weights_93_axes_0, x = var_1281_cast_fp16)[name = tensor("weights_93_cast_fp16")]; + tensor weights_93_cast_fp16 = expand_dims(axes = weights_93_axes_0, x = var_1281_cast_fp16)[name = tensor("weights_93_cast_fp16")]; tensor var_1285_axes_0 = const()[name = tensor("op_1285_axes_0"), val = tensor([2])]; tensor var_1285_keep_dims_0 = const()[name = tensor("op_1285_keep_dims_0"), val = tensor(false)]; tensor var_1285_cast_fp16 = reduce_sum(axes = var_1285_axes_0, keep_dims = var_1285_keep_dims_0, x = weights_93_cast_fp16)[name = tensor("op_1285_cast_fp16")]; tensor var_1286_to_fp16 = const()[name = tensor("op_1286_to_fp16"), val = tensor(0x1p-24)]; tensor v1_45_cast_fp16 = add(x = var_1285_cast_fp16, y = var_1286_to_fp16)[name = tensor("v1_45_cast_fp16")]; - tensor var_1288_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_93_cast_fp16)[name = tensor("op_1288_cast_fp16")]; + tensor var_1288_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_93_cast_fp16)[name = tensor("op_1288_cast_fp16")]; tensor var_1290_axes_0 = const()[name = tensor("op_1290_axes_0"), val = tensor([2])]; tensor var_1290_keep_dims_0 = const()[name = tensor("op_1290_keep_dims_0"), val = tensor(false)]; tensor var_1290_cast_fp16 = reduce_sum(axes = var_1290_axes_0, keep_dims = var_1290_keep_dims_0, x = var_1288_cast_fp16)[name = tensor("op_1290_cast_fp16")]; tensor mean_45_cast_fp16 = real_div(x = var_1290_cast_fp16, y = v1_45_cast_fp16)[name = tensor("mean_45_cast_fp16")]; tensor var_1292_axes_0 = const()[name = tensor("op_1292_axes_0"), val = tensor([2])]; tensor var_1292_cast_fp16 = expand_dims(axes = var_1292_axes_0, x = mean_45_cast_fp16)[name = tensor("op_1292_cast_fp16")]; - tensor var_1293_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1292_cast_fp16)[name = tensor("op_1293_cast_fp16")]; - tensor dx2_45_cast_fp16 = mul(x = var_1293_cast_fp16, y = var_1293_cast_fp16)[name = tensor("dx2_45_cast_fp16")]; - tensor var_1295_cast_fp16 = mul(x = weights_93_cast_fp16, y = weights_93_cast_fp16)[name = tensor("op_1295_cast_fp16")]; + tensor var_1293_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1292_cast_fp16)[name = tensor("op_1293_cast_fp16")]; + tensor dx2_45_cast_fp16 = mul(x = var_1293_cast_fp16, y = var_1293_cast_fp16)[name = tensor("dx2_45_cast_fp16")]; + tensor var_1295_cast_fp16 = mul(x = weights_93_cast_fp16, y = weights_93_cast_fp16)[name = tensor("op_1295_cast_fp16")]; tensor v2_45_axes_0 = const()[name = tensor("v2_45_axes_0"), val = tensor([2])]; tensor v2_45_keep_dims_0 = const()[name = tensor("v2_45_keep_dims_0"), val = tensor(false)]; tensor v2_45_cast_fp16 = reduce_sum(axes = v2_45_axes_0, keep_dims = v2_45_keep_dims_0, x = var_1295_cast_fp16)[name = tensor("v2_45_cast_fp16")]; - tensor var_1298_cast_fp16 = mul(x = dx2_45_cast_fp16, y = weights_93_cast_fp16)[name = tensor("op_1298_cast_fp16")]; + tensor var_1298_cast_fp16 = mul(x = dx2_45_cast_fp16, y = weights_93_cast_fp16)[name = tensor("op_1298_cast_fp16")]; tensor var_1300_axes_0 = const()[name = tensor("op_1300_axes_0"), val = tensor([2])]; tensor var_1300_keep_dims_0 = const()[name = tensor("op_1300_keep_dims_0"), val = tensor(false)]; tensor var_1300_cast_fp16 = reduce_sum(axes = var_1300_axes_0, keep_dims = var_1300_keep_dims_0, x = var_1298_cast_fp16)[name = tensor("op_1300_cast_fp16")]; @@ -1205,31 +1205,31 @@ program(1.0) tensor var_1308_interleave_0 = const()[name = tensor("op_1308_interleave_0"), val = tensor(false)]; tensor var_1308_cast_fp16 = concat(axis = var_72, interleave = var_1308_interleave_0, values = (mean_45_cast_fp16, std_45_cast_fp16))[name = tensor("op_1308_cast_fp16")]; tensor var_1310_begin_0 = const()[name = tensor("op_1310_begin_0"), val = tensor([0, 23, 0])]; - tensor var_1310_end_0 = const()[name = tensor("op_1310_end_0"), val = tensor([1, 24, 125])]; + tensor var_1310_end_0 = const()[name = tensor("op_1310_end_0"), val = tensor([1, 24, 375])]; tensor var_1310_end_mask_0 = const()[name = tensor("op_1310_end_mask_0"), val = tensor([true, false, true])]; tensor var_1310_squeeze_mask_0 = const()[name = tensor("op_1310_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1310_cast_fp16 = slice_by_index(begin = var_1310_begin_0, end = var_1310_end_0, end_mask = var_1310_end_mask_0, squeeze_mask = var_1310_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1310_cast_fp16")]; + tensor var_1310_cast_fp16 = slice_by_index(begin = var_1310_begin_0, end = var_1310_end_0, end_mask = var_1310_end_mask_0, squeeze_mask = var_1310_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1310_cast_fp16")]; tensor weights_97_axes_0 = const()[name = tensor("weights_97_axes_0"), val = tensor([1])]; - tensor weights_97_cast_fp16 = expand_dims(axes = weights_97_axes_0, x = var_1310_cast_fp16)[name = tensor("weights_97_cast_fp16")]; + tensor weights_97_cast_fp16 = expand_dims(axes = weights_97_axes_0, x = var_1310_cast_fp16)[name = tensor("weights_97_cast_fp16")]; tensor var_1314_axes_0 = const()[name = tensor("op_1314_axes_0"), val = tensor([2])]; tensor var_1314_keep_dims_0 = const()[name = tensor("op_1314_keep_dims_0"), val = tensor(false)]; tensor var_1314_cast_fp16 = reduce_sum(axes = var_1314_axes_0, keep_dims = var_1314_keep_dims_0, x = weights_97_cast_fp16)[name = tensor("op_1314_cast_fp16")]; tensor var_1315_to_fp16 = const()[name = tensor("op_1315_to_fp16"), val = tensor(0x1p-24)]; tensor v1_47_cast_fp16 = add(x = var_1314_cast_fp16, y = var_1315_to_fp16)[name = tensor("v1_47_cast_fp16")]; - tensor var_1317_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_97_cast_fp16)[name = tensor("op_1317_cast_fp16")]; + tensor var_1317_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_97_cast_fp16)[name = tensor("op_1317_cast_fp16")]; tensor var_1319_axes_0 = const()[name = tensor("op_1319_axes_0"), val = tensor([2])]; tensor var_1319_keep_dims_0 = const()[name = tensor("op_1319_keep_dims_0"), val = tensor(false)]; tensor var_1319_cast_fp16 = reduce_sum(axes = var_1319_axes_0, keep_dims = var_1319_keep_dims_0, x = var_1317_cast_fp16)[name = tensor("op_1319_cast_fp16")]; tensor mean_47_cast_fp16 = real_div(x = var_1319_cast_fp16, y = v1_47_cast_fp16)[name = tensor("mean_47_cast_fp16")]; tensor var_1321_axes_0 = const()[name = tensor("op_1321_axes_0"), val = tensor([2])]; tensor var_1321_cast_fp16 = expand_dims(axes = var_1321_axes_0, x = mean_47_cast_fp16)[name = tensor("op_1321_cast_fp16")]; - tensor var_1322_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1321_cast_fp16)[name = tensor("op_1322_cast_fp16")]; - tensor dx2_47_cast_fp16 = mul(x = var_1322_cast_fp16, y = var_1322_cast_fp16)[name = tensor("dx2_47_cast_fp16")]; - tensor var_1324_cast_fp16 = mul(x = weights_97_cast_fp16, y = weights_97_cast_fp16)[name = tensor("op_1324_cast_fp16")]; + tensor var_1322_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1321_cast_fp16)[name = tensor("op_1322_cast_fp16")]; + tensor dx2_47_cast_fp16 = mul(x = var_1322_cast_fp16, y = var_1322_cast_fp16)[name = tensor("dx2_47_cast_fp16")]; + tensor var_1324_cast_fp16 = mul(x = weights_97_cast_fp16, y = weights_97_cast_fp16)[name = tensor("op_1324_cast_fp16")]; tensor v2_47_axes_0 = const()[name = tensor("v2_47_axes_0"), val = tensor([2])]; tensor v2_47_keep_dims_0 = const()[name = tensor("v2_47_keep_dims_0"), val = tensor(false)]; tensor v2_47_cast_fp16 = reduce_sum(axes = v2_47_axes_0, keep_dims = v2_47_keep_dims_0, x = var_1324_cast_fp16)[name = tensor("v2_47_cast_fp16")]; - tensor var_1327_cast_fp16 = mul(x = dx2_47_cast_fp16, y = weights_97_cast_fp16)[name = tensor("op_1327_cast_fp16")]; + tensor var_1327_cast_fp16 = mul(x = dx2_47_cast_fp16, y = weights_97_cast_fp16)[name = tensor("op_1327_cast_fp16")]; tensor var_1329_axes_0 = const()[name = tensor("op_1329_axes_0"), val = tensor([2])]; tensor var_1329_keep_dims_0 = const()[name = tensor("op_1329_keep_dims_0"), val = tensor(false)]; tensor var_1329_cast_fp16 = reduce_sum(axes = var_1329_axes_0, keep_dims = var_1329_keep_dims_0, x = var_1327_cast_fp16)[name = tensor("op_1329_cast_fp16")]; @@ -1242,31 +1242,31 @@ program(1.0) tensor var_1337_interleave_0 = const()[name = tensor("op_1337_interleave_0"), val = tensor(false)]; tensor var_1337_cast_fp16 = concat(axis = var_72, interleave = var_1337_interleave_0, values = (mean_47_cast_fp16, std_47_cast_fp16))[name = tensor("op_1337_cast_fp16")]; tensor var_1339_begin_0 = const()[name = tensor("op_1339_begin_0"), val = tensor([0, 24, 0])]; - tensor var_1339_end_0 = const()[name = tensor("op_1339_end_0"), val = tensor([1, 25, 125])]; + tensor var_1339_end_0 = const()[name = tensor("op_1339_end_0"), val = tensor([1, 25, 375])]; tensor var_1339_end_mask_0 = const()[name = tensor("op_1339_end_mask_0"), val = tensor([true, false, true])]; tensor var_1339_squeeze_mask_0 = const()[name = tensor("op_1339_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1339_cast_fp16 = slice_by_index(begin = var_1339_begin_0, end = var_1339_end_0, end_mask = var_1339_end_mask_0, squeeze_mask = var_1339_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1339_cast_fp16")]; + tensor var_1339_cast_fp16 = slice_by_index(begin = var_1339_begin_0, end = var_1339_end_0, end_mask = var_1339_end_mask_0, squeeze_mask = var_1339_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1339_cast_fp16")]; tensor weights_101_axes_0 = const()[name = tensor("weights_101_axes_0"), val = tensor([1])]; - tensor weights_101_cast_fp16 = expand_dims(axes = weights_101_axes_0, x = var_1339_cast_fp16)[name = tensor("weights_101_cast_fp16")]; + tensor weights_101_cast_fp16 = expand_dims(axes = weights_101_axes_0, x = var_1339_cast_fp16)[name = tensor("weights_101_cast_fp16")]; tensor var_1343_axes_0 = const()[name = tensor("op_1343_axes_0"), val = tensor([2])]; tensor var_1343_keep_dims_0 = const()[name = tensor("op_1343_keep_dims_0"), val = tensor(false)]; tensor var_1343_cast_fp16 = reduce_sum(axes = var_1343_axes_0, keep_dims = var_1343_keep_dims_0, x = weights_101_cast_fp16)[name = tensor("op_1343_cast_fp16")]; tensor var_1344_to_fp16 = const()[name = tensor("op_1344_to_fp16"), val = tensor(0x1p-24)]; tensor v1_49_cast_fp16 = add(x = var_1343_cast_fp16, y = var_1344_to_fp16)[name = tensor("v1_49_cast_fp16")]; - tensor var_1346_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_101_cast_fp16)[name = tensor("op_1346_cast_fp16")]; + tensor var_1346_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_101_cast_fp16)[name = tensor("op_1346_cast_fp16")]; tensor var_1348_axes_0 = const()[name = tensor("op_1348_axes_0"), val = tensor([2])]; tensor var_1348_keep_dims_0 = const()[name = tensor("op_1348_keep_dims_0"), val = tensor(false)]; tensor var_1348_cast_fp16 = reduce_sum(axes = var_1348_axes_0, keep_dims = var_1348_keep_dims_0, x = var_1346_cast_fp16)[name = tensor("op_1348_cast_fp16")]; tensor mean_49_cast_fp16 = real_div(x = var_1348_cast_fp16, y = v1_49_cast_fp16)[name = tensor("mean_49_cast_fp16")]; tensor var_1350_axes_0 = const()[name = tensor("op_1350_axes_0"), val = tensor([2])]; tensor var_1350_cast_fp16 = expand_dims(axes = var_1350_axes_0, x = mean_49_cast_fp16)[name = tensor("op_1350_cast_fp16")]; - tensor var_1351_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1350_cast_fp16)[name = tensor("op_1351_cast_fp16")]; - tensor dx2_49_cast_fp16 = mul(x = var_1351_cast_fp16, y = var_1351_cast_fp16)[name = tensor("dx2_49_cast_fp16")]; - tensor var_1353_cast_fp16 = mul(x = weights_101_cast_fp16, y = weights_101_cast_fp16)[name = tensor("op_1353_cast_fp16")]; + tensor var_1351_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1350_cast_fp16)[name = tensor("op_1351_cast_fp16")]; + tensor dx2_49_cast_fp16 = mul(x = var_1351_cast_fp16, y = var_1351_cast_fp16)[name = tensor("dx2_49_cast_fp16")]; + tensor var_1353_cast_fp16 = mul(x = weights_101_cast_fp16, y = weights_101_cast_fp16)[name = tensor("op_1353_cast_fp16")]; tensor v2_49_axes_0 = const()[name = tensor("v2_49_axes_0"), val = tensor([2])]; tensor v2_49_keep_dims_0 = const()[name = tensor("v2_49_keep_dims_0"), val = tensor(false)]; tensor v2_49_cast_fp16 = reduce_sum(axes = v2_49_axes_0, keep_dims = v2_49_keep_dims_0, x = var_1353_cast_fp16)[name = tensor("v2_49_cast_fp16")]; - tensor var_1356_cast_fp16 = mul(x = dx2_49_cast_fp16, y = weights_101_cast_fp16)[name = tensor("op_1356_cast_fp16")]; + tensor var_1356_cast_fp16 = mul(x = dx2_49_cast_fp16, y = weights_101_cast_fp16)[name = tensor("op_1356_cast_fp16")]; tensor var_1358_axes_0 = const()[name = tensor("op_1358_axes_0"), val = tensor([2])]; tensor var_1358_keep_dims_0 = const()[name = tensor("op_1358_keep_dims_0"), val = tensor(false)]; tensor var_1358_cast_fp16 = reduce_sum(axes = var_1358_axes_0, keep_dims = var_1358_keep_dims_0, x = var_1356_cast_fp16)[name = tensor("op_1358_cast_fp16")]; @@ -1279,31 +1279,31 @@ program(1.0) tensor var_1366_interleave_0 = const()[name = tensor("op_1366_interleave_0"), val = tensor(false)]; tensor var_1366_cast_fp16 = concat(axis = var_72, interleave = var_1366_interleave_0, values = (mean_49_cast_fp16, std_49_cast_fp16))[name = tensor("op_1366_cast_fp16")]; tensor var_1368_begin_0 = const()[name = tensor("op_1368_begin_0"), val = tensor([0, 25, 0])]; - tensor var_1368_end_0 = const()[name = tensor("op_1368_end_0"), val = tensor([1, 26, 125])]; + tensor var_1368_end_0 = const()[name = tensor("op_1368_end_0"), val = tensor([1, 26, 375])]; tensor var_1368_end_mask_0 = const()[name = tensor("op_1368_end_mask_0"), val = tensor([true, false, true])]; tensor var_1368_squeeze_mask_0 = const()[name = tensor("op_1368_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1368_cast_fp16 = slice_by_index(begin = var_1368_begin_0, end = var_1368_end_0, end_mask = var_1368_end_mask_0, squeeze_mask = var_1368_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1368_cast_fp16")]; + tensor var_1368_cast_fp16 = slice_by_index(begin = var_1368_begin_0, end = var_1368_end_0, end_mask = var_1368_end_mask_0, squeeze_mask = var_1368_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1368_cast_fp16")]; tensor weights_105_axes_0 = const()[name = tensor("weights_105_axes_0"), val = tensor([1])]; - tensor weights_105_cast_fp16 = expand_dims(axes = weights_105_axes_0, x = var_1368_cast_fp16)[name = tensor("weights_105_cast_fp16")]; + tensor weights_105_cast_fp16 = expand_dims(axes = weights_105_axes_0, x = var_1368_cast_fp16)[name = tensor("weights_105_cast_fp16")]; tensor var_1372_axes_0 = const()[name = tensor("op_1372_axes_0"), val = tensor([2])]; tensor var_1372_keep_dims_0 = const()[name = tensor("op_1372_keep_dims_0"), val = tensor(false)]; tensor var_1372_cast_fp16 = reduce_sum(axes = var_1372_axes_0, keep_dims = var_1372_keep_dims_0, x = weights_105_cast_fp16)[name = tensor("op_1372_cast_fp16")]; tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(0x1p-24)]; tensor v1_51_cast_fp16 = add(x = var_1372_cast_fp16, y = var_1373_to_fp16)[name = tensor("v1_51_cast_fp16")]; - tensor var_1375_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_105_cast_fp16)[name = tensor("op_1375_cast_fp16")]; + tensor var_1375_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_105_cast_fp16)[name = tensor("op_1375_cast_fp16")]; tensor var_1377_axes_0 = const()[name = tensor("op_1377_axes_0"), val = tensor([2])]; tensor var_1377_keep_dims_0 = const()[name = tensor("op_1377_keep_dims_0"), val = tensor(false)]; tensor var_1377_cast_fp16 = reduce_sum(axes = var_1377_axes_0, keep_dims = var_1377_keep_dims_0, x = var_1375_cast_fp16)[name = tensor("op_1377_cast_fp16")]; tensor mean_51_cast_fp16 = real_div(x = var_1377_cast_fp16, y = v1_51_cast_fp16)[name = tensor("mean_51_cast_fp16")]; tensor var_1379_axes_0 = const()[name = tensor("op_1379_axes_0"), val = tensor([2])]; tensor var_1379_cast_fp16 = expand_dims(axes = var_1379_axes_0, x = mean_51_cast_fp16)[name = tensor("op_1379_cast_fp16")]; - tensor var_1380_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1379_cast_fp16)[name = tensor("op_1380_cast_fp16")]; - tensor dx2_51_cast_fp16 = mul(x = var_1380_cast_fp16, y = var_1380_cast_fp16)[name = tensor("dx2_51_cast_fp16")]; - tensor var_1382_cast_fp16 = mul(x = weights_105_cast_fp16, y = weights_105_cast_fp16)[name = tensor("op_1382_cast_fp16")]; + tensor var_1380_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1379_cast_fp16)[name = tensor("op_1380_cast_fp16")]; + tensor dx2_51_cast_fp16 = mul(x = var_1380_cast_fp16, y = var_1380_cast_fp16)[name = tensor("dx2_51_cast_fp16")]; + tensor var_1382_cast_fp16 = mul(x = weights_105_cast_fp16, y = weights_105_cast_fp16)[name = tensor("op_1382_cast_fp16")]; tensor v2_51_axes_0 = const()[name = tensor("v2_51_axes_0"), val = tensor([2])]; tensor v2_51_keep_dims_0 = const()[name = tensor("v2_51_keep_dims_0"), val = tensor(false)]; tensor v2_51_cast_fp16 = reduce_sum(axes = v2_51_axes_0, keep_dims = v2_51_keep_dims_0, x = var_1382_cast_fp16)[name = tensor("v2_51_cast_fp16")]; - tensor var_1385_cast_fp16 = mul(x = dx2_51_cast_fp16, y = weights_105_cast_fp16)[name = tensor("op_1385_cast_fp16")]; + tensor var_1385_cast_fp16 = mul(x = dx2_51_cast_fp16, y = weights_105_cast_fp16)[name = tensor("op_1385_cast_fp16")]; tensor var_1387_axes_0 = const()[name = tensor("op_1387_axes_0"), val = tensor([2])]; tensor var_1387_keep_dims_0 = const()[name = tensor("op_1387_keep_dims_0"), val = tensor(false)]; tensor var_1387_cast_fp16 = reduce_sum(axes = var_1387_axes_0, keep_dims = var_1387_keep_dims_0, x = var_1385_cast_fp16)[name = tensor("op_1387_cast_fp16")]; @@ -1316,31 +1316,31 @@ program(1.0) tensor var_1395_interleave_0 = const()[name = tensor("op_1395_interleave_0"), val = tensor(false)]; tensor var_1395_cast_fp16 = concat(axis = var_72, interleave = var_1395_interleave_0, values = (mean_51_cast_fp16, std_51_cast_fp16))[name = tensor("op_1395_cast_fp16")]; tensor var_1397_begin_0 = const()[name = tensor("op_1397_begin_0"), val = tensor([0, 26, 0])]; - tensor var_1397_end_0 = const()[name = tensor("op_1397_end_0"), val = tensor([1, 27, 125])]; + tensor var_1397_end_0 = const()[name = tensor("op_1397_end_0"), val = tensor([1, 27, 375])]; tensor var_1397_end_mask_0 = const()[name = tensor("op_1397_end_mask_0"), val = tensor([true, false, true])]; tensor var_1397_squeeze_mask_0 = const()[name = tensor("op_1397_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1397_cast_fp16 = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, squeeze_mask = var_1397_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1397_cast_fp16")]; + tensor var_1397_cast_fp16 = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, squeeze_mask = var_1397_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1397_cast_fp16")]; tensor weights_109_axes_0 = const()[name = tensor("weights_109_axes_0"), val = tensor([1])]; - tensor weights_109_cast_fp16 = expand_dims(axes = weights_109_axes_0, x = var_1397_cast_fp16)[name = tensor("weights_109_cast_fp16")]; + tensor weights_109_cast_fp16 = expand_dims(axes = weights_109_axes_0, x = var_1397_cast_fp16)[name = tensor("weights_109_cast_fp16")]; tensor var_1401_axes_0 = const()[name = tensor("op_1401_axes_0"), val = tensor([2])]; tensor var_1401_keep_dims_0 = const()[name = tensor("op_1401_keep_dims_0"), val = tensor(false)]; tensor var_1401_cast_fp16 = reduce_sum(axes = var_1401_axes_0, keep_dims = var_1401_keep_dims_0, x = weights_109_cast_fp16)[name = tensor("op_1401_cast_fp16")]; tensor var_1402_to_fp16 = const()[name = tensor("op_1402_to_fp16"), val = tensor(0x1p-24)]; tensor v1_53_cast_fp16 = add(x = var_1401_cast_fp16, y = var_1402_to_fp16)[name = tensor("v1_53_cast_fp16")]; - tensor var_1404_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_109_cast_fp16)[name = tensor("op_1404_cast_fp16")]; + tensor var_1404_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_109_cast_fp16)[name = tensor("op_1404_cast_fp16")]; tensor var_1406_axes_0 = const()[name = tensor("op_1406_axes_0"), val = tensor([2])]; tensor var_1406_keep_dims_0 = const()[name = tensor("op_1406_keep_dims_0"), val = tensor(false)]; tensor var_1406_cast_fp16 = reduce_sum(axes = var_1406_axes_0, keep_dims = var_1406_keep_dims_0, x = var_1404_cast_fp16)[name = tensor("op_1406_cast_fp16")]; tensor mean_53_cast_fp16 = real_div(x = var_1406_cast_fp16, y = v1_53_cast_fp16)[name = tensor("mean_53_cast_fp16")]; tensor var_1408_axes_0 = const()[name = tensor("op_1408_axes_0"), val = tensor([2])]; tensor var_1408_cast_fp16 = expand_dims(axes = var_1408_axes_0, x = mean_53_cast_fp16)[name = tensor("op_1408_cast_fp16")]; - tensor var_1409_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1408_cast_fp16)[name = tensor("op_1409_cast_fp16")]; - tensor dx2_53_cast_fp16 = mul(x = var_1409_cast_fp16, y = var_1409_cast_fp16)[name = tensor("dx2_53_cast_fp16")]; - tensor var_1411_cast_fp16 = mul(x = weights_109_cast_fp16, y = weights_109_cast_fp16)[name = tensor("op_1411_cast_fp16")]; + tensor var_1409_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1408_cast_fp16)[name = tensor("op_1409_cast_fp16")]; + tensor dx2_53_cast_fp16 = mul(x = var_1409_cast_fp16, y = var_1409_cast_fp16)[name = tensor("dx2_53_cast_fp16")]; + tensor var_1411_cast_fp16 = mul(x = weights_109_cast_fp16, y = weights_109_cast_fp16)[name = tensor("op_1411_cast_fp16")]; tensor v2_53_axes_0 = const()[name = tensor("v2_53_axes_0"), val = tensor([2])]; tensor v2_53_keep_dims_0 = const()[name = tensor("v2_53_keep_dims_0"), val = tensor(false)]; tensor v2_53_cast_fp16 = reduce_sum(axes = v2_53_axes_0, keep_dims = v2_53_keep_dims_0, x = var_1411_cast_fp16)[name = tensor("v2_53_cast_fp16")]; - tensor var_1414_cast_fp16 = mul(x = dx2_53_cast_fp16, y = weights_109_cast_fp16)[name = tensor("op_1414_cast_fp16")]; + tensor var_1414_cast_fp16 = mul(x = dx2_53_cast_fp16, y = weights_109_cast_fp16)[name = tensor("op_1414_cast_fp16")]; tensor var_1416_axes_0 = const()[name = tensor("op_1416_axes_0"), val = tensor([2])]; tensor var_1416_keep_dims_0 = const()[name = tensor("op_1416_keep_dims_0"), val = tensor(false)]; tensor var_1416_cast_fp16 = reduce_sum(axes = var_1416_axes_0, keep_dims = var_1416_keep_dims_0, x = var_1414_cast_fp16)[name = tensor("op_1416_cast_fp16")]; @@ -1353,31 +1353,31 @@ program(1.0) tensor var_1424_interleave_0 = const()[name = tensor("op_1424_interleave_0"), val = tensor(false)]; tensor var_1424_cast_fp16 = concat(axis = var_72, interleave = var_1424_interleave_0, values = (mean_53_cast_fp16, std_53_cast_fp16))[name = tensor("op_1424_cast_fp16")]; tensor var_1426_begin_0 = const()[name = tensor("op_1426_begin_0"), val = tensor([0, 27, 0])]; - tensor var_1426_end_0 = const()[name = tensor("op_1426_end_0"), val = tensor([1, 28, 125])]; + tensor var_1426_end_0 = const()[name = tensor("op_1426_end_0"), val = tensor([1, 28, 375])]; tensor var_1426_end_mask_0 = const()[name = tensor("op_1426_end_mask_0"), val = tensor([true, false, true])]; tensor var_1426_squeeze_mask_0 = const()[name = tensor("op_1426_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1426_cast_fp16 = slice_by_index(begin = var_1426_begin_0, end = var_1426_end_0, end_mask = var_1426_end_mask_0, squeeze_mask = var_1426_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1426_cast_fp16")]; + tensor var_1426_cast_fp16 = slice_by_index(begin = var_1426_begin_0, end = var_1426_end_0, end_mask = var_1426_end_mask_0, squeeze_mask = var_1426_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1426_cast_fp16")]; tensor weights_113_axes_0 = const()[name = tensor("weights_113_axes_0"), val = tensor([1])]; - tensor weights_113_cast_fp16 = expand_dims(axes = weights_113_axes_0, x = var_1426_cast_fp16)[name = tensor("weights_113_cast_fp16")]; + tensor weights_113_cast_fp16 = expand_dims(axes = weights_113_axes_0, x = var_1426_cast_fp16)[name = tensor("weights_113_cast_fp16")]; tensor var_1430_axes_0 = const()[name = tensor("op_1430_axes_0"), val = tensor([2])]; tensor var_1430_keep_dims_0 = const()[name = tensor("op_1430_keep_dims_0"), val = tensor(false)]; tensor var_1430_cast_fp16 = reduce_sum(axes = var_1430_axes_0, keep_dims = var_1430_keep_dims_0, x = weights_113_cast_fp16)[name = tensor("op_1430_cast_fp16")]; tensor var_1431_to_fp16 = const()[name = tensor("op_1431_to_fp16"), val = tensor(0x1p-24)]; tensor v1_55_cast_fp16 = add(x = var_1430_cast_fp16, y = var_1431_to_fp16)[name = tensor("v1_55_cast_fp16")]; - tensor var_1433_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_113_cast_fp16)[name = tensor("op_1433_cast_fp16")]; + tensor var_1433_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_113_cast_fp16)[name = tensor("op_1433_cast_fp16")]; tensor var_1435_axes_0 = const()[name = tensor("op_1435_axes_0"), val = tensor([2])]; tensor var_1435_keep_dims_0 = const()[name = tensor("op_1435_keep_dims_0"), val = tensor(false)]; tensor var_1435_cast_fp16 = reduce_sum(axes = var_1435_axes_0, keep_dims = var_1435_keep_dims_0, x = var_1433_cast_fp16)[name = tensor("op_1435_cast_fp16")]; tensor mean_55_cast_fp16 = real_div(x = var_1435_cast_fp16, y = v1_55_cast_fp16)[name = tensor("mean_55_cast_fp16")]; tensor var_1437_axes_0 = const()[name = tensor("op_1437_axes_0"), val = tensor([2])]; tensor var_1437_cast_fp16 = expand_dims(axes = var_1437_axes_0, x = mean_55_cast_fp16)[name = tensor("op_1437_cast_fp16")]; - tensor var_1438_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1437_cast_fp16)[name = tensor("op_1438_cast_fp16")]; - tensor dx2_55_cast_fp16 = mul(x = var_1438_cast_fp16, y = var_1438_cast_fp16)[name = tensor("dx2_55_cast_fp16")]; - tensor var_1440_cast_fp16 = mul(x = weights_113_cast_fp16, y = weights_113_cast_fp16)[name = tensor("op_1440_cast_fp16")]; + tensor var_1438_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1437_cast_fp16)[name = tensor("op_1438_cast_fp16")]; + tensor dx2_55_cast_fp16 = mul(x = var_1438_cast_fp16, y = var_1438_cast_fp16)[name = tensor("dx2_55_cast_fp16")]; + tensor var_1440_cast_fp16 = mul(x = weights_113_cast_fp16, y = weights_113_cast_fp16)[name = tensor("op_1440_cast_fp16")]; tensor v2_55_axes_0 = const()[name = tensor("v2_55_axes_0"), val = tensor([2])]; tensor v2_55_keep_dims_0 = const()[name = tensor("v2_55_keep_dims_0"), val = tensor(false)]; tensor v2_55_cast_fp16 = reduce_sum(axes = v2_55_axes_0, keep_dims = v2_55_keep_dims_0, x = var_1440_cast_fp16)[name = tensor("v2_55_cast_fp16")]; - tensor var_1443_cast_fp16 = mul(x = dx2_55_cast_fp16, y = weights_113_cast_fp16)[name = tensor("op_1443_cast_fp16")]; + tensor var_1443_cast_fp16 = mul(x = dx2_55_cast_fp16, y = weights_113_cast_fp16)[name = tensor("op_1443_cast_fp16")]; tensor var_1445_axes_0 = const()[name = tensor("op_1445_axes_0"), val = tensor([2])]; tensor var_1445_keep_dims_0 = const()[name = tensor("op_1445_keep_dims_0"), val = tensor(false)]; tensor var_1445_cast_fp16 = reduce_sum(axes = var_1445_axes_0, keep_dims = var_1445_keep_dims_0, x = var_1443_cast_fp16)[name = tensor("op_1445_cast_fp16")]; @@ -1390,31 +1390,31 @@ program(1.0) tensor var_1453_interleave_0 = const()[name = tensor("op_1453_interleave_0"), val = tensor(false)]; tensor var_1453_cast_fp16 = concat(axis = var_72, interleave = var_1453_interleave_0, values = (mean_55_cast_fp16, std_55_cast_fp16))[name = tensor("op_1453_cast_fp16")]; tensor var_1455_begin_0 = const()[name = tensor("op_1455_begin_0"), val = tensor([0, 28, 0])]; - tensor var_1455_end_0 = const()[name = tensor("op_1455_end_0"), val = tensor([1, 29, 125])]; + tensor var_1455_end_0 = const()[name = tensor("op_1455_end_0"), val = tensor([1, 29, 375])]; tensor var_1455_end_mask_0 = const()[name = tensor("op_1455_end_mask_0"), val = tensor([true, false, true])]; tensor var_1455_squeeze_mask_0 = const()[name = tensor("op_1455_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1455_cast_fp16 = slice_by_index(begin = var_1455_begin_0, end = var_1455_end_0, end_mask = var_1455_end_mask_0, squeeze_mask = var_1455_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1455_cast_fp16")]; + tensor var_1455_cast_fp16 = slice_by_index(begin = var_1455_begin_0, end = var_1455_end_0, end_mask = var_1455_end_mask_0, squeeze_mask = var_1455_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1455_cast_fp16")]; tensor weights_117_axes_0 = const()[name = tensor("weights_117_axes_0"), val = tensor([1])]; - tensor weights_117_cast_fp16 = expand_dims(axes = weights_117_axes_0, x = var_1455_cast_fp16)[name = tensor("weights_117_cast_fp16")]; + tensor weights_117_cast_fp16 = expand_dims(axes = weights_117_axes_0, x = var_1455_cast_fp16)[name = tensor("weights_117_cast_fp16")]; tensor var_1459_axes_0 = const()[name = tensor("op_1459_axes_0"), val = tensor([2])]; tensor var_1459_keep_dims_0 = const()[name = tensor("op_1459_keep_dims_0"), val = tensor(false)]; tensor var_1459_cast_fp16 = reduce_sum(axes = var_1459_axes_0, keep_dims = var_1459_keep_dims_0, x = weights_117_cast_fp16)[name = tensor("op_1459_cast_fp16")]; tensor var_1460_to_fp16 = const()[name = tensor("op_1460_to_fp16"), val = tensor(0x1p-24)]; tensor v1_57_cast_fp16 = add(x = var_1459_cast_fp16, y = var_1460_to_fp16)[name = tensor("v1_57_cast_fp16")]; - tensor var_1462_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_117_cast_fp16)[name = tensor("op_1462_cast_fp16")]; + tensor var_1462_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_117_cast_fp16)[name = tensor("op_1462_cast_fp16")]; tensor var_1464_axes_0 = const()[name = tensor("op_1464_axes_0"), val = tensor([2])]; tensor var_1464_keep_dims_0 = const()[name = tensor("op_1464_keep_dims_0"), val = tensor(false)]; tensor var_1464_cast_fp16 = reduce_sum(axes = var_1464_axes_0, keep_dims = var_1464_keep_dims_0, x = var_1462_cast_fp16)[name = tensor("op_1464_cast_fp16")]; tensor mean_57_cast_fp16 = real_div(x = var_1464_cast_fp16, y = v1_57_cast_fp16)[name = tensor("mean_57_cast_fp16")]; tensor var_1466_axes_0 = const()[name = tensor("op_1466_axes_0"), val = tensor([2])]; tensor var_1466_cast_fp16 = expand_dims(axes = var_1466_axes_0, x = mean_57_cast_fp16)[name = tensor("op_1466_cast_fp16")]; - tensor var_1467_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1466_cast_fp16)[name = tensor("op_1467_cast_fp16")]; - tensor dx2_57_cast_fp16 = mul(x = var_1467_cast_fp16, y = var_1467_cast_fp16)[name = tensor("dx2_57_cast_fp16")]; - tensor var_1469_cast_fp16 = mul(x = weights_117_cast_fp16, y = weights_117_cast_fp16)[name = tensor("op_1469_cast_fp16")]; + tensor var_1467_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1466_cast_fp16)[name = tensor("op_1467_cast_fp16")]; + tensor dx2_57_cast_fp16 = mul(x = var_1467_cast_fp16, y = var_1467_cast_fp16)[name = tensor("dx2_57_cast_fp16")]; + tensor var_1469_cast_fp16 = mul(x = weights_117_cast_fp16, y = weights_117_cast_fp16)[name = tensor("op_1469_cast_fp16")]; tensor v2_57_axes_0 = const()[name = tensor("v2_57_axes_0"), val = tensor([2])]; tensor v2_57_keep_dims_0 = const()[name = tensor("v2_57_keep_dims_0"), val = tensor(false)]; tensor v2_57_cast_fp16 = reduce_sum(axes = v2_57_axes_0, keep_dims = v2_57_keep_dims_0, x = var_1469_cast_fp16)[name = tensor("v2_57_cast_fp16")]; - tensor var_1472_cast_fp16 = mul(x = dx2_57_cast_fp16, y = weights_117_cast_fp16)[name = tensor("op_1472_cast_fp16")]; + tensor var_1472_cast_fp16 = mul(x = dx2_57_cast_fp16, y = weights_117_cast_fp16)[name = tensor("op_1472_cast_fp16")]; tensor var_1474_axes_0 = const()[name = tensor("op_1474_axes_0"), val = tensor([2])]; tensor var_1474_keep_dims_0 = const()[name = tensor("op_1474_keep_dims_0"), val = tensor(false)]; tensor var_1474_cast_fp16 = reduce_sum(axes = var_1474_axes_0, keep_dims = var_1474_keep_dims_0, x = var_1472_cast_fp16)[name = tensor("op_1474_cast_fp16")]; @@ -1427,31 +1427,31 @@ program(1.0) tensor var_1482_interleave_0 = const()[name = tensor("op_1482_interleave_0"), val = tensor(false)]; tensor var_1482_cast_fp16 = concat(axis = var_72, interleave = var_1482_interleave_0, values = (mean_57_cast_fp16, std_57_cast_fp16))[name = tensor("op_1482_cast_fp16")]; tensor var_1484_begin_0 = const()[name = tensor("op_1484_begin_0"), val = tensor([0, 29, 0])]; - tensor var_1484_end_0 = const()[name = tensor("op_1484_end_0"), val = tensor([1, 30, 125])]; + tensor var_1484_end_0 = const()[name = tensor("op_1484_end_0"), val = tensor([1, 30, 375])]; tensor var_1484_end_mask_0 = const()[name = tensor("op_1484_end_mask_0"), val = tensor([true, false, true])]; tensor var_1484_squeeze_mask_0 = const()[name = tensor("op_1484_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1484_cast_fp16 = slice_by_index(begin = var_1484_begin_0, end = var_1484_end_0, end_mask = var_1484_end_mask_0, squeeze_mask = var_1484_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1484_cast_fp16")]; + tensor var_1484_cast_fp16 = slice_by_index(begin = var_1484_begin_0, end = var_1484_end_0, end_mask = var_1484_end_mask_0, squeeze_mask = var_1484_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1484_cast_fp16")]; tensor weights_121_axes_0 = const()[name = tensor("weights_121_axes_0"), val = tensor([1])]; - tensor weights_121_cast_fp16 = expand_dims(axes = weights_121_axes_0, x = var_1484_cast_fp16)[name = tensor("weights_121_cast_fp16")]; + tensor weights_121_cast_fp16 = expand_dims(axes = weights_121_axes_0, x = var_1484_cast_fp16)[name = tensor("weights_121_cast_fp16")]; tensor var_1488_axes_0 = const()[name = tensor("op_1488_axes_0"), val = tensor([2])]; tensor var_1488_keep_dims_0 = const()[name = tensor("op_1488_keep_dims_0"), val = tensor(false)]; tensor var_1488_cast_fp16 = reduce_sum(axes = var_1488_axes_0, keep_dims = var_1488_keep_dims_0, x = weights_121_cast_fp16)[name = tensor("op_1488_cast_fp16")]; tensor var_1489_to_fp16 = const()[name = tensor("op_1489_to_fp16"), val = tensor(0x1p-24)]; tensor v1_59_cast_fp16 = add(x = var_1488_cast_fp16, y = var_1489_to_fp16)[name = tensor("v1_59_cast_fp16")]; - tensor var_1491_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_121_cast_fp16)[name = tensor("op_1491_cast_fp16")]; + tensor var_1491_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_121_cast_fp16)[name = tensor("op_1491_cast_fp16")]; tensor var_1493_axes_0 = const()[name = tensor("op_1493_axes_0"), val = tensor([2])]; tensor var_1493_keep_dims_0 = const()[name = tensor("op_1493_keep_dims_0"), val = tensor(false)]; tensor var_1493_cast_fp16 = reduce_sum(axes = var_1493_axes_0, keep_dims = var_1493_keep_dims_0, x = var_1491_cast_fp16)[name = tensor("op_1493_cast_fp16")]; tensor mean_59_cast_fp16 = real_div(x = var_1493_cast_fp16, y = v1_59_cast_fp16)[name = tensor("mean_59_cast_fp16")]; tensor var_1495_axes_0 = const()[name = tensor("op_1495_axes_0"), val = tensor([2])]; tensor var_1495_cast_fp16 = expand_dims(axes = var_1495_axes_0, x = mean_59_cast_fp16)[name = tensor("op_1495_cast_fp16")]; - tensor var_1496_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1495_cast_fp16)[name = tensor("op_1496_cast_fp16")]; - tensor dx2_59_cast_fp16 = mul(x = var_1496_cast_fp16, y = var_1496_cast_fp16)[name = tensor("dx2_59_cast_fp16")]; - tensor var_1498_cast_fp16 = mul(x = weights_121_cast_fp16, y = weights_121_cast_fp16)[name = tensor("op_1498_cast_fp16")]; + tensor var_1496_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1495_cast_fp16)[name = tensor("op_1496_cast_fp16")]; + tensor dx2_59_cast_fp16 = mul(x = var_1496_cast_fp16, y = var_1496_cast_fp16)[name = tensor("dx2_59_cast_fp16")]; + tensor var_1498_cast_fp16 = mul(x = weights_121_cast_fp16, y = weights_121_cast_fp16)[name = tensor("op_1498_cast_fp16")]; tensor v2_59_axes_0 = const()[name = tensor("v2_59_axes_0"), val = tensor([2])]; tensor v2_59_keep_dims_0 = const()[name = tensor("v2_59_keep_dims_0"), val = tensor(false)]; tensor v2_59_cast_fp16 = reduce_sum(axes = v2_59_axes_0, keep_dims = v2_59_keep_dims_0, x = var_1498_cast_fp16)[name = tensor("v2_59_cast_fp16")]; - tensor var_1501_cast_fp16 = mul(x = dx2_59_cast_fp16, y = weights_121_cast_fp16)[name = tensor("op_1501_cast_fp16")]; + tensor var_1501_cast_fp16 = mul(x = dx2_59_cast_fp16, y = weights_121_cast_fp16)[name = tensor("op_1501_cast_fp16")]; tensor var_1503_axes_0 = const()[name = tensor("op_1503_axes_0"), val = tensor([2])]; tensor var_1503_keep_dims_0 = const()[name = tensor("op_1503_keep_dims_0"), val = tensor(false)]; tensor var_1503_cast_fp16 = reduce_sum(axes = var_1503_axes_0, keep_dims = var_1503_keep_dims_0, x = var_1501_cast_fp16)[name = tensor("op_1503_cast_fp16")]; @@ -1464,31 +1464,31 @@ program(1.0) tensor var_1511_interleave_0 = const()[name = tensor("op_1511_interleave_0"), val = tensor(false)]; tensor var_1511_cast_fp16 = concat(axis = var_72, interleave = var_1511_interleave_0, values = (mean_59_cast_fp16, std_59_cast_fp16))[name = tensor("op_1511_cast_fp16")]; tensor var_1513_begin_0 = const()[name = tensor("op_1513_begin_0"), val = tensor([0, 30, 0])]; - tensor var_1513_end_0 = const()[name = tensor("op_1513_end_0"), val = tensor([1, 31, 125])]; + tensor var_1513_end_0 = const()[name = tensor("op_1513_end_0"), val = tensor([1, 31, 375])]; tensor var_1513_end_mask_0 = const()[name = tensor("op_1513_end_mask_0"), val = tensor([true, false, true])]; tensor var_1513_squeeze_mask_0 = const()[name = tensor("op_1513_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1513_cast_fp16 = slice_by_index(begin = var_1513_begin_0, end = var_1513_end_0, end_mask = var_1513_end_mask_0, squeeze_mask = var_1513_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1513_cast_fp16")]; + tensor var_1513_cast_fp16 = slice_by_index(begin = var_1513_begin_0, end = var_1513_end_0, end_mask = var_1513_end_mask_0, squeeze_mask = var_1513_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1513_cast_fp16")]; tensor weights_125_axes_0 = const()[name = tensor("weights_125_axes_0"), val = tensor([1])]; - tensor weights_125_cast_fp16 = expand_dims(axes = weights_125_axes_0, x = var_1513_cast_fp16)[name = tensor("weights_125_cast_fp16")]; + tensor weights_125_cast_fp16 = expand_dims(axes = weights_125_axes_0, x = var_1513_cast_fp16)[name = tensor("weights_125_cast_fp16")]; tensor var_1517_axes_0 = const()[name = tensor("op_1517_axes_0"), val = tensor([2])]; tensor var_1517_keep_dims_0 = const()[name = tensor("op_1517_keep_dims_0"), val = tensor(false)]; tensor var_1517_cast_fp16 = reduce_sum(axes = var_1517_axes_0, keep_dims = var_1517_keep_dims_0, x = weights_125_cast_fp16)[name = tensor("op_1517_cast_fp16")]; tensor var_1518_to_fp16 = const()[name = tensor("op_1518_to_fp16"), val = tensor(0x1p-24)]; tensor v1_61_cast_fp16 = add(x = var_1517_cast_fp16, y = var_1518_to_fp16)[name = tensor("v1_61_cast_fp16")]; - tensor var_1520_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_125_cast_fp16)[name = tensor("op_1520_cast_fp16")]; + tensor var_1520_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_125_cast_fp16)[name = tensor("op_1520_cast_fp16")]; tensor var_1522_axes_0 = const()[name = tensor("op_1522_axes_0"), val = tensor([2])]; tensor var_1522_keep_dims_0 = const()[name = tensor("op_1522_keep_dims_0"), val = tensor(false)]; tensor var_1522_cast_fp16 = reduce_sum(axes = var_1522_axes_0, keep_dims = var_1522_keep_dims_0, x = var_1520_cast_fp16)[name = tensor("op_1522_cast_fp16")]; tensor mean_61_cast_fp16 = real_div(x = var_1522_cast_fp16, y = v1_61_cast_fp16)[name = tensor("mean_61_cast_fp16")]; tensor var_1524_axes_0 = const()[name = tensor("op_1524_axes_0"), val = tensor([2])]; tensor var_1524_cast_fp16 = expand_dims(axes = var_1524_axes_0, x = mean_61_cast_fp16)[name = tensor("op_1524_cast_fp16")]; - tensor var_1525_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1524_cast_fp16)[name = tensor("op_1525_cast_fp16")]; - tensor dx2_61_cast_fp16 = mul(x = var_1525_cast_fp16, y = var_1525_cast_fp16)[name = tensor("dx2_61_cast_fp16")]; - tensor var_1527_cast_fp16 = mul(x = weights_125_cast_fp16, y = weights_125_cast_fp16)[name = tensor("op_1527_cast_fp16")]; + tensor var_1525_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1524_cast_fp16)[name = tensor("op_1525_cast_fp16")]; + tensor dx2_61_cast_fp16 = mul(x = var_1525_cast_fp16, y = var_1525_cast_fp16)[name = tensor("dx2_61_cast_fp16")]; + tensor var_1527_cast_fp16 = mul(x = weights_125_cast_fp16, y = weights_125_cast_fp16)[name = tensor("op_1527_cast_fp16")]; tensor v2_61_axes_0 = const()[name = tensor("v2_61_axes_0"), val = tensor([2])]; tensor v2_61_keep_dims_0 = const()[name = tensor("v2_61_keep_dims_0"), val = tensor(false)]; tensor v2_61_cast_fp16 = reduce_sum(axes = v2_61_axes_0, keep_dims = v2_61_keep_dims_0, x = var_1527_cast_fp16)[name = tensor("v2_61_cast_fp16")]; - tensor var_1530_cast_fp16 = mul(x = dx2_61_cast_fp16, y = weights_125_cast_fp16)[name = tensor("op_1530_cast_fp16")]; + tensor var_1530_cast_fp16 = mul(x = dx2_61_cast_fp16, y = weights_125_cast_fp16)[name = tensor("op_1530_cast_fp16")]; tensor var_1532_axes_0 = const()[name = tensor("op_1532_axes_0"), val = tensor([2])]; tensor var_1532_keep_dims_0 = const()[name = tensor("op_1532_keep_dims_0"), val = tensor(false)]; tensor var_1532_cast_fp16 = reduce_sum(axes = var_1532_axes_0, keep_dims = var_1532_keep_dims_0, x = var_1530_cast_fp16)[name = tensor("op_1532_cast_fp16")]; @@ -1501,31 +1501,31 @@ program(1.0) tensor var_1540_interleave_0 = const()[name = tensor("op_1540_interleave_0"), val = tensor(false)]; tensor var_1540_cast_fp16 = concat(axis = var_72, interleave = var_1540_interleave_0, values = (mean_61_cast_fp16, std_61_cast_fp16))[name = tensor("op_1540_cast_fp16")]; tensor var_1542_begin_0 = const()[name = tensor("op_1542_begin_0"), val = tensor([0, 31, 0])]; - tensor var_1542_end_0 = const()[name = tensor("op_1542_end_0"), val = tensor([1, 32, 125])]; + tensor var_1542_end_0 = const()[name = tensor("op_1542_end_0"), val = tensor([1, 32, 375])]; tensor var_1542_end_mask_0 = const()[name = tensor("op_1542_end_mask_0"), val = tensor([true, false, true])]; tensor var_1542_squeeze_mask_0 = const()[name = tensor("op_1542_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1542_cast_fp16 = slice_by_index(begin = var_1542_begin_0, end = var_1542_end_0, end_mask = var_1542_end_mask_0, squeeze_mask = var_1542_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1542_cast_fp16")]; + tensor var_1542_cast_fp16 = slice_by_index(begin = var_1542_begin_0, end = var_1542_end_0, end_mask = var_1542_end_mask_0, squeeze_mask = var_1542_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1542_cast_fp16")]; tensor weights_129_axes_0 = const()[name = tensor("weights_129_axes_0"), val = tensor([1])]; - tensor weights_129_cast_fp16 = expand_dims(axes = weights_129_axes_0, x = var_1542_cast_fp16)[name = tensor("weights_129_cast_fp16")]; + tensor weights_129_cast_fp16 = expand_dims(axes = weights_129_axes_0, x = var_1542_cast_fp16)[name = tensor("weights_129_cast_fp16")]; tensor var_1546_axes_0 = const()[name = tensor("op_1546_axes_0"), val = tensor([2])]; tensor var_1546_keep_dims_0 = const()[name = tensor("op_1546_keep_dims_0"), val = tensor(false)]; tensor var_1546_cast_fp16 = reduce_sum(axes = var_1546_axes_0, keep_dims = var_1546_keep_dims_0, x = weights_129_cast_fp16)[name = tensor("op_1546_cast_fp16")]; tensor var_1547_to_fp16 = const()[name = tensor("op_1547_to_fp16"), val = tensor(0x1p-24)]; tensor v1_63_cast_fp16 = add(x = var_1546_cast_fp16, y = var_1547_to_fp16)[name = tensor("v1_63_cast_fp16")]; - tensor var_1549_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_129_cast_fp16)[name = tensor("op_1549_cast_fp16")]; + tensor var_1549_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_129_cast_fp16)[name = tensor("op_1549_cast_fp16")]; tensor var_1551_axes_0 = const()[name = tensor("op_1551_axes_0"), val = tensor([2])]; tensor var_1551_keep_dims_0 = const()[name = tensor("op_1551_keep_dims_0"), val = tensor(false)]; tensor var_1551_cast_fp16 = reduce_sum(axes = var_1551_axes_0, keep_dims = var_1551_keep_dims_0, x = var_1549_cast_fp16)[name = tensor("op_1551_cast_fp16")]; tensor mean_63_cast_fp16 = real_div(x = var_1551_cast_fp16, y = v1_63_cast_fp16)[name = tensor("mean_63_cast_fp16")]; tensor var_1553_axes_0 = const()[name = tensor("op_1553_axes_0"), val = tensor([2])]; tensor var_1553_cast_fp16 = expand_dims(axes = var_1553_axes_0, x = mean_63_cast_fp16)[name = tensor("op_1553_cast_fp16")]; - tensor var_1554_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1553_cast_fp16)[name = tensor("op_1554_cast_fp16")]; - tensor dx2_63_cast_fp16 = mul(x = var_1554_cast_fp16, y = var_1554_cast_fp16)[name = tensor("dx2_63_cast_fp16")]; - tensor var_1556_cast_fp16 = mul(x = weights_129_cast_fp16, y = weights_129_cast_fp16)[name = tensor("op_1556_cast_fp16")]; + tensor var_1554_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1553_cast_fp16)[name = tensor("op_1554_cast_fp16")]; + tensor dx2_63_cast_fp16 = mul(x = var_1554_cast_fp16, y = var_1554_cast_fp16)[name = tensor("dx2_63_cast_fp16")]; + tensor var_1556_cast_fp16 = mul(x = weights_129_cast_fp16, y = weights_129_cast_fp16)[name = tensor("op_1556_cast_fp16")]; tensor v2_63_axes_0 = const()[name = tensor("v2_63_axes_0"), val = tensor([2])]; tensor v2_63_keep_dims_0 = const()[name = tensor("v2_63_keep_dims_0"), val = tensor(false)]; tensor v2_63_cast_fp16 = reduce_sum(axes = v2_63_axes_0, keep_dims = v2_63_keep_dims_0, x = var_1556_cast_fp16)[name = tensor("v2_63_cast_fp16")]; - tensor var_1559_cast_fp16 = mul(x = dx2_63_cast_fp16, y = weights_129_cast_fp16)[name = tensor("op_1559_cast_fp16")]; + tensor var_1559_cast_fp16 = mul(x = dx2_63_cast_fp16, y = weights_129_cast_fp16)[name = tensor("op_1559_cast_fp16")]; tensor var_1561_axes_0 = const()[name = tensor("op_1561_axes_0"), val = tensor([2])]; tensor var_1561_keep_dims_0 = const()[name = tensor("op_1561_keep_dims_0"), val = tensor(false)]; tensor var_1561_cast_fp16 = reduce_sum(axes = var_1561_axes_0, keep_dims = var_1561_keep_dims_0, x = var_1559_cast_fp16)[name = tensor("op_1561_cast_fp16")]; @@ -1538,31 +1538,31 @@ program(1.0) tensor var_1569_interleave_0 = const()[name = tensor("op_1569_interleave_0"), val = tensor(false)]; tensor var_1569_cast_fp16 = concat(axis = var_72, interleave = var_1569_interleave_0, values = (mean_63_cast_fp16, std_63_cast_fp16))[name = tensor("op_1569_cast_fp16")]; tensor var_1571_begin_0 = const()[name = tensor("op_1571_begin_0"), val = tensor([0, 32, 0])]; - tensor var_1571_end_0 = const()[name = tensor("op_1571_end_0"), val = tensor([1, 33, 125])]; + tensor var_1571_end_0 = const()[name = tensor("op_1571_end_0"), val = tensor([1, 33, 375])]; tensor var_1571_end_mask_0 = const()[name = tensor("op_1571_end_mask_0"), val = tensor([true, false, true])]; tensor var_1571_squeeze_mask_0 = const()[name = tensor("op_1571_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1571_cast_fp16 = slice_by_index(begin = var_1571_begin_0, end = var_1571_end_0, end_mask = var_1571_end_mask_0, squeeze_mask = var_1571_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1571_cast_fp16")]; + tensor var_1571_cast_fp16 = slice_by_index(begin = var_1571_begin_0, end = var_1571_end_0, end_mask = var_1571_end_mask_0, squeeze_mask = var_1571_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1571_cast_fp16")]; tensor weights_133_axes_0 = const()[name = tensor("weights_133_axes_0"), val = tensor([1])]; - tensor weights_133_cast_fp16 = expand_dims(axes = weights_133_axes_0, x = var_1571_cast_fp16)[name = tensor("weights_133_cast_fp16")]; + tensor weights_133_cast_fp16 = expand_dims(axes = weights_133_axes_0, x = var_1571_cast_fp16)[name = tensor("weights_133_cast_fp16")]; tensor var_1575_axes_0 = const()[name = tensor("op_1575_axes_0"), val = tensor([2])]; tensor var_1575_keep_dims_0 = const()[name = tensor("op_1575_keep_dims_0"), val = tensor(false)]; tensor var_1575_cast_fp16 = reduce_sum(axes = var_1575_axes_0, keep_dims = var_1575_keep_dims_0, x = weights_133_cast_fp16)[name = tensor("op_1575_cast_fp16")]; tensor var_1576_to_fp16 = const()[name = tensor("op_1576_to_fp16"), val = tensor(0x1p-24)]; tensor v1_65_cast_fp16 = add(x = var_1575_cast_fp16, y = var_1576_to_fp16)[name = tensor("v1_65_cast_fp16")]; - tensor var_1578_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_133_cast_fp16)[name = tensor("op_1578_cast_fp16")]; + tensor var_1578_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_133_cast_fp16)[name = tensor("op_1578_cast_fp16")]; tensor var_1580_axes_0 = const()[name = tensor("op_1580_axes_0"), val = tensor([2])]; tensor var_1580_keep_dims_0 = const()[name = tensor("op_1580_keep_dims_0"), val = tensor(false)]; tensor var_1580_cast_fp16 = reduce_sum(axes = var_1580_axes_0, keep_dims = var_1580_keep_dims_0, x = var_1578_cast_fp16)[name = tensor("op_1580_cast_fp16")]; tensor mean_65_cast_fp16 = real_div(x = var_1580_cast_fp16, y = v1_65_cast_fp16)[name = tensor("mean_65_cast_fp16")]; tensor var_1582_axes_0 = const()[name = tensor("op_1582_axes_0"), val = tensor([2])]; tensor var_1582_cast_fp16 = expand_dims(axes = var_1582_axes_0, x = mean_65_cast_fp16)[name = tensor("op_1582_cast_fp16")]; - tensor var_1583_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1582_cast_fp16)[name = tensor("op_1583_cast_fp16")]; - tensor dx2_65_cast_fp16 = mul(x = var_1583_cast_fp16, y = var_1583_cast_fp16)[name = tensor("dx2_65_cast_fp16")]; - tensor var_1585_cast_fp16 = mul(x = weights_133_cast_fp16, y = weights_133_cast_fp16)[name = tensor("op_1585_cast_fp16")]; + tensor var_1583_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1582_cast_fp16)[name = tensor("op_1583_cast_fp16")]; + tensor dx2_65_cast_fp16 = mul(x = var_1583_cast_fp16, y = var_1583_cast_fp16)[name = tensor("dx2_65_cast_fp16")]; + tensor var_1585_cast_fp16 = mul(x = weights_133_cast_fp16, y = weights_133_cast_fp16)[name = tensor("op_1585_cast_fp16")]; tensor v2_65_axes_0 = const()[name = tensor("v2_65_axes_0"), val = tensor([2])]; tensor v2_65_keep_dims_0 = const()[name = tensor("v2_65_keep_dims_0"), val = tensor(false)]; tensor v2_65_cast_fp16 = reduce_sum(axes = v2_65_axes_0, keep_dims = v2_65_keep_dims_0, x = var_1585_cast_fp16)[name = tensor("v2_65_cast_fp16")]; - tensor var_1588_cast_fp16 = mul(x = dx2_65_cast_fp16, y = weights_133_cast_fp16)[name = tensor("op_1588_cast_fp16")]; + tensor var_1588_cast_fp16 = mul(x = dx2_65_cast_fp16, y = weights_133_cast_fp16)[name = tensor("op_1588_cast_fp16")]; tensor var_1590_axes_0 = const()[name = tensor("op_1590_axes_0"), val = tensor([2])]; tensor var_1590_keep_dims_0 = const()[name = tensor("op_1590_keep_dims_0"), val = tensor(false)]; tensor var_1590_cast_fp16 = reduce_sum(axes = var_1590_axes_0, keep_dims = var_1590_keep_dims_0, x = var_1588_cast_fp16)[name = tensor("op_1590_cast_fp16")]; @@ -1575,31 +1575,31 @@ program(1.0) tensor var_1598_interleave_0 = const()[name = tensor("op_1598_interleave_0"), val = tensor(false)]; tensor var_1598_cast_fp16 = concat(axis = var_72, interleave = var_1598_interleave_0, values = (mean_65_cast_fp16, std_65_cast_fp16))[name = tensor("op_1598_cast_fp16")]; tensor var_1600_begin_0 = const()[name = tensor("op_1600_begin_0"), val = tensor([0, 33, 0])]; - tensor var_1600_end_0 = const()[name = tensor("op_1600_end_0"), val = tensor([1, 34, 125])]; + tensor var_1600_end_0 = const()[name = tensor("op_1600_end_0"), val = tensor([1, 34, 375])]; tensor var_1600_end_mask_0 = const()[name = tensor("op_1600_end_mask_0"), val = tensor([true, false, true])]; tensor var_1600_squeeze_mask_0 = const()[name = tensor("op_1600_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1600_cast_fp16 = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, squeeze_mask = var_1600_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1600_cast_fp16")]; + tensor var_1600_cast_fp16 = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, squeeze_mask = var_1600_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1600_cast_fp16")]; tensor weights_137_axes_0 = const()[name = tensor("weights_137_axes_0"), val = tensor([1])]; - tensor weights_137_cast_fp16 = expand_dims(axes = weights_137_axes_0, x = var_1600_cast_fp16)[name = tensor("weights_137_cast_fp16")]; + tensor weights_137_cast_fp16 = expand_dims(axes = weights_137_axes_0, x = var_1600_cast_fp16)[name = tensor("weights_137_cast_fp16")]; tensor var_1604_axes_0 = const()[name = tensor("op_1604_axes_0"), val = tensor([2])]; tensor var_1604_keep_dims_0 = const()[name = tensor("op_1604_keep_dims_0"), val = tensor(false)]; tensor var_1604_cast_fp16 = reduce_sum(axes = var_1604_axes_0, keep_dims = var_1604_keep_dims_0, x = weights_137_cast_fp16)[name = tensor("op_1604_cast_fp16")]; tensor var_1605_to_fp16 = const()[name = tensor("op_1605_to_fp16"), val = tensor(0x1p-24)]; tensor v1_67_cast_fp16 = add(x = var_1604_cast_fp16, y = var_1605_to_fp16)[name = tensor("v1_67_cast_fp16")]; - tensor var_1607_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_137_cast_fp16)[name = tensor("op_1607_cast_fp16")]; + tensor var_1607_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_137_cast_fp16)[name = tensor("op_1607_cast_fp16")]; tensor var_1609_axes_0 = const()[name = tensor("op_1609_axes_0"), val = tensor([2])]; tensor var_1609_keep_dims_0 = const()[name = tensor("op_1609_keep_dims_0"), val = tensor(false)]; tensor var_1609_cast_fp16 = reduce_sum(axes = var_1609_axes_0, keep_dims = var_1609_keep_dims_0, x = var_1607_cast_fp16)[name = tensor("op_1609_cast_fp16")]; tensor mean_67_cast_fp16 = real_div(x = var_1609_cast_fp16, y = v1_67_cast_fp16)[name = tensor("mean_67_cast_fp16")]; tensor var_1611_axes_0 = const()[name = tensor("op_1611_axes_0"), val = tensor([2])]; tensor var_1611_cast_fp16 = expand_dims(axes = var_1611_axes_0, x = mean_67_cast_fp16)[name = tensor("op_1611_cast_fp16")]; - tensor var_1612_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1611_cast_fp16)[name = tensor("op_1612_cast_fp16")]; - tensor dx2_67_cast_fp16 = mul(x = var_1612_cast_fp16, y = var_1612_cast_fp16)[name = tensor("dx2_67_cast_fp16")]; - tensor var_1614_cast_fp16 = mul(x = weights_137_cast_fp16, y = weights_137_cast_fp16)[name = tensor("op_1614_cast_fp16")]; + tensor var_1612_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1611_cast_fp16)[name = tensor("op_1612_cast_fp16")]; + tensor dx2_67_cast_fp16 = mul(x = var_1612_cast_fp16, y = var_1612_cast_fp16)[name = tensor("dx2_67_cast_fp16")]; + tensor var_1614_cast_fp16 = mul(x = weights_137_cast_fp16, y = weights_137_cast_fp16)[name = tensor("op_1614_cast_fp16")]; tensor v2_67_axes_0 = const()[name = tensor("v2_67_axes_0"), val = tensor([2])]; tensor v2_67_keep_dims_0 = const()[name = tensor("v2_67_keep_dims_0"), val = tensor(false)]; tensor v2_67_cast_fp16 = reduce_sum(axes = v2_67_axes_0, keep_dims = v2_67_keep_dims_0, x = var_1614_cast_fp16)[name = tensor("v2_67_cast_fp16")]; - tensor var_1617_cast_fp16 = mul(x = dx2_67_cast_fp16, y = weights_137_cast_fp16)[name = tensor("op_1617_cast_fp16")]; + tensor var_1617_cast_fp16 = mul(x = dx2_67_cast_fp16, y = weights_137_cast_fp16)[name = tensor("op_1617_cast_fp16")]; tensor var_1619_axes_0 = const()[name = tensor("op_1619_axes_0"), val = tensor([2])]; tensor var_1619_keep_dims_0 = const()[name = tensor("op_1619_keep_dims_0"), val = tensor(false)]; tensor var_1619_cast_fp16 = reduce_sum(axes = var_1619_axes_0, keep_dims = var_1619_keep_dims_0, x = var_1617_cast_fp16)[name = tensor("op_1619_cast_fp16")]; @@ -1612,31 +1612,31 @@ program(1.0) tensor var_1627_interleave_0 = const()[name = tensor("op_1627_interleave_0"), val = tensor(false)]; tensor var_1627_cast_fp16 = concat(axis = var_72, interleave = var_1627_interleave_0, values = (mean_67_cast_fp16, std_67_cast_fp16))[name = tensor("op_1627_cast_fp16")]; tensor var_1629_begin_0 = const()[name = tensor("op_1629_begin_0"), val = tensor([0, 34, 0])]; - tensor var_1629_end_0 = const()[name = tensor("op_1629_end_0"), val = tensor([1, 35, 125])]; + tensor var_1629_end_0 = const()[name = tensor("op_1629_end_0"), val = tensor([1, 35, 375])]; tensor var_1629_end_mask_0 = const()[name = tensor("op_1629_end_mask_0"), val = tensor([true, false, true])]; tensor var_1629_squeeze_mask_0 = const()[name = tensor("op_1629_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1629_cast_fp16 = slice_by_index(begin = var_1629_begin_0, end = var_1629_end_0, end_mask = var_1629_end_mask_0, squeeze_mask = var_1629_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1629_cast_fp16")]; + tensor var_1629_cast_fp16 = slice_by_index(begin = var_1629_begin_0, end = var_1629_end_0, end_mask = var_1629_end_mask_0, squeeze_mask = var_1629_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1629_cast_fp16")]; tensor weights_141_axes_0 = const()[name = tensor("weights_141_axes_0"), val = tensor([1])]; - tensor weights_141_cast_fp16 = expand_dims(axes = weights_141_axes_0, x = var_1629_cast_fp16)[name = tensor("weights_141_cast_fp16")]; + tensor weights_141_cast_fp16 = expand_dims(axes = weights_141_axes_0, x = var_1629_cast_fp16)[name = tensor("weights_141_cast_fp16")]; tensor var_1633_axes_0 = const()[name = tensor("op_1633_axes_0"), val = tensor([2])]; tensor var_1633_keep_dims_0 = const()[name = tensor("op_1633_keep_dims_0"), val = tensor(false)]; tensor var_1633_cast_fp16 = reduce_sum(axes = var_1633_axes_0, keep_dims = var_1633_keep_dims_0, x = weights_141_cast_fp16)[name = tensor("op_1633_cast_fp16")]; tensor var_1634_to_fp16 = const()[name = tensor("op_1634_to_fp16"), val = tensor(0x1p-24)]; tensor v1_69_cast_fp16 = add(x = var_1633_cast_fp16, y = var_1634_to_fp16)[name = tensor("v1_69_cast_fp16")]; - tensor var_1636_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_141_cast_fp16)[name = tensor("op_1636_cast_fp16")]; + tensor var_1636_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_141_cast_fp16)[name = tensor("op_1636_cast_fp16")]; tensor var_1638_axes_0 = const()[name = tensor("op_1638_axes_0"), val = tensor([2])]; tensor var_1638_keep_dims_0 = const()[name = tensor("op_1638_keep_dims_0"), val = tensor(false)]; tensor var_1638_cast_fp16 = reduce_sum(axes = var_1638_axes_0, keep_dims = var_1638_keep_dims_0, x = var_1636_cast_fp16)[name = tensor("op_1638_cast_fp16")]; tensor mean_69_cast_fp16 = real_div(x = var_1638_cast_fp16, y = v1_69_cast_fp16)[name = tensor("mean_69_cast_fp16")]; tensor var_1640_axes_0 = const()[name = tensor("op_1640_axes_0"), val = tensor([2])]; tensor var_1640_cast_fp16 = expand_dims(axes = var_1640_axes_0, x = mean_69_cast_fp16)[name = tensor("op_1640_cast_fp16")]; - tensor var_1641_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1640_cast_fp16)[name = tensor("op_1641_cast_fp16")]; - tensor dx2_69_cast_fp16 = mul(x = var_1641_cast_fp16, y = var_1641_cast_fp16)[name = tensor("dx2_69_cast_fp16")]; - tensor var_1643_cast_fp16 = mul(x = weights_141_cast_fp16, y = weights_141_cast_fp16)[name = tensor("op_1643_cast_fp16")]; + tensor var_1641_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1640_cast_fp16)[name = tensor("op_1641_cast_fp16")]; + tensor dx2_69_cast_fp16 = mul(x = var_1641_cast_fp16, y = var_1641_cast_fp16)[name = tensor("dx2_69_cast_fp16")]; + tensor var_1643_cast_fp16 = mul(x = weights_141_cast_fp16, y = weights_141_cast_fp16)[name = tensor("op_1643_cast_fp16")]; tensor v2_69_axes_0 = const()[name = tensor("v2_69_axes_0"), val = tensor([2])]; tensor v2_69_keep_dims_0 = const()[name = tensor("v2_69_keep_dims_0"), val = tensor(false)]; tensor v2_69_cast_fp16 = reduce_sum(axes = v2_69_axes_0, keep_dims = v2_69_keep_dims_0, x = var_1643_cast_fp16)[name = tensor("v2_69_cast_fp16")]; - tensor var_1646_cast_fp16 = mul(x = dx2_69_cast_fp16, y = weights_141_cast_fp16)[name = tensor("op_1646_cast_fp16")]; + tensor var_1646_cast_fp16 = mul(x = dx2_69_cast_fp16, y = weights_141_cast_fp16)[name = tensor("op_1646_cast_fp16")]; tensor var_1648_axes_0 = const()[name = tensor("op_1648_axes_0"), val = tensor([2])]; tensor var_1648_keep_dims_0 = const()[name = tensor("op_1648_keep_dims_0"), val = tensor(false)]; tensor var_1648_cast_fp16 = reduce_sum(axes = var_1648_axes_0, keep_dims = var_1648_keep_dims_0, x = var_1646_cast_fp16)[name = tensor("op_1648_cast_fp16")]; @@ -1649,31 +1649,31 @@ program(1.0) tensor var_1656_interleave_0 = const()[name = tensor("op_1656_interleave_0"), val = tensor(false)]; tensor var_1656_cast_fp16 = concat(axis = var_72, interleave = var_1656_interleave_0, values = (mean_69_cast_fp16, std_69_cast_fp16))[name = tensor("op_1656_cast_fp16")]; tensor var_1658_begin_0 = const()[name = tensor("op_1658_begin_0"), val = tensor([0, 35, 0])]; - tensor var_1658_end_0 = const()[name = tensor("op_1658_end_0"), val = tensor([1, 36, 125])]; + tensor var_1658_end_0 = const()[name = tensor("op_1658_end_0"), val = tensor([1, 36, 375])]; tensor var_1658_end_mask_0 = const()[name = tensor("op_1658_end_mask_0"), val = tensor([true, false, true])]; tensor var_1658_squeeze_mask_0 = const()[name = tensor("op_1658_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1658_cast_fp16 = slice_by_index(begin = var_1658_begin_0, end = var_1658_end_0, end_mask = var_1658_end_mask_0, squeeze_mask = var_1658_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1658_cast_fp16")]; + tensor var_1658_cast_fp16 = slice_by_index(begin = var_1658_begin_0, end = var_1658_end_0, end_mask = var_1658_end_mask_0, squeeze_mask = var_1658_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1658_cast_fp16")]; tensor weights_145_axes_0 = const()[name = tensor("weights_145_axes_0"), val = tensor([1])]; - tensor weights_145_cast_fp16 = expand_dims(axes = weights_145_axes_0, x = var_1658_cast_fp16)[name = tensor("weights_145_cast_fp16")]; + tensor weights_145_cast_fp16 = expand_dims(axes = weights_145_axes_0, x = var_1658_cast_fp16)[name = tensor("weights_145_cast_fp16")]; tensor var_1662_axes_0 = const()[name = tensor("op_1662_axes_0"), val = tensor([2])]; tensor var_1662_keep_dims_0 = const()[name = tensor("op_1662_keep_dims_0"), val = tensor(false)]; tensor var_1662_cast_fp16 = reduce_sum(axes = var_1662_axes_0, keep_dims = var_1662_keep_dims_0, x = weights_145_cast_fp16)[name = tensor("op_1662_cast_fp16")]; tensor var_1663_to_fp16 = const()[name = tensor("op_1663_to_fp16"), val = tensor(0x1p-24)]; tensor v1_71_cast_fp16 = add(x = var_1662_cast_fp16, y = var_1663_to_fp16)[name = tensor("v1_71_cast_fp16")]; - tensor var_1665_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_145_cast_fp16)[name = tensor("op_1665_cast_fp16")]; + tensor var_1665_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_145_cast_fp16)[name = tensor("op_1665_cast_fp16")]; tensor var_1667_axes_0 = const()[name = tensor("op_1667_axes_0"), val = tensor([2])]; tensor var_1667_keep_dims_0 = const()[name = tensor("op_1667_keep_dims_0"), val = tensor(false)]; tensor var_1667_cast_fp16 = reduce_sum(axes = var_1667_axes_0, keep_dims = var_1667_keep_dims_0, x = var_1665_cast_fp16)[name = tensor("op_1667_cast_fp16")]; tensor mean_71_cast_fp16 = real_div(x = var_1667_cast_fp16, y = v1_71_cast_fp16)[name = tensor("mean_71_cast_fp16")]; tensor var_1669_axes_0 = const()[name = tensor("op_1669_axes_0"), val = tensor([2])]; tensor var_1669_cast_fp16 = expand_dims(axes = var_1669_axes_0, x = mean_71_cast_fp16)[name = tensor("op_1669_cast_fp16")]; - tensor var_1670_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1669_cast_fp16)[name = tensor("op_1670_cast_fp16")]; - tensor dx2_71_cast_fp16 = mul(x = var_1670_cast_fp16, y = var_1670_cast_fp16)[name = tensor("dx2_71_cast_fp16")]; - tensor var_1672_cast_fp16 = mul(x = weights_145_cast_fp16, y = weights_145_cast_fp16)[name = tensor("op_1672_cast_fp16")]; + tensor var_1670_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1669_cast_fp16)[name = tensor("op_1670_cast_fp16")]; + tensor dx2_71_cast_fp16 = mul(x = var_1670_cast_fp16, y = var_1670_cast_fp16)[name = tensor("dx2_71_cast_fp16")]; + tensor var_1672_cast_fp16 = mul(x = weights_145_cast_fp16, y = weights_145_cast_fp16)[name = tensor("op_1672_cast_fp16")]; tensor v2_71_axes_0 = const()[name = tensor("v2_71_axes_0"), val = tensor([2])]; tensor v2_71_keep_dims_0 = const()[name = tensor("v2_71_keep_dims_0"), val = tensor(false)]; tensor v2_71_cast_fp16 = reduce_sum(axes = v2_71_axes_0, keep_dims = v2_71_keep_dims_0, x = var_1672_cast_fp16)[name = tensor("v2_71_cast_fp16")]; - tensor var_1675_cast_fp16 = mul(x = dx2_71_cast_fp16, y = weights_145_cast_fp16)[name = tensor("op_1675_cast_fp16")]; + tensor var_1675_cast_fp16 = mul(x = dx2_71_cast_fp16, y = weights_145_cast_fp16)[name = tensor("op_1675_cast_fp16")]; tensor var_1677_axes_0 = const()[name = tensor("op_1677_axes_0"), val = tensor([2])]; tensor var_1677_keep_dims_0 = const()[name = tensor("op_1677_keep_dims_0"), val = tensor(false)]; tensor var_1677_cast_fp16 = reduce_sum(axes = var_1677_axes_0, keep_dims = var_1677_keep_dims_0, x = var_1675_cast_fp16)[name = tensor("op_1677_cast_fp16")]; @@ -1686,31 +1686,31 @@ program(1.0) tensor var_1685_interleave_0 = const()[name = tensor("op_1685_interleave_0"), val = tensor(false)]; tensor var_1685_cast_fp16 = concat(axis = var_72, interleave = var_1685_interleave_0, values = (mean_71_cast_fp16, std_71_cast_fp16))[name = tensor("op_1685_cast_fp16")]; tensor var_1687_begin_0 = const()[name = tensor("op_1687_begin_0"), val = tensor([0, 36, 0])]; - tensor var_1687_end_0 = const()[name = tensor("op_1687_end_0"), val = tensor([1, 37, 125])]; + tensor var_1687_end_0 = const()[name = tensor("op_1687_end_0"), val = tensor([1, 37, 375])]; tensor var_1687_end_mask_0 = const()[name = tensor("op_1687_end_mask_0"), val = tensor([true, false, true])]; tensor var_1687_squeeze_mask_0 = const()[name = tensor("op_1687_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1687_cast_fp16 = slice_by_index(begin = var_1687_begin_0, end = var_1687_end_0, end_mask = var_1687_end_mask_0, squeeze_mask = var_1687_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1687_cast_fp16")]; + tensor var_1687_cast_fp16 = slice_by_index(begin = var_1687_begin_0, end = var_1687_end_0, end_mask = var_1687_end_mask_0, squeeze_mask = var_1687_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1687_cast_fp16")]; tensor weights_149_axes_0 = const()[name = tensor("weights_149_axes_0"), val = tensor([1])]; - tensor weights_149_cast_fp16 = expand_dims(axes = weights_149_axes_0, x = var_1687_cast_fp16)[name = tensor("weights_149_cast_fp16")]; + tensor weights_149_cast_fp16 = expand_dims(axes = weights_149_axes_0, x = var_1687_cast_fp16)[name = tensor("weights_149_cast_fp16")]; tensor var_1691_axes_0 = const()[name = tensor("op_1691_axes_0"), val = tensor([2])]; tensor var_1691_keep_dims_0 = const()[name = tensor("op_1691_keep_dims_0"), val = tensor(false)]; tensor var_1691_cast_fp16 = reduce_sum(axes = var_1691_axes_0, keep_dims = var_1691_keep_dims_0, x = weights_149_cast_fp16)[name = tensor("op_1691_cast_fp16")]; tensor var_1692_to_fp16 = const()[name = tensor("op_1692_to_fp16"), val = tensor(0x1p-24)]; tensor v1_73_cast_fp16 = add(x = var_1691_cast_fp16, y = var_1692_to_fp16)[name = tensor("v1_73_cast_fp16")]; - tensor var_1694_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_149_cast_fp16)[name = tensor("op_1694_cast_fp16")]; + tensor var_1694_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_149_cast_fp16)[name = tensor("op_1694_cast_fp16")]; tensor var_1696_axes_0 = const()[name = tensor("op_1696_axes_0"), val = tensor([2])]; tensor var_1696_keep_dims_0 = const()[name = tensor("op_1696_keep_dims_0"), val = tensor(false)]; tensor var_1696_cast_fp16 = reduce_sum(axes = var_1696_axes_0, keep_dims = var_1696_keep_dims_0, x = var_1694_cast_fp16)[name = tensor("op_1696_cast_fp16")]; tensor mean_73_cast_fp16 = real_div(x = var_1696_cast_fp16, y = v1_73_cast_fp16)[name = tensor("mean_73_cast_fp16")]; tensor var_1698_axes_0 = const()[name = tensor("op_1698_axes_0"), val = tensor([2])]; tensor var_1698_cast_fp16 = expand_dims(axes = var_1698_axes_0, x = mean_73_cast_fp16)[name = tensor("op_1698_cast_fp16")]; - tensor var_1699_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1698_cast_fp16)[name = tensor("op_1699_cast_fp16")]; - tensor dx2_73_cast_fp16 = mul(x = var_1699_cast_fp16, y = var_1699_cast_fp16)[name = tensor("dx2_73_cast_fp16")]; - tensor var_1701_cast_fp16 = mul(x = weights_149_cast_fp16, y = weights_149_cast_fp16)[name = tensor("op_1701_cast_fp16")]; + tensor var_1699_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1698_cast_fp16)[name = tensor("op_1699_cast_fp16")]; + tensor dx2_73_cast_fp16 = mul(x = var_1699_cast_fp16, y = var_1699_cast_fp16)[name = tensor("dx2_73_cast_fp16")]; + tensor var_1701_cast_fp16 = mul(x = weights_149_cast_fp16, y = weights_149_cast_fp16)[name = tensor("op_1701_cast_fp16")]; tensor v2_73_axes_0 = const()[name = tensor("v2_73_axes_0"), val = tensor([2])]; tensor v2_73_keep_dims_0 = const()[name = tensor("v2_73_keep_dims_0"), val = tensor(false)]; tensor v2_73_cast_fp16 = reduce_sum(axes = v2_73_axes_0, keep_dims = v2_73_keep_dims_0, x = var_1701_cast_fp16)[name = tensor("v2_73_cast_fp16")]; - tensor var_1704_cast_fp16 = mul(x = dx2_73_cast_fp16, y = weights_149_cast_fp16)[name = tensor("op_1704_cast_fp16")]; + tensor var_1704_cast_fp16 = mul(x = dx2_73_cast_fp16, y = weights_149_cast_fp16)[name = tensor("op_1704_cast_fp16")]; tensor var_1706_axes_0 = const()[name = tensor("op_1706_axes_0"), val = tensor([2])]; tensor var_1706_keep_dims_0 = const()[name = tensor("op_1706_keep_dims_0"), val = tensor(false)]; tensor var_1706_cast_fp16 = reduce_sum(axes = var_1706_axes_0, keep_dims = var_1706_keep_dims_0, x = var_1704_cast_fp16)[name = tensor("op_1706_cast_fp16")]; @@ -1723,31 +1723,31 @@ program(1.0) tensor var_1714_interleave_0 = const()[name = tensor("op_1714_interleave_0"), val = tensor(false)]; tensor var_1714_cast_fp16 = concat(axis = var_72, interleave = var_1714_interleave_0, values = (mean_73_cast_fp16, std_73_cast_fp16))[name = tensor("op_1714_cast_fp16")]; tensor var_1716_begin_0 = const()[name = tensor("op_1716_begin_0"), val = tensor([0, 37, 0])]; - tensor var_1716_end_0 = const()[name = tensor("op_1716_end_0"), val = tensor([1, 38, 125])]; + tensor var_1716_end_0 = const()[name = tensor("op_1716_end_0"), val = tensor([1, 38, 375])]; tensor var_1716_end_mask_0 = const()[name = tensor("op_1716_end_mask_0"), val = tensor([true, false, true])]; tensor var_1716_squeeze_mask_0 = const()[name = tensor("op_1716_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1716_cast_fp16 = slice_by_index(begin = var_1716_begin_0, end = var_1716_end_0, end_mask = var_1716_end_mask_0, squeeze_mask = var_1716_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1716_cast_fp16")]; + tensor var_1716_cast_fp16 = slice_by_index(begin = var_1716_begin_0, end = var_1716_end_0, end_mask = var_1716_end_mask_0, squeeze_mask = var_1716_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1716_cast_fp16")]; tensor weights_153_axes_0 = const()[name = tensor("weights_153_axes_0"), val = tensor([1])]; - tensor weights_153_cast_fp16 = expand_dims(axes = weights_153_axes_0, x = var_1716_cast_fp16)[name = tensor("weights_153_cast_fp16")]; + tensor weights_153_cast_fp16 = expand_dims(axes = weights_153_axes_0, x = var_1716_cast_fp16)[name = tensor("weights_153_cast_fp16")]; tensor var_1720_axes_0 = const()[name = tensor("op_1720_axes_0"), val = tensor([2])]; tensor var_1720_keep_dims_0 = const()[name = tensor("op_1720_keep_dims_0"), val = tensor(false)]; tensor var_1720_cast_fp16 = reduce_sum(axes = var_1720_axes_0, keep_dims = var_1720_keep_dims_0, x = weights_153_cast_fp16)[name = tensor("op_1720_cast_fp16")]; tensor var_1721_to_fp16 = const()[name = tensor("op_1721_to_fp16"), val = tensor(0x1p-24)]; tensor v1_75_cast_fp16 = add(x = var_1720_cast_fp16, y = var_1721_to_fp16)[name = tensor("v1_75_cast_fp16")]; - tensor var_1723_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_153_cast_fp16)[name = tensor("op_1723_cast_fp16")]; + tensor var_1723_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_153_cast_fp16)[name = tensor("op_1723_cast_fp16")]; tensor var_1725_axes_0 = const()[name = tensor("op_1725_axes_0"), val = tensor([2])]; tensor var_1725_keep_dims_0 = const()[name = tensor("op_1725_keep_dims_0"), val = tensor(false)]; tensor var_1725_cast_fp16 = reduce_sum(axes = var_1725_axes_0, keep_dims = var_1725_keep_dims_0, x = var_1723_cast_fp16)[name = tensor("op_1725_cast_fp16")]; tensor mean_75_cast_fp16 = real_div(x = var_1725_cast_fp16, y = v1_75_cast_fp16)[name = tensor("mean_75_cast_fp16")]; tensor var_1727_axes_0 = const()[name = tensor("op_1727_axes_0"), val = tensor([2])]; tensor var_1727_cast_fp16 = expand_dims(axes = var_1727_axes_0, x = mean_75_cast_fp16)[name = tensor("op_1727_cast_fp16")]; - tensor var_1728_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1727_cast_fp16)[name = tensor("op_1728_cast_fp16")]; - tensor dx2_75_cast_fp16 = mul(x = var_1728_cast_fp16, y = var_1728_cast_fp16)[name = tensor("dx2_75_cast_fp16")]; - tensor var_1730_cast_fp16 = mul(x = weights_153_cast_fp16, y = weights_153_cast_fp16)[name = tensor("op_1730_cast_fp16")]; + tensor var_1728_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1727_cast_fp16)[name = tensor("op_1728_cast_fp16")]; + tensor dx2_75_cast_fp16 = mul(x = var_1728_cast_fp16, y = var_1728_cast_fp16)[name = tensor("dx2_75_cast_fp16")]; + tensor var_1730_cast_fp16 = mul(x = weights_153_cast_fp16, y = weights_153_cast_fp16)[name = tensor("op_1730_cast_fp16")]; tensor v2_75_axes_0 = const()[name = tensor("v2_75_axes_0"), val = tensor([2])]; tensor v2_75_keep_dims_0 = const()[name = tensor("v2_75_keep_dims_0"), val = tensor(false)]; tensor v2_75_cast_fp16 = reduce_sum(axes = v2_75_axes_0, keep_dims = v2_75_keep_dims_0, x = var_1730_cast_fp16)[name = tensor("v2_75_cast_fp16")]; - tensor var_1733_cast_fp16 = mul(x = dx2_75_cast_fp16, y = weights_153_cast_fp16)[name = tensor("op_1733_cast_fp16")]; + tensor var_1733_cast_fp16 = mul(x = dx2_75_cast_fp16, y = weights_153_cast_fp16)[name = tensor("op_1733_cast_fp16")]; tensor var_1735_axes_0 = const()[name = tensor("op_1735_axes_0"), val = tensor([2])]; tensor var_1735_keep_dims_0 = const()[name = tensor("op_1735_keep_dims_0"), val = tensor(false)]; tensor var_1735_cast_fp16 = reduce_sum(axes = var_1735_axes_0, keep_dims = var_1735_keep_dims_0, x = var_1733_cast_fp16)[name = tensor("op_1735_cast_fp16")]; @@ -1760,31 +1760,31 @@ program(1.0) tensor var_1743_interleave_0 = const()[name = tensor("op_1743_interleave_0"), val = tensor(false)]; tensor var_1743_cast_fp16 = concat(axis = var_72, interleave = var_1743_interleave_0, values = (mean_75_cast_fp16, std_75_cast_fp16))[name = tensor("op_1743_cast_fp16")]; tensor var_1745_begin_0 = const()[name = tensor("op_1745_begin_0"), val = tensor([0, 38, 0])]; - tensor var_1745_end_0 = const()[name = tensor("op_1745_end_0"), val = tensor([1, 39, 125])]; + tensor var_1745_end_0 = const()[name = tensor("op_1745_end_0"), val = tensor([1, 39, 375])]; tensor var_1745_end_mask_0 = const()[name = tensor("op_1745_end_mask_0"), val = tensor([true, false, true])]; tensor var_1745_squeeze_mask_0 = const()[name = tensor("op_1745_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1745_cast_fp16 = slice_by_index(begin = var_1745_begin_0, end = var_1745_end_0, end_mask = var_1745_end_mask_0, squeeze_mask = var_1745_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1745_cast_fp16")]; + tensor var_1745_cast_fp16 = slice_by_index(begin = var_1745_begin_0, end = var_1745_end_0, end_mask = var_1745_end_mask_0, squeeze_mask = var_1745_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1745_cast_fp16")]; tensor weights_157_axes_0 = const()[name = tensor("weights_157_axes_0"), val = tensor([1])]; - tensor weights_157_cast_fp16 = expand_dims(axes = weights_157_axes_0, x = var_1745_cast_fp16)[name = tensor("weights_157_cast_fp16")]; + tensor weights_157_cast_fp16 = expand_dims(axes = weights_157_axes_0, x = var_1745_cast_fp16)[name = tensor("weights_157_cast_fp16")]; tensor var_1749_axes_0 = const()[name = tensor("op_1749_axes_0"), val = tensor([2])]; tensor var_1749_keep_dims_0 = const()[name = tensor("op_1749_keep_dims_0"), val = tensor(false)]; tensor var_1749_cast_fp16 = reduce_sum(axes = var_1749_axes_0, keep_dims = var_1749_keep_dims_0, x = weights_157_cast_fp16)[name = tensor("op_1749_cast_fp16")]; tensor var_1750_to_fp16 = const()[name = tensor("op_1750_to_fp16"), val = tensor(0x1p-24)]; tensor v1_77_cast_fp16 = add(x = var_1749_cast_fp16, y = var_1750_to_fp16)[name = tensor("v1_77_cast_fp16")]; - tensor var_1752_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_157_cast_fp16)[name = tensor("op_1752_cast_fp16")]; + tensor var_1752_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_157_cast_fp16)[name = tensor("op_1752_cast_fp16")]; tensor var_1754_axes_0 = const()[name = tensor("op_1754_axes_0"), val = tensor([2])]; tensor var_1754_keep_dims_0 = const()[name = tensor("op_1754_keep_dims_0"), val = tensor(false)]; tensor var_1754_cast_fp16 = reduce_sum(axes = var_1754_axes_0, keep_dims = var_1754_keep_dims_0, x = var_1752_cast_fp16)[name = tensor("op_1754_cast_fp16")]; tensor mean_77_cast_fp16 = real_div(x = var_1754_cast_fp16, y = v1_77_cast_fp16)[name = tensor("mean_77_cast_fp16")]; tensor var_1756_axes_0 = const()[name = tensor("op_1756_axes_0"), val = tensor([2])]; tensor var_1756_cast_fp16 = expand_dims(axes = var_1756_axes_0, x = mean_77_cast_fp16)[name = tensor("op_1756_cast_fp16")]; - tensor var_1757_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1756_cast_fp16)[name = tensor("op_1757_cast_fp16")]; - tensor dx2_77_cast_fp16 = mul(x = var_1757_cast_fp16, y = var_1757_cast_fp16)[name = tensor("dx2_77_cast_fp16")]; - tensor var_1759_cast_fp16 = mul(x = weights_157_cast_fp16, y = weights_157_cast_fp16)[name = tensor("op_1759_cast_fp16")]; + tensor var_1757_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1756_cast_fp16)[name = tensor("op_1757_cast_fp16")]; + tensor dx2_77_cast_fp16 = mul(x = var_1757_cast_fp16, y = var_1757_cast_fp16)[name = tensor("dx2_77_cast_fp16")]; + tensor var_1759_cast_fp16 = mul(x = weights_157_cast_fp16, y = weights_157_cast_fp16)[name = tensor("op_1759_cast_fp16")]; tensor v2_77_axes_0 = const()[name = tensor("v2_77_axes_0"), val = tensor([2])]; tensor v2_77_keep_dims_0 = const()[name = tensor("v2_77_keep_dims_0"), val = tensor(false)]; tensor v2_77_cast_fp16 = reduce_sum(axes = v2_77_axes_0, keep_dims = v2_77_keep_dims_0, x = var_1759_cast_fp16)[name = tensor("v2_77_cast_fp16")]; - tensor var_1762_cast_fp16 = mul(x = dx2_77_cast_fp16, y = weights_157_cast_fp16)[name = tensor("op_1762_cast_fp16")]; + tensor var_1762_cast_fp16 = mul(x = dx2_77_cast_fp16, y = weights_157_cast_fp16)[name = tensor("op_1762_cast_fp16")]; tensor var_1764_axes_0 = const()[name = tensor("op_1764_axes_0"), val = tensor([2])]; tensor var_1764_keep_dims_0 = const()[name = tensor("op_1764_keep_dims_0"), val = tensor(false)]; tensor var_1764_cast_fp16 = reduce_sum(axes = var_1764_axes_0, keep_dims = var_1764_keep_dims_0, x = var_1762_cast_fp16)[name = tensor("op_1764_cast_fp16")]; @@ -1797,31 +1797,31 @@ program(1.0) tensor var_1772_interleave_0 = const()[name = tensor("op_1772_interleave_0"), val = tensor(false)]; tensor var_1772_cast_fp16 = concat(axis = var_72, interleave = var_1772_interleave_0, values = (mean_77_cast_fp16, std_77_cast_fp16))[name = tensor("op_1772_cast_fp16")]; tensor var_1774_begin_0 = const()[name = tensor("op_1774_begin_0"), val = tensor([0, 39, 0])]; - tensor var_1774_end_0 = const()[name = tensor("op_1774_end_0"), val = tensor([1, 40, 125])]; + tensor var_1774_end_0 = const()[name = tensor("op_1774_end_0"), val = tensor([1, 40, 375])]; tensor var_1774_end_mask_0 = const()[name = tensor("op_1774_end_mask_0"), val = tensor([true, false, true])]; tensor var_1774_squeeze_mask_0 = const()[name = tensor("op_1774_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1774_cast_fp16 = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, squeeze_mask = var_1774_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1774_cast_fp16")]; + tensor var_1774_cast_fp16 = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, squeeze_mask = var_1774_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1774_cast_fp16")]; tensor weights_161_axes_0 = const()[name = tensor("weights_161_axes_0"), val = tensor([1])]; - tensor weights_161_cast_fp16 = expand_dims(axes = weights_161_axes_0, x = var_1774_cast_fp16)[name = tensor("weights_161_cast_fp16")]; + tensor weights_161_cast_fp16 = expand_dims(axes = weights_161_axes_0, x = var_1774_cast_fp16)[name = tensor("weights_161_cast_fp16")]; tensor var_1778_axes_0 = const()[name = tensor("op_1778_axes_0"), val = tensor([2])]; tensor var_1778_keep_dims_0 = const()[name = tensor("op_1778_keep_dims_0"), val = tensor(false)]; tensor var_1778_cast_fp16 = reduce_sum(axes = var_1778_axes_0, keep_dims = var_1778_keep_dims_0, x = weights_161_cast_fp16)[name = tensor("op_1778_cast_fp16")]; tensor var_1779_to_fp16 = const()[name = tensor("op_1779_to_fp16"), val = tensor(0x1p-24)]; tensor v1_79_cast_fp16 = add(x = var_1778_cast_fp16, y = var_1779_to_fp16)[name = tensor("v1_79_cast_fp16")]; - tensor var_1781_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_161_cast_fp16)[name = tensor("op_1781_cast_fp16")]; + tensor var_1781_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_161_cast_fp16)[name = tensor("op_1781_cast_fp16")]; tensor var_1783_axes_0 = const()[name = tensor("op_1783_axes_0"), val = tensor([2])]; tensor var_1783_keep_dims_0 = const()[name = tensor("op_1783_keep_dims_0"), val = tensor(false)]; tensor var_1783_cast_fp16 = reduce_sum(axes = var_1783_axes_0, keep_dims = var_1783_keep_dims_0, x = var_1781_cast_fp16)[name = tensor("op_1783_cast_fp16")]; tensor mean_79_cast_fp16 = real_div(x = var_1783_cast_fp16, y = v1_79_cast_fp16)[name = tensor("mean_79_cast_fp16")]; tensor var_1785_axes_0 = const()[name = tensor("op_1785_axes_0"), val = tensor([2])]; tensor var_1785_cast_fp16 = expand_dims(axes = var_1785_axes_0, x = mean_79_cast_fp16)[name = tensor("op_1785_cast_fp16")]; - tensor var_1786_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1785_cast_fp16)[name = tensor("op_1786_cast_fp16")]; - tensor dx2_79_cast_fp16 = mul(x = var_1786_cast_fp16, y = var_1786_cast_fp16)[name = tensor("dx2_79_cast_fp16")]; - tensor var_1788_cast_fp16 = mul(x = weights_161_cast_fp16, y = weights_161_cast_fp16)[name = tensor("op_1788_cast_fp16")]; + tensor var_1786_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1785_cast_fp16)[name = tensor("op_1786_cast_fp16")]; + tensor dx2_79_cast_fp16 = mul(x = var_1786_cast_fp16, y = var_1786_cast_fp16)[name = tensor("dx2_79_cast_fp16")]; + tensor var_1788_cast_fp16 = mul(x = weights_161_cast_fp16, y = weights_161_cast_fp16)[name = tensor("op_1788_cast_fp16")]; tensor v2_79_axes_0 = const()[name = tensor("v2_79_axes_0"), val = tensor([2])]; tensor v2_79_keep_dims_0 = const()[name = tensor("v2_79_keep_dims_0"), val = tensor(false)]; tensor v2_79_cast_fp16 = reduce_sum(axes = v2_79_axes_0, keep_dims = v2_79_keep_dims_0, x = var_1788_cast_fp16)[name = tensor("v2_79_cast_fp16")]; - tensor var_1791_cast_fp16 = mul(x = dx2_79_cast_fp16, y = weights_161_cast_fp16)[name = tensor("op_1791_cast_fp16")]; + tensor var_1791_cast_fp16 = mul(x = dx2_79_cast_fp16, y = weights_161_cast_fp16)[name = tensor("op_1791_cast_fp16")]; tensor var_1793_axes_0 = const()[name = tensor("op_1793_axes_0"), val = tensor([2])]; tensor var_1793_keep_dims_0 = const()[name = tensor("op_1793_keep_dims_0"), val = tensor(false)]; tensor var_1793_cast_fp16 = reduce_sum(axes = var_1793_axes_0, keep_dims = var_1793_keep_dims_0, x = var_1791_cast_fp16)[name = tensor("op_1793_cast_fp16")]; @@ -1834,31 +1834,31 @@ program(1.0) tensor var_1801_interleave_0 = const()[name = tensor("op_1801_interleave_0"), val = tensor(false)]; tensor var_1801_cast_fp16 = concat(axis = var_72, interleave = var_1801_interleave_0, values = (mean_79_cast_fp16, std_79_cast_fp16))[name = tensor("op_1801_cast_fp16")]; tensor var_1803_begin_0 = const()[name = tensor("op_1803_begin_0"), val = tensor([0, 40, 0])]; - tensor var_1803_end_0 = const()[name = tensor("op_1803_end_0"), val = tensor([1, 41, 125])]; + tensor var_1803_end_0 = const()[name = tensor("op_1803_end_0"), val = tensor([1, 41, 375])]; tensor var_1803_end_mask_0 = const()[name = tensor("op_1803_end_mask_0"), val = tensor([true, false, true])]; tensor var_1803_squeeze_mask_0 = const()[name = tensor("op_1803_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1803_cast_fp16 = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, squeeze_mask = var_1803_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1803_cast_fp16")]; + tensor var_1803_cast_fp16 = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, squeeze_mask = var_1803_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1803_cast_fp16")]; tensor weights_165_axes_0 = const()[name = tensor("weights_165_axes_0"), val = tensor([1])]; - tensor weights_165_cast_fp16 = expand_dims(axes = weights_165_axes_0, x = var_1803_cast_fp16)[name = tensor("weights_165_cast_fp16")]; + tensor weights_165_cast_fp16 = expand_dims(axes = weights_165_axes_0, x = var_1803_cast_fp16)[name = tensor("weights_165_cast_fp16")]; tensor var_1807_axes_0 = const()[name = tensor("op_1807_axes_0"), val = tensor([2])]; tensor var_1807_keep_dims_0 = const()[name = tensor("op_1807_keep_dims_0"), val = tensor(false)]; tensor var_1807_cast_fp16 = reduce_sum(axes = var_1807_axes_0, keep_dims = var_1807_keep_dims_0, x = weights_165_cast_fp16)[name = tensor("op_1807_cast_fp16")]; tensor var_1808_to_fp16 = const()[name = tensor("op_1808_to_fp16"), val = tensor(0x1p-24)]; tensor v1_81_cast_fp16 = add(x = var_1807_cast_fp16, y = var_1808_to_fp16)[name = tensor("v1_81_cast_fp16")]; - tensor var_1810_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_165_cast_fp16)[name = tensor("op_1810_cast_fp16")]; + tensor var_1810_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_165_cast_fp16)[name = tensor("op_1810_cast_fp16")]; tensor var_1812_axes_0 = const()[name = tensor("op_1812_axes_0"), val = tensor([2])]; tensor var_1812_keep_dims_0 = const()[name = tensor("op_1812_keep_dims_0"), val = tensor(false)]; tensor var_1812_cast_fp16 = reduce_sum(axes = var_1812_axes_0, keep_dims = var_1812_keep_dims_0, x = var_1810_cast_fp16)[name = tensor("op_1812_cast_fp16")]; tensor mean_81_cast_fp16 = real_div(x = var_1812_cast_fp16, y = v1_81_cast_fp16)[name = tensor("mean_81_cast_fp16")]; tensor var_1814_axes_0 = const()[name = tensor("op_1814_axes_0"), val = tensor([2])]; tensor var_1814_cast_fp16 = expand_dims(axes = var_1814_axes_0, x = mean_81_cast_fp16)[name = tensor("op_1814_cast_fp16")]; - tensor var_1815_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1814_cast_fp16)[name = tensor("op_1815_cast_fp16")]; - tensor dx2_81_cast_fp16 = mul(x = var_1815_cast_fp16, y = var_1815_cast_fp16)[name = tensor("dx2_81_cast_fp16")]; - tensor var_1817_cast_fp16 = mul(x = weights_165_cast_fp16, y = weights_165_cast_fp16)[name = tensor("op_1817_cast_fp16")]; + tensor var_1815_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1814_cast_fp16)[name = tensor("op_1815_cast_fp16")]; + tensor dx2_81_cast_fp16 = mul(x = var_1815_cast_fp16, y = var_1815_cast_fp16)[name = tensor("dx2_81_cast_fp16")]; + tensor var_1817_cast_fp16 = mul(x = weights_165_cast_fp16, y = weights_165_cast_fp16)[name = tensor("op_1817_cast_fp16")]; tensor v2_81_axes_0 = const()[name = tensor("v2_81_axes_0"), val = tensor([2])]; tensor v2_81_keep_dims_0 = const()[name = tensor("v2_81_keep_dims_0"), val = tensor(false)]; tensor v2_81_cast_fp16 = reduce_sum(axes = v2_81_axes_0, keep_dims = v2_81_keep_dims_0, x = var_1817_cast_fp16)[name = tensor("v2_81_cast_fp16")]; - tensor var_1820_cast_fp16 = mul(x = dx2_81_cast_fp16, y = weights_165_cast_fp16)[name = tensor("op_1820_cast_fp16")]; + tensor var_1820_cast_fp16 = mul(x = dx2_81_cast_fp16, y = weights_165_cast_fp16)[name = tensor("op_1820_cast_fp16")]; tensor var_1822_axes_0 = const()[name = tensor("op_1822_axes_0"), val = tensor([2])]; tensor var_1822_keep_dims_0 = const()[name = tensor("op_1822_keep_dims_0"), val = tensor(false)]; tensor var_1822_cast_fp16 = reduce_sum(axes = var_1822_axes_0, keep_dims = var_1822_keep_dims_0, x = var_1820_cast_fp16)[name = tensor("op_1822_cast_fp16")]; @@ -1871,31 +1871,31 @@ program(1.0) tensor var_1830_interleave_0 = const()[name = tensor("op_1830_interleave_0"), val = tensor(false)]; tensor var_1830_cast_fp16 = concat(axis = var_72, interleave = var_1830_interleave_0, values = (mean_81_cast_fp16, std_81_cast_fp16))[name = tensor("op_1830_cast_fp16")]; tensor var_1832_begin_0 = const()[name = tensor("op_1832_begin_0"), val = tensor([0, 41, 0])]; - tensor var_1832_end_0 = const()[name = tensor("op_1832_end_0"), val = tensor([1, 42, 125])]; + tensor var_1832_end_0 = const()[name = tensor("op_1832_end_0"), val = tensor([1, 42, 375])]; tensor var_1832_end_mask_0 = const()[name = tensor("op_1832_end_mask_0"), val = tensor([true, false, true])]; tensor var_1832_squeeze_mask_0 = const()[name = tensor("op_1832_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1832_cast_fp16 = slice_by_index(begin = var_1832_begin_0, end = var_1832_end_0, end_mask = var_1832_end_mask_0, squeeze_mask = var_1832_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1832_cast_fp16")]; + tensor var_1832_cast_fp16 = slice_by_index(begin = var_1832_begin_0, end = var_1832_end_0, end_mask = var_1832_end_mask_0, squeeze_mask = var_1832_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1832_cast_fp16")]; tensor weights_169_axes_0 = const()[name = tensor("weights_169_axes_0"), val = tensor([1])]; - tensor weights_169_cast_fp16 = expand_dims(axes = weights_169_axes_0, x = var_1832_cast_fp16)[name = tensor("weights_169_cast_fp16")]; + tensor weights_169_cast_fp16 = expand_dims(axes = weights_169_axes_0, x = var_1832_cast_fp16)[name = tensor("weights_169_cast_fp16")]; tensor var_1836_axes_0 = const()[name = tensor("op_1836_axes_0"), val = tensor([2])]; tensor var_1836_keep_dims_0 = const()[name = tensor("op_1836_keep_dims_0"), val = tensor(false)]; tensor var_1836_cast_fp16 = reduce_sum(axes = var_1836_axes_0, keep_dims = var_1836_keep_dims_0, x = weights_169_cast_fp16)[name = tensor("op_1836_cast_fp16")]; tensor var_1837_to_fp16 = const()[name = tensor("op_1837_to_fp16"), val = tensor(0x1p-24)]; tensor v1_83_cast_fp16 = add(x = var_1836_cast_fp16, y = var_1837_to_fp16)[name = tensor("v1_83_cast_fp16")]; - tensor var_1839_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_169_cast_fp16)[name = tensor("op_1839_cast_fp16")]; + tensor var_1839_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_169_cast_fp16)[name = tensor("op_1839_cast_fp16")]; tensor var_1841_axes_0 = const()[name = tensor("op_1841_axes_0"), val = tensor([2])]; tensor var_1841_keep_dims_0 = const()[name = tensor("op_1841_keep_dims_0"), val = tensor(false)]; tensor var_1841_cast_fp16 = reduce_sum(axes = var_1841_axes_0, keep_dims = var_1841_keep_dims_0, x = var_1839_cast_fp16)[name = tensor("op_1841_cast_fp16")]; tensor mean_83_cast_fp16 = real_div(x = var_1841_cast_fp16, y = v1_83_cast_fp16)[name = tensor("mean_83_cast_fp16")]; tensor var_1843_axes_0 = const()[name = tensor("op_1843_axes_0"), val = tensor([2])]; tensor var_1843_cast_fp16 = expand_dims(axes = var_1843_axes_0, x = mean_83_cast_fp16)[name = tensor("op_1843_cast_fp16")]; - tensor var_1844_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1843_cast_fp16)[name = tensor("op_1844_cast_fp16")]; - tensor dx2_83_cast_fp16 = mul(x = var_1844_cast_fp16, y = var_1844_cast_fp16)[name = tensor("dx2_83_cast_fp16")]; - tensor var_1846_cast_fp16 = mul(x = weights_169_cast_fp16, y = weights_169_cast_fp16)[name = tensor("op_1846_cast_fp16")]; + tensor var_1844_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1843_cast_fp16)[name = tensor("op_1844_cast_fp16")]; + tensor dx2_83_cast_fp16 = mul(x = var_1844_cast_fp16, y = var_1844_cast_fp16)[name = tensor("dx2_83_cast_fp16")]; + tensor var_1846_cast_fp16 = mul(x = weights_169_cast_fp16, y = weights_169_cast_fp16)[name = tensor("op_1846_cast_fp16")]; tensor v2_83_axes_0 = const()[name = tensor("v2_83_axes_0"), val = tensor([2])]; tensor v2_83_keep_dims_0 = const()[name = tensor("v2_83_keep_dims_0"), val = tensor(false)]; tensor v2_83_cast_fp16 = reduce_sum(axes = v2_83_axes_0, keep_dims = v2_83_keep_dims_0, x = var_1846_cast_fp16)[name = tensor("v2_83_cast_fp16")]; - tensor var_1849_cast_fp16 = mul(x = dx2_83_cast_fp16, y = weights_169_cast_fp16)[name = tensor("op_1849_cast_fp16")]; + tensor var_1849_cast_fp16 = mul(x = dx2_83_cast_fp16, y = weights_169_cast_fp16)[name = tensor("op_1849_cast_fp16")]; tensor var_1851_axes_0 = const()[name = tensor("op_1851_axes_0"), val = tensor([2])]; tensor var_1851_keep_dims_0 = const()[name = tensor("op_1851_keep_dims_0"), val = tensor(false)]; tensor var_1851_cast_fp16 = reduce_sum(axes = var_1851_axes_0, keep_dims = var_1851_keep_dims_0, x = var_1849_cast_fp16)[name = tensor("op_1851_cast_fp16")]; @@ -1908,31 +1908,31 @@ program(1.0) tensor var_1859_interleave_0 = const()[name = tensor("op_1859_interleave_0"), val = tensor(false)]; tensor var_1859_cast_fp16 = concat(axis = var_72, interleave = var_1859_interleave_0, values = (mean_83_cast_fp16, std_83_cast_fp16))[name = tensor("op_1859_cast_fp16")]; tensor var_1861_begin_0 = const()[name = tensor("op_1861_begin_0"), val = tensor([0, 42, 0])]; - tensor var_1861_end_0 = const()[name = tensor("op_1861_end_0"), val = tensor([1, 43, 125])]; + tensor var_1861_end_0 = const()[name = tensor("op_1861_end_0"), val = tensor([1, 43, 375])]; tensor var_1861_end_mask_0 = const()[name = tensor("op_1861_end_mask_0"), val = tensor([true, false, true])]; tensor var_1861_squeeze_mask_0 = const()[name = tensor("op_1861_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, squeeze_mask = var_1861_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1861_cast_fp16")]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, squeeze_mask = var_1861_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1861_cast_fp16")]; tensor weights_173_axes_0 = const()[name = tensor("weights_173_axes_0"), val = tensor([1])]; - tensor weights_173_cast_fp16 = expand_dims(axes = weights_173_axes_0, x = var_1861_cast_fp16)[name = tensor("weights_173_cast_fp16")]; + tensor weights_173_cast_fp16 = expand_dims(axes = weights_173_axes_0, x = var_1861_cast_fp16)[name = tensor("weights_173_cast_fp16")]; tensor var_1865_axes_0 = const()[name = tensor("op_1865_axes_0"), val = tensor([2])]; tensor var_1865_keep_dims_0 = const()[name = tensor("op_1865_keep_dims_0"), val = tensor(false)]; tensor var_1865_cast_fp16 = reduce_sum(axes = var_1865_axes_0, keep_dims = var_1865_keep_dims_0, x = weights_173_cast_fp16)[name = tensor("op_1865_cast_fp16")]; tensor var_1866_to_fp16 = const()[name = tensor("op_1866_to_fp16"), val = tensor(0x1p-24)]; tensor v1_85_cast_fp16 = add(x = var_1865_cast_fp16, y = var_1866_to_fp16)[name = tensor("v1_85_cast_fp16")]; - tensor var_1868_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_173_cast_fp16)[name = tensor("op_1868_cast_fp16")]; + tensor var_1868_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_173_cast_fp16)[name = tensor("op_1868_cast_fp16")]; tensor var_1870_axes_0 = const()[name = tensor("op_1870_axes_0"), val = tensor([2])]; tensor var_1870_keep_dims_0 = const()[name = tensor("op_1870_keep_dims_0"), val = tensor(false)]; tensor var_1870_cast_fp16 = reduce_sum(axes = var_1870_axes_0, keep_dims = var_1870_keep_dims_0, x = var_1868_cast_fp16)[name = tensor("op_1870_cast_fp16")]; tensor mean_85_cast_fp16 = real_div(x = var_1870_cast_fp16, y = v1_85_cast_fp16)[name = tensor("mean_85_cast_fp16")]; tensor var_1872_axes_0 = const()[name = tensor("op_1872_axes_0"), val = tensor([2])]; tensor var_1872_cast_fp16 = expand_dims(axes = var_1872_axes_0, x = mean_85_cast_fp16)[name = tensor("op_1872_cast_fp16")]; - tensor var_1873_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1872_cast_fp16)[name = tensor("op_1873_cast_fp16")]; - tensor dx2_85_cast_fp16 = mul(x = var_1873_cast_fp16, y = var_1873_cast_fp16)[name = tensor("dx2_85_cast_fp16")]; - tensor var_1875_cast_fp16 = mul(x = weights_173_cast_fp16, y = weights_173_cast_fp16)[name = tensor("op_1875_cast_fp16")]; + tensor var_1873_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1872_cast_fp16)[name = tensor("op_1873_cast_fp16")]; + tensor dx2_85_cast_fp16 = mul(x = var_1873_cast_fp16, y = var_1873_cast_fp16)[name = tensor("dx2_85_cast_fp16")]; + tensor var_1875_cast_fp16 = mul(x = weights_173_cast_fp16, y = weights_173_cast_fp16)[name = tensor("op_1875_cast_fp16")]; tensor v2_85_axes_0 = const()[name = tensor("v2_85_axes_0"), val = tensor([2])]; tensor v2_85_keep_dims_0 = const()[name = tensor("v2_85_keep_dims_0"), val = tensor(false)]; tensor v2_85_cast_fp16 = reduce_sum(axes = v2_85_axes_0, keep_dims = v2_85_keep_dims_0, x = var_1875_cast_fp16)[name = tensor("v2_85_cast_fp16")]; - tensor var_1878_cast_fp16 = mul(x = dx2_85_cast_fp16, y = weights_173_cast_fp16)[name = tensor("op_1878_cast_fp16")]; + tensor var_1878_cast_fp16 = mul(x = dx2_85_cast_fp16, y = weights_173_cast_fp16)[name = tensor("op_1878_cast_fp16")]; tensor var_1880_axes_0 = const()[name = tensor("op_1880_axes_0"), val = tensor([2])]; tensor var_1880_keep_dims_0 = const()[name = tensor("op_1880_keep_dims_0"), val = tensor(false)]; tensor var_1880_cast_fp16 = reduce_sum(axes = var_1880_axes_0, keep_dims = var_1880_keep_dims_0, x = var_1878_cast_fp16)[name = tensor("op_1880_cast_fp16")]; @@ -1945,31 +1945,31 @@ program(1.0) tensor var_1888_interleave_0 = const()[name = tensor("op_1888_interleave_0"), val = tensor(false)]; tensor var_1888_cast_fp16 = concat(axis = var_72, interleave = var_1888_interleave_0, values = (mean_85_cast_fp16, std_85_cast_fp16))[name = tensor("op_1888_cast_fp16")]; tensor var_1890_begin_0 = const()[name = tensor("op_1890_begin_0"), val = tensor([0, 43, 0])]; - tensor var_1890_end_0 = const()[name = tensor("op_1890_end_0"), val = tensor([1, 44, 125])]; + tensor var_1890_end_0 = const()[name = tensor("op_1890_end_0"), val = tensor([1, 44, 375])]; tensor var_1890_end_mask_0 = const()[name = tensor("op_1890_end_mask_0"), val = tensor([true, false, true])]; tensor var_1890_squeeze_mask_0 = const()[name = tensor("op_1890_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1890_cast_fp16 = slice_by_index(begin = var_1890_begin_0, end = var_1890_end_0, end_mask = var_1890_end_mask_0, squeeze_mask = var_1890_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1890_cast_fp16")]; + tensor var_1890_cast_fp16 = slice_by_index(begin = var_1890_begin_0, end = var_1890_end_0, end_mask = var_1890_end_mask_0, squeeze_mask = var_1890_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1890_cast_fp16")]; tensor weights_177_axes_0 = const()[name = tensor("weights_177_axes_0"), val = tensor([1])]; - tensor weights_177_cast_fp16 = expand_dims(axes = weights_177_axes_0, x = var_1890_cast_fp16)[name = tensor("weights_177_cast_fp16")]; + tensor weights_177_cast_fp16 = expand_dims(axes = weights_177_axes_0, x = var_1890_cast_fp16)[name = tensor("weights_177_cast_fp16")]; tensor var_1894_axes_0 = const()[name = tensor("op_1894_axes_0"), val = tensor([2])]; tensor var_1894_keep_dims_0 = const()[name = tensor("op_1894_keep_dims_0"), val = tensor(false)]; tensor var_1894_cast_fp16 = reduce_sum(axes = var_1894_axes_0, keep_dims = var_1894_keep_dims_0, x = weights_177_cast_fp16)[name = tensor("op_1894_cast_fp16")]; tensor var_1895_to_fp16 = const()[name = tensor("op_1895_to_fp16"), val = tensor(0x1p-24)]; tensor v1_87_cast_fp16 = add(x = var_1894_cast_fp16, y = var_1895_to_fp16)[name = tensor("v1_87_cast_fp16")]; - tensor var_1897_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_177_cast_fp16)[name = tensor("op_1897_cast_fp16")]; + tensor var_1897_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_177_cast_fp16)[name = tensor("op_1897_cast_fp16")]; tensor var_1899_axes_0 = const()[name = tensor("op_1899_axes_0"), val = tensor([2])]; tensor var_1899_keep_dims_0 = const()[name = tensor("op_1899_keep_dims_0"), val = tensor(false)]; tensor var_1899_cast_fp16 = reduce_sum(axes = var_1899_axes_0, keep_dims = var_1899_keep_dims_0, x = var_1897_cast_fp16)[name = tensor("op_1899_cast_fp16")]; tensor mean_87_cast_fp16 = real_div(x = var_1899_cast_fp16, y = v1_87_cast_fp16)[name = tensor("mean_87_cast_fp16")]; tensor var_1901_axes_0 = const()[name = tensor("op_1901_axes_0"), val = tensor([2])]; tensor var_1901_cast_fp16 = expand_dims(axes = var_1901_axes_0, x = mean_87_cast_fp16)[name = tensor("op_1901_cast_fp16")]; - tensor var_1902_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1901_cast_fp16)[name = tensor("op_1902_cast_fp16")]; - tensor dx2_87_cast_fp16 = mul(x = var_1902_cast_fp16, y = var_1902_cast_fp16)[name = tensor("dx2_87_cast_fp16")]; - tensor var_1904_cast_fp16 = mul(x = weights_177_cast_fp16, y = weights_177_cast_fp16)[name = tensor("op_1904_cast_fp16")]; + tensor var_1902_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1901_cast_fp16)[name = tensor("op_1902_cast_fp16")]; + tensor dx2_87_cast_fp16 = mul(x = var_1902_cast_fp16, y = var_1902_cast_fp16)[name = tensor("dx2_87_cast_fp16")]; + tensor var_1904_cast_fp16 = mul(x = weights_177_cast_fp16, y = weights_177_cast_fp16)[name = tensor("op_1904_cast_fp16")]; tensor v2_87_axes_0 = const()[name = tensor("v2_87_axes_0"), val = tensor([2])]; tensor v2_87_keep_dims_0 = const()[name = tensor("v2_87_keep_dims_0"), val = tensor(false)]; tensor v2_87_cast_fp16 = reduce_sum(axes = v2_87_axes_0, keep_dims = v2_87_keep_dims_0, x = var_1904_cast_fp16)[name = tensor("v2_87_cast_fp16")]; - tensor var_1907_cast_fp16 = mul(x = dx2_87_cast_fp16, y = weights_177_cast_fp16)[name = tensor("op_1907_cast_fp16")]; + tensor var_1907_cast_fp16 = mul(x = dx2_87_cast_fp16, y = weights_177_cast_fp16)[name = tensor("op_1907_cast_fp16")]; tensor var_1909_axes_0 = const()[name = tensor("op_1909_axes_0"), val = tensor([2])]; tensor var_1909_keep_dims_0 = const()[name = tensor("op_1909_keep_dims_0"), val = tensor(false)]; tensor var_1909_cast_fp16 = reduce_sum(axes = var_1909_axes_0, keep_dims = var_1909_keep_dims_0, x = var_1907_cast_fp16)[name = tensor("op_1909_cast_fp16")]; @@ -1982,31 +1982,31 @@ program(1.0) tensor var_1917_interleave_0 = const()[name = tensor("op_1917_interleave_0"), val = tensor(false)]; tensor var_1917_cast_fp16 = concat(axis = var_72, interleave = var_1917_interleave_0, values = (mean_87_cast_fp16, std_87_cast_fp16))[name = tensor("op_1917_cast_fp16")]; tensor var_1919_begin_0 = const()[name = tensor("op_1919_begin_0"), val = tensor([0, 44, 0])]; - tensor var_1919_end_0 = const()[name = tensor("op_1919_end_0"), val = tensor([1, 45, 125])]; + tensor var_1919_end_0 = const()[name = tensor("op_1919_end_0"), val = tensor([1, 45, 375])]; tensor var_1919_end_mask_0 = const()[name = tensor("op_1919_end_mask_0"), val = tensor([true, false, true])]; tensor var_1919_squeeze_mask_0 = const()[name = tensor("op_1919_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1919_cast_fp16 = slice_by_index(begin = var_1919_begin_0, end = var_1919_end_0, end_mask = var_1919_end_mask_0, squeeze_mask = var_1919_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1919_cast_fp16")]; + tensor var_1919_cast_fp16 = slice_by_index(begin = var_1919_begin_0, end = var_1919_end_0, end_mask = var_1919_end_mask_0, squeeze_mask = var_1919_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1919_cast_fp16")]; tensor weights_181_axes_0 = const()[name = tensor("weights_181_axes_0"), val = tensor([1])]; - tensor weights_181_cast_fp16 = expand_dims(axes = weights_181_axes_0, x = var_1919_cast_fp16)[name = tensor("weights_181_cast_fp16")]; + tensor weights_181_cast_fp16 = expand_dims(axes = weights_181_axes_0, x = var_1919_cast_fp16)[name = tensor("weights_181_cast_fp16")]; tensor var_1923_axes_0 = const()[name = tensor("op_1923_axes_0"), val = tensor([2])]; tensor var_1923_keep_dims_0 = const()[name = tensor("op_1923_keep_dims_0"), val = tensor(false)]; tensor var_1923_cast_fp16 = reduce_sum(axes = var_1923_axes_0, keep_dims = var_1923_keep_dims_0, x = weights_181_cast_fp16)[name = tensor("op_1923_cast_fp16")]; tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(0x1p-24)]; tensor v1_89_cast_fp16 = add(x = var_1923_cast_fp16, y = var_1924_to_fp16)[name = tensor("v1_89_cast_fp16")]; - tensor var_1926_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_181_cast_fp16)[name = tensor("op_1926_cast_fp16")]; + tensor var_1926_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_181_cast_fp16)[name = tensor("op_1926_cast_fp16")]; tensor var_1928_axes_0 = const()[name = tensor("op_1928_axes_0"), val = tensor([2])]; tensor var_1928_keep_dims_0 = const()[name = tensor("op_1928_keep_dims_0"), val = tensor(false)]; tensor var_1928_cast_fp16 = reduce_sum(axes = var_1928_axes_0, keep_dims = var_1928_keep_dims_0, x = var_1926_cast_fp16)[name = tensor("op_1928_cast_fp16")]; tensor mean_89_cast_fp16 = real_div(x = var_1928_cast_fp16, y = v1_89_cast_fp16)[name = tensor("mean_89_cast_fp16")]; tensor var_1930_axes_0 = const()[name = tensor("op_1930_axes_0"), val = tensor([2])]; tensor var_1930_cast_fp16 = expand_dims(axes = var_1930_axes_0, x = mean_89_cast_fp16)[name = tensor("op_1930_cast_fp16")]; - tensor var_1931_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1930_cast_fp16)[name = tensor("op_1931_cast_fp16")]; - tensor dx2_89_cast_fp16 = mul(x = var_1931_cast_fp16, y = var_1931_cast_fp16)[name = tensor("dx2_89_cast_fp16")]; - tensor var_1933_cast_fp16 = mul(x = weights_181_cast_fp16, y = weights_181_cast_fp16)[name = tensor("op_1933_cast_fp16")]; + tensor var_1931_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1930_cast_fp16)[name = tensor("op_1931_cast_fp16")]; + tensor dx2_89_cast_fp16 = mul(x = var_1931_cast_fp16, y = var_1931_cast_fp16)[name = tensor("dx2_89_cast_fp16")]; + tensor var_1933_cast_fp16 = mul(x = weights_181_cast_fp16, y = weights_181_cast_fp16)[name = tensor("op_1933_cast_fp16")]; tensor v2_89_axes_0 = const()[name = tensor("v2_89_axes_0"), val = tensor([2])]; tensor v2_89_keep_dims_0 = const()[name = tensor("v2_89_keep_dims_0"), val = tensor(false)]; tensor v2_89_cast_fp16 = reduce_sum(axes = v2_89_axes_0, keep_dims = v2_89_keep_dims_0, x = var_1933_cast_fp16)[name = tensor("v2_89_cast_fp16")]; - tensor var_1936_cast_fp16 = mul(x = dx2_89_cast_fp16, y = weights_181_cast_fp16)[name = tensor("op_1936_cast_fp16")]; + tensor var_1936_cast_fp16 = mul(x = dx2_89_cast_fp16, y = weights_181_cast_fp16)[name = tensor("op_1936_cast_fp16")]; tensor var_1938_axes_0 = const()[name = tensor("op_1938_axes_0"), val = tensor([2])]; tensor var_1938_keep_dims_0 = const()[name = tensor("op_1938_keep_dims_0"), val = tensor(false)]; tensor var_1938_cast_fp16 = reduce_sum(axes = var_1938_axes_0, keep_dims = var_1938_keep_dims_0, x = var_1936_cast_fp16)[name = tensor("op_1938_cast_fp16")]; @@ -2019,31 +2019,31 @@ program(1.0) tensor var_1946_interleave_0 = const()[name = tensor("op_1946_interleave_0"), val = tensor(false)]; tensor var_1946_cast_fp16 = concat(axis = var_72, interleave = var_1946_interleave_0, values = (mean_89_cast_fp16, std_89_cast_fp16))[name = tensor("op_1946_cast_fp16")]; tensor var_1948_begin_0 = const()[name = tensor("op_1948_begin_0"), val = tensor([0, 45, 0])]; - tensor var_1948_end_0 = const()[name = tensor("op_1948_end_0"), val = tensor([1, 46, 125])]; + tensor var_1948_end_0 = const()[name = tensor("op_1948_end_0"), val = tensor([1, 46, 375])]; tensor var_1948_end_mask_0 = const()[name = tensor("op_1948_end_mask_0"), val = tensor([true, false, true])]; tensor var_1948_squeeze_mask_0 = const()[name = tensor("op_1948_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1948_cast_fp16 = slice_by_index(begin = var_1948_begin_0, end = var_1948_end_0, end_mask = var_1948_end_mask_0, squeeze_mask = var_1948_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1948_cast_fp16")]; + tensor var_1948_cast_fp16 = slice_by_index(begin = var_1948_begin_0, end = var_1948_end_0, end_mask = var_1948_end_mask_0, squeeze_mask = var_1948_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1948_cast_fp16")]; tensor weights_185_axes_0 = const()[name = tensor("weights_185_axes_0"), val = tensor([1])]; - tensor weights_185_cast_fp16 = expand_dims(axes = weights_185_axes_0, x = var_1948_cast_fp16)[name = tensor("weights_185_cast_fp16")]; + tensor weights_185_cast_fp16 = expand_dims(axes = weights_185_axes_0, x = var_1948_cast_fp16)[name = tensor("weights_185_cast_fp16")]; tensor var_1952_axes_0 = const()[name = tensor("op_1952_axes_0"), val = tensor([2])]; tensor var_1952_keep_dims_0 = const()[name = tensor("op_1952_keep_dims_0"), val = tensor(false)]; tensor var_1952_cast_fp16 = reduce_sum(axes = var_1952_axes_0, keep_dims = var_1952_keep_dims_0, x = weights_185_cast_fp16)[name = tensor("op_1952_cast_fp16")]; tensor var_1953_to_fp16 = const()[name = tensor("op_1953_to_fp16"), val = tensor(0x1p-24)]; tensor v1_91_cast_fp16 = add(x = var_1952_cast_fp16, y = var_1953_to_fp16)[name = tensor("v1_91_cast_fp16")]; - tensor var_1955_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_185_cast_fp16)[name = tensor("op_1955_cast_fp16")]; + tensor var_1955_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_185_cast_fp16)[name = tensor("op_1955_cast_fp16")]; tensor var_1957_axes_0 = const()[name = tensor("op_1957_axes_0"), val = tensor([2])]; tensor var_1957_keep_dims_0 = const()[name = tensor("op_1957_keep_dims_0"), val = tensor(false)]; tensor var_1957_cast_fp16 = reduce_sum(axes = var_1957_axes_0, keep_dims = var_1957_keep_dims_0, x = var_1955_cast_fp16)[name = tensor("op_1957_cast_fp16")]; tensor mean_91_cast_fp16 = real_div(x = var_1957_cast_fp16, y = v1_91_cast_fp16)[name = tensor("mean_91_cast_fp16")]; tensor var_1959_axes_0 = const()[name = tensor("op_1959_axes_0"), val = tensor([2])]; tensor var_1959_cast_fp16 = expand_dims(axes = var_1959_axes_0, x = mean_91_cast_fp16)[name = tensor("op_1959_cast_fp16")]; - tensor var_1960_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1959_cast_fp16)[name = tensor("op_1960_cast_fp16")]; - tensor dx2_91_cast_fp16 = mul(x = var_1960_cast_fp16, y = var_1960_cast_fp16)[name = tensor("dx2_91_cast_fp16")]; - tensor var_1962_cast_fp16 = mul(x = weights_185_cast_fp16, y = weights_185_cast_fp16)[name = tensor("op_1962_cast_fp16")]; + tensor var_1960_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1959_cast_fp16)[name = tensor("op_1960_cast_fp16")]; + tensor dx2_91_cast_fp16 = mul(x = var_1960_cast_fp16, y = var_1960_cast_fp16)[name = tensor("dx2_91_cast_fp16")]; + tensor var_1962_cast_fp16 = mul(x = weights_185_cast_fp16, y = weights_185_cast_fp16)[name = tensor("op_1962_cast_fp16")]; tensor v2_91_axes_0 = const()[name = tensor("v2_91_axes_0"), val = tensor([2])]; tensor v2_91_keep_dims_0 = const()[name = tensor("v2_91_keep_dims_0"), val = tensor(false)]; tensor v2_91_cast_fp16 = reduce_sum(axes = v2_91_axes_0, keep_dims = v2_91_keep_dims_0, x = var_1962_cast_fp16)[name = tensor("v2_91_cast_fp16")]; - tensor var_1965_cast_fp16 = mul(x = dx2_91_cast_fp16, y = weights_185_cast_fp16)[name = tensor("op_1965_cast_fp16")]; + tensor var_1965_cast_fp16 = mul(x = dx2_91_cast_fp16, y = weights_185_cast_fp16)[name = tensor("op_1965_cast_fp16")]; tensor var_1967_axes_0 = const()[name = tensor("op_1967_axes_0"), val = tensor([2])]; tensor var_1967_keep_dims_0 = const()[name = tensor("op_1967_keep_dims_0"), val = tensor(false)]; tensor var_1967_cast_fp16 = reduce_sum(axes = var_1967_axes_0, keep_dims = var_1967_keep_dims_0, x = var_1965_cast_fp16)[name = tensor("op_1967_cast_fp16")]; @@ -2056,31 +2056,31 @@ program(1.0) tensor var_1975_interleave_0 = const()[name = tensor("op_1975_interleave_0"), val = tensor(false)]; tensor var_1975_cast_fp16 = concat(axis = var_72, interleave = var_1975_interleave_0, values = (mean_91_cast_fp16, std_91_cast_fp16))[name = tensor("op_1975_cast_fp16")]; tensor var_1977_begin_0 = const()[name = tensor("op_1977_begin_0"), val = tensor([0, 46, 0])]; - tensor var_1977_end_0 = const()[name = tensor("op_1977_end_0"), val = tensor([1, 47, 125])]; + tensor var_1977_end_0 = const()[name = tensor("op_1977_end_0"), val = tensor([1, 47, 375])]; tensor var_1977_end_mask_0 = const()[name = tensor("op_1977_end_mask_0"), val = tensor([true, false, true])]; tensor var_1977_squeeze_mask_0 = const()[name = tensor("op_1977_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_1977_cast_fp16 = slice_by_index(begin = var_1977_begin_0, end = var_1977_end_0, end_mask = var_1977_end_mask_0, squeeze_mask = var_1977_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1977_cast_fp16")]; + tensor var_1977_cast_fp16 = slice_by_index(begin = var_1977_begin_0, end = var_1977_end_0, end_mask = var_1977_end_mask_0, squeeze_mask = var_1977_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_1977_cast_fp16")]; tensor weights_189_axes_0 = const()[name = tensor("weights_189_axes_0"), val = tensor([1])]; - tensor weights_189_cast_fp16 = expand_dims(axes = weights_189_axes_0, x = var_1977_cast_fp16)[name = tensor("weights_189_cast_fp16")]; + tensor weights_189_cast_fp16 = expand_dims(axes = weights_189_axes_0, x = var_1977_cast_fp16)[name = tensor("weights_189_cast_fp16")]; tensor var_1981_axes_0 = const()[name = tensor("op_1981_axes_0"), val = tensor([2])]; tensor var_1981_keep_dims_0 = const()[name = tensor("op_1981_keep_dims_0"), val = tensor(false)]; tensor var_1981_cast_fp16 = reduce_sum(axes = var_1981_axes_0, keep_dims = var_1981_keep_dims_0, x = weights_189_cast_fp16)[name = tensor("op_1981_cast_fp16")]; tensor var_1982_to_fp16 = const()[name = tensor("op_1982_to_fp16"), val = tensor(0x1p-24)]; tensor v1_93_cast_fp16 = add(x = var_1981_cast_fp16, y = var_1982_to_fp16)[name = tensor("v1_93_cast_fp16")]; - tensor var_1984_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_189_cast_fp16)[name = tensor("op_1984_cast_fp16")]; + tensor var_1984_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_189_cast_fp16)[name = tensor("op_1984_cast_fp16")]; tensor var_1986_axes_0 = const()[name = tensor("op_1986_axes_0"), val = tensor([2])]; tensor var_1986_keep_dims_0 = const()[name = tensor("op_1986_keep_dims_0"), val = tensor(false)]; tensor var_1986_cast_fp16 = reduce_sum(axes = var_1986_axes_0, keep_dims = var_1986_keep_dims_0, x = var_1984_cast_fp16)[name = tensor("op_1986_cast_fp16")]; tensor mean_93_cast_fp16 = real_div(x = var_1986_cast_fp16, y = v1_93_cast_fp16)[name = tensor("mean_93_cast_fp16")]; tensor var_1988_axes_0 = const()[name = tensor("op_1988_axes_0"), val = tensor([2])]; tensor var_1988_cast_fp16 = expand_dims(axes = var_1988_axes_0, x = mean_93_cast_fp16)[name = tensor("op_1988_cast_fp16")]; - tensor var_1989_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1988_cast_fp16)[name = tensor("op_1989_cast_fp16")]; - tensor dx2_93_cast_fp16 = mul(x = var_1989_cast_fp16, y = var_1989_cast_fp16)[name = tensor("dx2_93_cast_fp16")]; - tensor var_1991_cast_fp16 = mul(x = weights_189_cast_fp16, y = weights_189_cast_fp16)[name = tensor("op_1991_cast_fp16")]; + tensor var_1989_cast_fp16 = sub(x = sequences_cast_fp16, y = var_1988_cast_fp16)[name = tensor("op_1989_cast_fp16")]; + tensor dx2_93_cast_fp16 = mul(x = var_1989_cast_fp16, y = var_1989_cast_fp16)[name = tensor("dx2_93_cast_fp16")]; + tensor var_1991_cast_fp16 = mul(x = weights_189_cast_fp16, y = weights_189_cast_fp16)[name = tensor("op_1991_cast_fp16")]; tensor v2_93_axes_0 = const()[name = tensor("v2_93_axes_0"), val = tensor([2])]; tensor v2_93_keep_dims_0 = const()[name = tensor("v2_93_keep_dims_0"), val = tensor(false)]; tensor v2_93_cast_fp16 = reduce_sum(axes = v2_93_axes_0, keep_dims = v2_93_keep_dims_0, x = var_1991_cast_fp16)[name = tensor("v2_93_cast_fp16")]; - tensor var_1994_cast_fp16 = mul(x = dx2_93_cast_fp16, y = weights_189_cast_fp16)[name = tensor("op_1994_cast_fp16")]; + tensor var_1994_cast_fp16 = mul(x = dx2_93_cast_fp16, y = weights_189_cast_fp16)[name = tensor("op_1994_cast_fp16")]; tensor var_1996_axes_0 = const()[name = tensor("op_1996_axes_0"), val = tensor([2])]; tensor var_1996_keep_dims_0 = const()[name = tensor("op_1996_keep_dims_0"), val = tensor(false)]; tensor var_1996_cast_fp16 = reduce_sum(axes = var_1996_axes_0, keep_dims = var_1996_keep_dims_0, x = var_1994_cast_fp16)[name = tensor("op_1996_cast_fp16")]; @@ -2093,31 +2093,31 @@ program(1.0) tensor var_2004_interleave_0 = const()[name = tensor("op_2004_interleave_0"), val = tensor(false)]; tensor var_2004_cast_fp16 = concat(axis = var_72, interleave = var_2004_interleave_0, values = (mean_93_cast_fp16, std_93_cast_fp16))[name = tensor("op_2004_cast_fp16")]; tensor var_2006_begin_0 = const()[name = tensor("op_2006_begin_0"), val = tensor([0, 47, 0])]; - tensor var_2006_end_0 = const()[name = tensor("op_2006_end_0"), val = tensor([1, 48, 125])]; + tensor var_2006_end_0 = const()[name = tensor("op_2006_end_0"), val = tensor([1, 48, 375])]; tensor var_2006_end_mask_0 = const()[name = tensor("op_2006_end_mask_0"), val = tensor([true, false, true])]; tensor var_2006_squeeze_mask_0 = const()[name = tensor("op_2006_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2006_cast_fp16 = slice_by_index(begin = var_2006_begin_0, end = var_2006_end_0, end_mask = var_2006_end_mask_0, squeeze_mask = var_2006_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2006_cast_fp16")]; + tensor var_2006_cast_fp16 = slice_by_index(begin = var_2006_begin_0, end = var_2006_end_0, end_mask = var_2006_end_mask_0, squeeze_mask = var_2006_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2006_cast_fp16")]; tensor weights_193_axes_0 = const()[name = tensor("weights_193_axes_0"), val = tensor([1])]; - tensor weights_193_cast_fp16 = expand_dims(axes = weights_193_axes_0, x = var_2006_cast_fp16)[name = tensor("weights_193_cast_fp16")]; + tensor weights_193_cast_fp16 = expand_dims(axes = weights_193_axes_0, x = var_2006_cast_fp16)[name = tensor("weights_193_cast_fp16")]; tensor var_2010_axes_0 = const()[name = tensor("op_2010_axes_0"), val = tensor([2])]; tensor var_2010_keep_dims_0 = const()[name = tensor("op_2010_keep_dims_0"), val = tensor(false)]; tensor var_2010_cast_fp16 = reduce_sum(axes = var_2010_axes_0, keep_dims = var_2010_keep_dims_0, x = weights_193_cast_fp16)[name = tensor("op_2010_cast_fp16")]; tensor var_2011_to_fp16 = const()[name = tensor("op_2011_to_fp16"), val = tensor(0x1p-24)]; tensor v1_95_cast_fp16 = add(x = var_2010_cast_fp16, y = var_2011_to_fp16)[name = tensor("v1_95_cast_fp16")]; - tensor var_2013_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_193_cast_fp16)[name = tensor("op_2013_cast_fp16")]; + tensor var_2013_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_193_cast_fp16)[name = tensor("op_2013_cast_fp16")]; tensor var_2015_axes_0 = const()[name = tensor("op_2015_axes_0"), val = tensor([2])]; tensor var_2015_keep_dims_0 = const()[name = tensor("op_2015_keep_dims_0"), val = tensor(false)]; tensor var_2015_cast_fp16 = reduce_sum(axes = var_2015_axes_0, keep_dims = var_2015_keep_dims_0, x = var_2013_cast_fp16)[name = tensor("op_2015_cast_fp16")]; tensor mean_95_cast_fp16 = real_div(x = var_2015_cast_fp16, y = v1_95_cast_fp16)[name = tensor("mean_95_cast_fp16")]; tensor var_2017_axes_0 = const()[name = tensor("op_2017_axes_0"), val = tensor([2])]; tensor var_2017_cast_fp16 = expand_dims(axes = var_2017_axes_0, x = mean_95_cast_fp16)[name = tensor("op_2017_cast_fp16")]; - tensor var_2018_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2017_cast_fp16)[name = tensor("op_2018_cast_fp16")]; - tensor dx2_95_cast_fp16 = mul(x = var_2018_cast_fp16, y = var_2018_cast_fp16)[name = tensor("dx2_95_cast_fp16")]; - tensor var_2020_cast_fp16 = mul(x = weights_193_cast_fp16, y = weights_193_cast_fp16)[name = tensor("op_2020_cast_fp16")]; + tensor var_2018_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2017_cast_fp16)[name = tensor("op_2018_cast_fp16")]; + tensor dx2_95_cast_fp16 = mul(x = var_2018_cast_fp16, y = var_2018_cast_fp16)[name = tensor("dx2_95_cast_fp16")]; + tensor var_2020_cast_fp16 = mul(x = weights_193_cast_fp16, y = weights_193_cast_fp16)[name = tensor("op_2020_cast_fp16")]; tensor v2_95_axes_0 = const()[name = tensor("v2_95_axes_0"), val = tensor([2])]; tensor v2_95_keep_dims_0 = const()[name = tensor("v2_95_keep_dims_0"), val = tensor(false)]; tensor v2_95_cast_fp16 = reduce_sum(axes = v2_95_axes_0, keep_dims = v2_95_keep_dims_0, x = var_2020_cast_fp16)[name = tensor("v2_95_cast_fp16")]; - tensor var_2023_cast_fp16 = mul(x = dx2_95_cast_fp16, y = weights_193_cast_fp16)[name = tensor("op_2023_cast_fp16")]; + tensor var_2023_cast_fp16 = mul(x = dx2_95_cast_fp16, y = weights_193_cast_fp16)[name = tensor("op_2023_cast_fp16")]; tensor var_2025_axes_0 = const()[name = tensor("op_2025_axes_0"), val = tensor([2])]; tensor var_2025_keep_dims_0 = const()[name = tensor("op_2025_keep_dims_0"), val = tensor(false)]; tensor var_2025_cast_fp16 = reduce_sum(axes = var_2025_axes_0, keep_dims = var_2025_keep_dims_0, x = var_2023_cast_fp16)[name = tensor("op_2025_cast_fp16")]; @@ -2130,31 +2130,31 @@ program(1.0) tensor var_2033_interleave_0 = const()[name = tensor("op_2033_interleave_0"), val = tensor(false)]; tensor var_2033_cast_fp16 = concat(axis = var_72, interleave = var_2033_interleave_0, values = (mean_95_cast_fp16, std_95_cast_fp16))[name = tensor("op_2033_cast_fp16")]; tensor var_2035_begin_0 = const()[name = tensor("op_2035_begin_0"), val = tensor([0, 48, 0])]; - tensor var_2035_end_0 = const()[name = tensor("op_2035_end_0"), val = tensor([1, 49, 125])]; + tensor var_2035_end_0 = const()[name = tensor("op_2035_end_0"), val = tensor([1, 49, 375])]; tensor var_2035_end_mask_0 = const()[name = tensor("op_2035_end_mask_0"), val = tensor([true, false, true])]; tensor var_2035_squeeze_mask_0 = const()[name = tensor("op_2035_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2035_cast_fp16 = slice_by_index(begin = var_2035_begin_0, end = var_2035_end_0, end_mask = var_2035_end_mask_0, squeeze_mask = var_2035_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2035_cast_fp16")]; + tensor var_2035_cast_fp16 = slice_by_index(begin = var_2035_begin_0, end = var_2035_end_0, end_mask = var_2035_end_mask_0, squeeze_mask = var_2035_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2035_cast_fp16")]; tensor weights_197_axes_0 = const()[name = tensor("weights_197_axes_0"), val = tensor([1])]; - tensor weights_197_cast_fp16 = expand_dims(axes = weights_197_axes_0, x = var_2035_cast_fp16)[name = tensor("weights_197_cast_fp16")]; + tensor weights_197_cast_fp16 = expand_dims(axes = weights_197_axes_0, x = var_2035_cast_fp16)[name = tensor("weights_197_cast_fp16")]; tensor var_2039_axes_0 = const()[name = tensor("op_2039_axes_0"), val = tensor([2])]; tensor var_2039_keep_dims_0 = const()[name = tensor("op_2039_keep_dims_0"), val = tensor(false)]; tensor var_2039_cast_fp16 = reduce_sum(axes = var_2039_axes_0, keep_dims = var_2039_keep_dims_0, x = weights_197_cast_fp16)[name = tensor("op_2039_cast_fp16")]; tensor var_2040_to_fp16 = const()[name = tensor("op_2040_to_fp16"), val = tensor(0x1p-24)]; tensor v1_97_cast_fp16 = add(x = var_2039_cast_fp16, y = var_2040_to_fp16)[name = tensor("v1_97_cast_fp16")]; - tensor var_2042_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_197_cast_fp16)[name = tensor("op_2042_cast_fp16")]; + tensor var_2042_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_197_cast_fp16)[name = tensor("op_2042_cast_fp16")]; tensor var_2044_axes_0 = const()[name = tensor("op_2044_axes_0"), val = tensor([2])]; tensor var_2044_keep_dims_0 = const()[name = tensor("op_2044_keep_dims_0"), val = tensor(false)]; tensor var_2044_cast_fp16 = reduce_sum(axes = var_2044_axes_0, keep_dims = var_2044_keep_dims_0, x = var_2042_cast_fp16)[name = tensor("op_2044_cast_fp16")]; tensor mean_97_cast_fp16 = real_div(x = var_2044_cast_fp16, y = v1_97_cast_fp16)[name = tensor("mean_97_cast_fp16")]; tensor var_2046_axes_0 = const()[name = tensor("op_2046_axes_0"), val = tensor([2])]; tensor var_2046_cast_fp16 = expand_dims(axes = var_2046_axes_0, x = mean_97_cast_fp16)[name = tensor("op_2046_cast_fp16")]; - tensor var_2047_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2046_cast_fp16)[name = tensor("op_2047_cast_fp16")]; - tensor dx2_97_cast_fp16 = mul(x = var_2047_cast_fp16, y = var_2047_cast_fp16)[name = tensor("dx2_97_cast_fp16")]; - tensor var_2049_cast_fp16 = mul(x = weights_197_cast_fp16, y = weights_197_cast_fp16)[name = tensor("op_2049_cast_fp16")]; + tensor var_2047_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2046_cast_fp16)[name = tensor("op_2047_cast_fp16")]; + tensor dx2_97_cast_fp16 = mul(x = var_2047_cast_fp16, y = var_2047_cast_fp16)[name = tensor("dx2_97_cast_fp16")]; + tensor var_2049_cast_fp16 = mul(x = weights_197_cast_fp16, y = weights_197_cast_fp16)[name = tensor("op_2049_cast_fp16")]; tensor v2_97_axes_0 = const()[name = tensor("v2_97_axes_0"), val = tensor([2])]; tensor v2_97_keep_dims_0 = const()[name = tensor("v2_97_keep_dims_0"), val = tensor(false)]; tensor v2_97_cast_fp16 = reduce_sum(axes = v2_97_axes_0, keep_dims = v2_97_keep_dims_0, x = var_2049_cast_fp16)[name = tensor("v2_97_cast_fp16")]; - tensor var_2052_cast_fp16 = mul(x = dx2_97_cast_fp16, y = weights_197_cast_fp16)[name = tensor("op_2052_cast_fp16")]; + tensor var_2052_cast_fp16 = mul(x = dx2_97_cast_fp16, y = weights_197_cast_fp16)[name = tensor("op_2052_cast_fp16")]; tensor var_2054_axes_0 = const()[name = tensor("op_2054_axes_0"), val = tensor([2])]; tensor var_2054_keep_dims_0 = const()[name = tensor("op_2054_keep_dims_0"), val = tensor(false)]; tensor var_2054_cast_fp16 = reduce_sum(axes = var_2054_axes_0, keep_dims = var_2054_keep_dims_0, x = var_2052_cast_fp16)[name = tensor("op_2054_cast_fp16")]; @@ -2167,31 +2167,31 @@ program(1.0) tensor var_2062_interleave_0 = const()[name = tensor("op_2062_interleave_0"), val = tensor(false)]; tensor var_2062_cast_fp16 = concat(axis = var_72, interleave = var_2062_interleave_0, values = (mean_97_cast_fp16, std_97_cast_fp16))[name = tensor("op_2062_cast_fp16")]; tensor var_2064_begin_0 = const()[name = tensor("op_2064_begin_0"), val = tensor([0, 49, 0])]; - tensor var_2064_end_0 = const()[name = tensor("op_2064_end_0"), val = tensor([1, 50, 125])]; + tensor var_2064_end_0 = const()[name = tensor("op_2064_end_0"), val = tensor([1, 50, 375])]; tensor var_2064_end_mask_0 = const()[name = tensor("op_2064_end_mask_0"), val = tensor([true, false, true])]; tensor var_2064_squeeze_mask_0 = const()[name = tensor("op_2064_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2064_cast_fp16 = slice_by_index(begin = var_2064_begin_0, end = var_2064_end_0, end_mask = var_2064_end_mask_0, squeeze_mask = var_2064_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2064_cast_fp16")]; + tensor var_2064_cast_fp16 = slice_by_index(begin = var_2064_begin_0, end = var_2064_end_0, end_mask = var_2064_end_mask_0, squeeze_mask = var_2064_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2064_cast_fp16")]; tensor weights_201_axes_0 = const()[name = tensor("weights_201_axes_0"), val = tensor([1])]; - tensor weights_201_cast_fp16 = expand_dims(axes = weights_201_axes_0, x = var_2064_cast_fp16)[name = tensor("weights_201_cast_fp16")]; + tensor weights_201_cast_fp16 = expand_dims(axes = weights_201_axes_0, x = var_2064_cast_fp16)[name = tensor("weights_201_cast_fp16")]; tensor var_2068_axes_0 = const()[name = tensor("op_2068_axes_0"), val = tensor([2])]; tensor var_2068_keep_dims_0 = const()[name = tensor("op_2068_keep_dims_0"), val = tensor(false)]; tensor var_2068_cast_fp16 = reduce_sum(axes = var_2068_axes_0, keep_dims = var_2068_keep_dims_0, x = weights_201_cast_fp16)[name = tensor("op_2068_cast_fp16")]; tensor var_2069_to_fp16 = const()[name = tensor("op_2069_to_fp16"), val = tensor(0x1p-24)]; tensor v1_99_cast_fp16 = add(x = var_2068_cast_fp16, y = var_2069_to_fp16)[name = tensor("v1_99_cast_fp16")]; - tensor var_2071_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_201_cast_fp16)[name = tensor("op_2071_cast_fp16")]; + tensor var_2071_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_201_cast_fp16)[name = tensor("op_2071_cast_fp16")]; tensor var_2073_axes_0 = const()[name = tensor("op_2073_axes_0"), val = tensor([2])]; tensor var_2073_keep_dims_0 = const()[name = tensor("op_2073_keep_dims_0"), val = tensor(false)]; tensor var_2073_cast_fp16 = reduce_sum(axes = var_2073_axes_0, keep_dims = var_2073_keep_dims_0, x = var_2071_cast_fp16)[name = tensor("op_2073_cast_fp16")]; tensor mean_99_cast_fp16 = real_div(x = var_2073_cast_fp16, y = v1_99_cast_fp16)[name = tensor("mean_99_cast_fp16")]; tensor var_2075_axes_0 = const()[name = tensor("op_2075_axes_0"), val = tensor([2])]; tensor var_2075_cast_fp16 = expand_dims(axes = var_2075_axes_0, x = mean_99_cast_fp16)[name = tensor("op_2075_cast_fp16")]; - tensor var_2076_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2075_cast_fp16)[name = tensor("op_2076_cast_fp16")]; - tensor dx2_99_cast_fp16 = mul(x = var_2076_cast_fp16, y = var_2076_cast_fp16)[name = tensor("dx2_99_cast_fp16")]; - tensor var_2078_cast_fp16 = mul(x = weights_201_cast_fp16, y = weights_201_cast_fp16)[name = tensor("op_2078_cast_fp16")]; + tensor var_2076_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2075_cast_fp16)[name = tensor("op_2076_cast_fp16")]; + tensor dx2_99_cast_fp16 = mul(x = var_2076_cast_fp16, y = var_2076_cast_fp16)[name = tensor("dx2_99_cast_fp16")]; + tensor var_2078_cast_fp16 = mul(x = weights_201_cast_fp16, y = weights_201_cast_fp16)[name = tensor("op_2078_cast_fp16")]; tensor v2_99_axes_0 = const()[name = tensor("v2_99_axes_0"), val = tensor([2])]; tensor v2_99_keep_dims_0 = const()[name = tensor("v2_99_keep_dims_0"), val = tensor(false)]; tensor v2_99_cast_fp16 = reduce_sum(axes = v2_99_axes_0, keep_dims = v2_99_keep_dims_0, x = var_2078_cast_fp16)[name = tensor("v2_99_cast_fp16")]; - tensor var_2081_cast_fp16 = mul(x = dx2_99_cast_fp16, y = weights_201_cast_fp16)[name = tensor("op_2081_cast_fp16")]; + tensor var_2081_cast_fp16 = mul(x = dx2_99_cast_fp16, y = weights_201_cast_fp16)[name = tensor("op_2081_cast_fp16")]; tensor var_2083_axes_0 = const()[name = tensor("op_2083_axes_0"), val = tensor([2])]; tensor var_2083_keep_dims_0 = const()[name = tensor("op_2083_keep_dims_0"), val = tensor(false)]; tensor var_2083_cast_fp16 = reduce_sum(axes = var_2083_axes_0, keep_dims = var_2083_keep_dims_0, x = var_2081_cast_fp16)[name = tensor("op_2083_cast_fp16")]; @@ -2204,31 +2204,31 @@ program(1.0) tensor var_2091_interleave_0 = const()[name = tensor("op_2091_interleave_0"), val = tensor(false)]; tensor var_2091_cast_fp16 = concat(axis = var_72, interleave = var_2091_interleave_0, values = (mean_99_cast_fp16, std_99_cast_fp16))[name = tensor("op_2091_cast_fp16")]; tensor var_2093_begin_0 = const()[name = tensor("op_2093_begin_0"), val = tensor([0, 50, 0])]; - tensor var_2093_end_0 = const()[name = tensor("op_2093_end_0"), val = tensor([1, 51, 125])]; + tensor var_2093_end_0 = const()[name = tensor("op_2093_end_0"), val = tensor([1, 51, 375])]; tensor var_2093_end_mask_0 = const()[name = tensor("op_2093_end_mask_0"), val = tensor([true, false, true])]; tensor var_2093_squeeze_mask_0 = const()[name = tensor("op_2093_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2093_cast_fp16 = slice_by_index(begin = var_2093_begin_0, end = var_2093_end_0, end_mask = var_2093_end_mask_0, squeeze_mask = var_2093_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2093_cast_fp16")]; + tensor var_2093_cast_fp16 = slice_by_index(begin = var_2093_begin_0, end = var_2093_end_0, end_mask = var_2093_end_mask_0, squeeze_mask = var_2093_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2093_cast_fp16")]; tensor weights_205_axes_0 = const()[name = tensor("weights_205_axes_0"), val = tensor([1])]; - tensor weights_205_cast_fp16 = expand_dims(axes = weights_205_axes_0, x = var_2093_cast_fp16)[name = tensor("weights_205_cast_fp16")]; + tensor weights_205_cast_fp16 = expand_dims(axes = weights_205_axes_0, x = var_2093_cast_fp16)[name = tensor("weights_205_cast_fp16")]; tensor var_2097_axes_0 = const()[name = tensor("op_2097_axes_0"), val = tensor([2])]; tensor var_2097_keep_dims_0 = const()[name = tensor("op_2097_keep_dims_0"), val = tensor(false)]; tensor var_2097_cast_fp16 = reduce_sum(axes = var_2097_axes_0, keep_dims = var_2097_keep_dims_0, x = weights_205_cast_fp16)[name = tensor("op_2097_cast_fp16")]; tensor var_2098_to_fp16 = const()[name = tensor("op_2098_to_fp16"), val = tensor(0x1p-24)]; tensor v1_101_cast_fp16 = add(x = var_2097_cast_fp16, y = var_2098_to_fp16)[name = tensor("v1_101_cast_fp16")]; - tensor var_2100_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_205_cast_fp16)[name = tensor("op_2100_cast_fp16")]; + tensor var_2100_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_205_cast_fp16)[name = tensor("op_2100_cast_fp16")]; tensor var_2102_axes_0 = const()[name = tensor("op_2102_axes_0"), val = tensor([2])]; tensor var_2102_keep_dims_0 = const()[name = tensor("op_2102_keep_dims_0"), val = tensor(false)]; tensor var_2102_cast_fp16 = reduce_sum(axes = var_2102_axes_0, keep_dims = var_2102_keep_dims_0, x = var_2100_cast_fp16)[name = tensor("op_2102_cast_fp16")]; tensor mean_101_cast_fp16 = real_div(x = var_2102_cast_fp16, y = v1_101_cast_fp16)[name = tensor("mean_101_cast_fp16")]; tensor var_2104_axes_0 = const()[name = tensor("op_2104_axes_0"), val = tensor([2])]; tensor var_2104_cast_fp16 = expand_dims(axes = var_2104_axes_0, x = mean_101_cast_fp16)[name = tensor("op_2104_cast_fp16")]; - tensor var_2105_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2104_cast_fp16)[name = tensor("op_2105_cast_fp16")]; - tensor dx2_101_cast_fp16 = mul(x = var_2105_cast_fp16, y = var_2105_cast_fp16)[name = tensor("dx2_101_cast_fp16")]; - tensor var_2107_cast_fp16 = mul(x = weights_205_cast_fp16, y = weights_205_cast_fp16)[name = tensor("op_2107_cast_fp16")]; + tensor var_2105_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2104_cast_fp16)[name = tensor("op_2105_cast_fp16")]; + tensor dx2_101_cast_fp16 = mul(x = var_2105_cast_fp16, y = var_2105_cast_fp16)[name = tensor("dx2_101_cast_fp16")]; + tensor var_2107_cast_fp16 = mul(x = weights_205_cast_fp16, y = weights_205_cast_fp16)[name = tensor("op_2107_cast_fp16")]; tensor v2_101_axes_0 = const()[name = tensor("v2_101_axes_0"), val = tensor([2])]; tensor v2_101_keep_dims_0 = const()[name = tensor("v2_101_keep_dims_0"), val = tensor(false)]; tensor v2_101_cast_fp16 = reduce_sum(axes = v2_101_axes_0, keep_dims = v2_101_keep_dims_0, x = var_2107_cast_fp16)[name = tensor("v2_101_cast_fp16")]; - tensor var_2110_cast_fp16 = mul(x = dx2_101_cast_fp16, y = weights_205_cast_fp16)[name = tensor("op_2110_cast_fp16")]; + tensor var_2110_cast_fp16 = mul(x = dx2_101_cast_fp16, y = weights_205_cast_fp16)[name = tensor("op_2110_cast_fp16")]; tensor var_2112_axes_0 = const()[name = tensor("op_2112_axes_0"), val = tensor([2])]; tensor var_2112_keep_dims_0 = const()[name = tensor("op_2112_keep_dims_0"), val = tensor(false)]; tensor var_2112_cast_fp16 = reduce_sum(axes = var_2112_axes_0, keep_dims = var_2112_keep_dims_0, x = var_2110_cast_fp16)[name = tensor("op_2112_cast_fp16")]; @@ -2241,31 +2241,31 @@ program(1.0) tensor var_2120_interleave_0 = const()[name = tensor("op_2120_interleave_0"), val = tensor(false)]; tensor var_2120_cast_fp16 = concat(axis = var_72, interleave = var_2120_interleave_0, values = (mean_101_cast_fp16, std_101_cast_fp16))[name = tensor("op_2120_cast_fp16")]; tensor var_2122_begin_0 = const()[name = tensor("op_2122_begin_0"), val = tensor([0, 51, 0])]; - tensor var_2122_end_0 = const()[name = tensor("op_2122_end_0"), val = tensor([1, 52, 125])]; + tensor var_2122_end_0 = const()[name = tensor("op_2122_end_0"), val = tensor([1, 52, 375])]; tensor var_2122_end_mask_0 = const()[name = tensor("op_2122_end_mask_0"), val = tensor([true, false, true])]; tensor var_2122_squeeze_mask_0 = const()[name = tensor("op_2122_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2122_cast_fp16 = slice_by_index(begin = var_2122_begin_0, end = var_2122_end_0, end_mask = var_2122_end_mask_0, squeeze_mask = var_2122_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2122_cast_fp16")]; + tensor var_2122_cast_fp16 = slice_by_index(begin = var_2122_begin_0, end = var_2122_end_0, end_mask = var_2122_end_mask_0, squeeze_mask = var_2122_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2122_cast_fp16")]; tensor weights_209_axes_0 = const()[name = tensor("weights_209_axes_0"), val = tensor([1])]; - tensor weights_209_cast_fp16 = expand_dims(axes = weights_209_axes_0, x = var_2122_cast_fp16)[name = tensor("weights_209_cast_fp16")]; + tensor weights_209_cast_fp16 = expand_dims(axes = weights_209_axes_0, x = var_2122_cast_fp16)[name = tensor("weights_209_cast_fp16")]; tensor var_2126_axes_0 = const()[name = tensor("op_2126_axes_0"), val = tensor([2])]; tensor var_2126_keep_dims_0 = const()[name = tensor("op_2126_keep_dims_0"), val = tensor(false)]; tensor var_2126_cast_fp16 = reduce_sum(axes = var_2126_axes_0, keep_dims = var_2126_keep_dims_0, x = weights_209_cast_fp16)[name = tensor("op_2126_cast_fp16")]; tensor var_2127_to_fp16 = const()[name = tensor("op_2127_to_fp16"), val = tensor(0x1p-24)]; tensor v1_103_cast_fp16 = add(x = var_2126_cast_fp16, y = var_2127_to_fp16)[name = tensor("v1_103_cast_fp16")]; - tensor var_2129_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_209_cast_fp16)[name = tensor("op_2129_cast_fp16")]; + tensor var_2129_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_209_cast_fp16)[name = tensor("op_2129_cast_fp16")]; tensor var_2131_axes_0 = const()[name = tensor("op_2131_axes_0"), val = tensor([2])]; tensor var_2131_keep_dims_0 = const()[name = tensor("op_2131_keep_dims_0"), val = tensor(false)]; tensor var_2131_cast_fp16 = reduce_sum(axes = var_2131_axes_0, keep_dims = var_2131_keep_dims_0, x = var_2129_cast_fp16)[name = tensor("op_2131_cast_fp16")]; tensor mean_103_cast_fp16 = real_div(x = var_2131_cast_fp16, y = v1_103_cast_fp16)[name = tensor("mean_103_cast_fp16")]; tensor var_2133_axes_0 = const()[name = tensor("op_2133_axes_0"), val = tensor([2])]; tensor var_2133_cast_fp16 = expand_dims(axes = var_2133_axes_0, x = mean_103_cast_fp16)[name = tensor("op_2133_cast_fp16")]; - tensor var_2134_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2133_cast_fp16)[name = tensor("op_2134_cast_fp16")]; - tensor dx2_103_cast_fp16 = mul(x = var_2134_cast_fp16, y = var_2134_cast_fp16)[name = tensor("dx2_103_cast_fp16")]; - tensor var_2136_cast_fp16 = mul(x = weights_209_cast_fp16, y = weights_209_cast_fp16)[name = tensor("op_2136_cast_fp16")]; + tensor var_2134_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2133_cast_fp16)[name = tensor("op_2134_cast_fp16")]; + tensor dx2_103_cast_fp16 = mul(x = var_2134_cast_fp16, y = var_2134_cast_fp16)[name = tensor("dx2_103_cast_fp16")]; + tensor var_2136_cast_fp16 = mul(x = weights_209_cast_fp16, y = weights_209_cast_fp16)[name = tensor("op_2136_cast_fp16")]; tensor v2_103_axes_0 = const()[name = tensor("v2_103_axes_0"), val = tensor([2])]; tensor v2_103_keep_dims_0 = const()[name = tensor("v2_103_keep_dims_0"), val = tensor(false)]; tensor v2_103_cast_fp16 = reduce_sum(axes = v2_103_axes_0, keep_dims = v2_103_keep_dims_0, x = var_2136_cast_fp16)[name = tensor("v2_103_cast_fp16")]; - tensor var_2139_cast_fp16 = mul(x = dx2_103_cast_fp16, y = weights_209_cast_fp16)[name = tensor("op_2139_cast_fp16")]; + tensor var_2139_cast_fp16 = mul(x = dx2_103_cast_fp16, y = weights_209_cast_fp16)[name = tensor("op_2139_cast_fp16")]; tensor var_2141_axes_0 = const()[name = tensor("op_2141_axes_0"), val = tensor([2])]; tensor var_2141_keep_dims_0 = const()[name = tensor("op_2141_keep_dims_0"), val = tensor(false)]; tensor var_2141_cast_fp16 = reduce_sum(axes = var_2141_axes_0, keep_dims = var_2141_keep_dims_0, x = var_2139_cast_fp16)[name = tensor("op_2141_cast_fp16")]; @@ -2278,31 +2278,31 @@ program(1.0) tensor var_2149_interleave_0 = const()[name = tensor("op_2149_interleave_0"), val = tensor(false)]; tensor var_2149_cast_fp16 = concat(axis = var_72, interleave = var_2149_interleave_0, values = (mean_103_cast_fp16, std_103_cast_fp16))[name = tensor("op_2149_cast_fp16")]; tensor var_2151_begin_0 = const()[name = tensor("op_2151_begin_0"), val = tensor([0, 52, 0])]; - tensor var_2151_end_0 = const()[name = tensor("op_2151_end_0"), val = tensor([1, 53, 125])]; + tensor var_2151_end_0 = const()[name = tensor("op_2151_end_0"), val = tensor([1, 53, 375])]; tensor var_2151_end_mask_0 = const()[name = tensor("op_2151_end_mask_0"), val = tensor([true, false, true])]; tensor var_2151_squeeze_mask_0 = const()[name = tensor("op_2151_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2151_cast_fp16 = slice_by_index(begin = var_2151_begin_0, end = var_2151_end_0, end_mask = var_2151_end_mask_0, squeeze_mask = var_2151_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2151_cast_fp16")]; + tensor var_2151_cast_fp16 = slice_by_index(begin = var_2151_begin_0, end = var_2151_end_0, end_mask = var_2151_end_mask_0, squeeze_mask = var_2151_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2151_cast_fp16")]; tensor weights_213_axes_0 = const()[name = tensor("weights_213_axes_0"), val = tensor([1])]; - tensor weights_213_cast_fp16 = expand_dims(axes = weights_213_axes_0, x = var_2151_cast_fp16)[name = tensor("weights_213_cast_fp16")]; + tensor weights_213_cast_fp16 = expand_dims(axes = weights_213_axes_0, x = var_2151_cast_fp16)[name = tensor("weights_213_cast_fp16")]; tensor var_2155_axes_0 = const()[name = tensor("op_2155_axes_0"), val = tensor([2])]; tensor var_2155_keep_dims_0 = const()[name = tensor("op_2155_keep_dims_0"), val = tensor(false)]; tensor var_2155_cast_fp16 = reduce_sum(axes = var_2155_axes_0, keep_dims = var_2155_keep_dims_0, x = weights_213_cast_fp16)[name = tensor("op_2155_cast_fp16")]; tensor var_2156_to_fp16 = const()[name = tensor("op_2156_to_fp16"), val = tensor(0x1p-24)]; tensor v1_105_cast_fp16 = add(x = var_2155_cast_fp16, y = var_2156_to_fp16)[name = tensor("v1_105_cast_fp16")]; - tensor var_2158_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_213_cast_fp16)[name = tensor("op_2158_cast_fp16")]; + tensor var_2158_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_213_cast_fp16)[name = tensor("op_2158_cast_fp16")]; tensor var_2160_axes_0 = const()[name = tensor("op_2160_axes_0"), val = tensor([2])]; tensor var_2160_keep_dims_0 = const()[name = tensor("op_2160_keep_dims_0"), val = tensor(false)]; tensor var_2160_cast_fp16 = reduce_sum(axes = var_2160_axes_0, keep_dims = var_2160_keep_dims_0, x = var_2158_cast_fp16)[name = tensor("op_2160_cast_fp16")]; tensor mean_105_cast_fp16 = real_div(x = var_2160_cast_fp16, y = v1_105_cast_fp16)[name = tensor("mean_105_cast_fp16")]; tensor var_2162_axes_0 = const()[name = tensor("op_2162_axes_0"), val = tensor([2])]; tensor var_2162_cast_fp16 = expand_dims(axes = var_2162_axes_0, x = mean_105_cast_fp16)[name = tensor("op_2162_cast_fp16")]; - tensor var_2163_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2162_cast_fp16)[name = tensor("op_2163_cast_fp16")]; - tensor dx2_105_cast_fp16 = mul(x = var_2163_cast_fp16, y = var_2163_cast_fp16)[name = tensor("dx2_105_cast_fp16")]; - tensor var_2165_cast_fp16 = mul(x = weights_213_cast_fp16, y = weights_213_cast_fp16)[name = tensor("op_2165_cast_fp16")]; + tensor var_2163_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2162_cast_fp16)[name = tensor("op_2163_cast_fp16")]; + tensor dx2_105_cast_fp16 = mul(x = var_2163_cast_fp16, y = var_2163_cast_fp16)[name = tensor("dx2_105_cast_fp16")]; + tensor var_2165_cast_fp16 = mul(x = weights_213_cast_fp16, y = weights_213_cast_fp16)[name = tensor("op_2165_cast_fp16")]; tensor v2_105_axes_0 = const()[name = tensor("v2_105_axes_0"), val = tensor([2])]; tensor v2_105_keep_dims_0 = const()[name = tensor("v2_105_keep_dims_0"), val = tensor(false)]; tensor v2_105_cast_fp16 = reduce_sum(axes = v2_105_axes_0, keep_dims = v2_105_keep_dims_0, x = var_2165_cast_fp16)[name = tensor("v2_105_cast_fp16")]; - tensor var_2168_cast_fp16 = mul(x = dx2_105_cast_fp16, y = weights_213_cast_fp16)[name = tensor("op_2168_cast_fp16")]; + tensor var_2168_cast_fp16 = mul(x = dx2_105_cast_fp16, y = weights_213_cast_fp16)[name = tensor("op_2168_cast_fp16")]; tensor var_2170_axes_0 = const()[name = tensor("op_2170_axes_0"), val = tensor([2])]; tensor var_2170_keep_dims_0 = const()[name = tensor("op_2170_keep_dims_0"), val = tensor(false)]; tensor var_2170_cast_fp16 = reduce_sum(axes = var_2170_axes_0, keep_dims = var_2170_keep_dims_0, x = var_2168_cast_fp16)[name = tensor("op_2170_cast_fp16")]; @@ -2315,31 +2315,31 @@ program(1.0) tensor var_2178_interleave_0 = const()[name = tensor("op_2178_interleave_0"), val = tensor(false)]; tensor var_2178_cast_fp16 = concat(axis = var_72, interleave = var_2178_interleave_0, values = (mean_105_cast_fp16, std_105_cast_fp16))[name = tensor("op_2178_cast_fp16")]; tensor var_2180_begin_0 = const()[name = tensor("op_2180_begin_0"), val = tensor([0, 53, 0])]; - tensor var_2180_end_0 = const()[name = tensor("op_2180_end_0"), val = tensor([1, 54, 125])]; + tensor var_2180_end_0 = const()[name = tensor("op_2180_end_0"), val = tensor([1, 54, 375])]; tensor var_2180_end_mask_0 = const()[name = tensor("op_2180_end_mask_0"), val = tensor([true, false, true])]; tensor var_2180_squeeze_mask_0 = const()[name = tensor("op_2180_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2180_cast_fp16 = slice_by_index(begin = var_2180_begin_0, end = var_2180_end_0, end_mask = var_2180_end_mask_0, squeeze_mask = var_2180_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2180_cast_fp16")]; + tensor var_2180_cast_fp16 = slice_by_index(begin = var_2180_begin_0, end = var_2180_end_0, end_mask = var_2180_end_mask_0, squeeze_mask = var_2180_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2180_cast_fp16")]; tensor weights_217_axes_0 = const()[name = tensor("weights_217_axes_0"), val = tensor([1])]; - tensor weights_217_cast_fp16 = expand_dims(axes = weights_217_axes_0, x = var_2180_cast_fp16)[name = tensor("weights_217_cast_fp16")]; + tensor weights_217_cast_fp16 = expand_dims(axes = weights_217_axes_0, x = var_2180_cast_fp16)[name = tensor("weights_217_cast_fp16")]; tensor var_2184_axes_0 = const()[name = tensor("op_2184_axes_0"), val = tensor([2])]; tensor var_2184_keep_dims_0 = const()[name = tensor("op_2184_keep_dims_0"), val = tensor(false)]; tensor var_2184_cast_fp16 = reduce_sum(axes = var_2184_axes_0, keep_dims = var_2184_keep_dims_0, x = weights_217_cast_fp16)[name = tensor("op_2184_cast_fp16")]; tensor var_2185_to_fp16 = const()[name = tensor("op_2185_to_fp16"), val = tensor(0x1p-24)]; tensor v1_107_cast_fp16 = add(x = var_2184_cast_fp16, y = var_2185_to_fp16)[name = tensor("v1_107_cast_fp16")]; - tensor var_2187_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_217_cast_fp16)[name = tensor("op_2187_cast_fp16")]; + tensor var_2187_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_217_cast_fp16)[name = tensor("op_2187_cast_fp16")]; tensor var_2189_axes_0 = const()[name = tensor("op_2189_axes_0"), val = tensor([2])]; tensor var_2189_keep_dims_0 = const()[name = tensor("op_2189_keep_dims_0"), val = tensor(false)]; tensor var_2189_cast_fp16 = reduce_sum(axes = var_2189_axes_0, keep_dims = var_2189_keep_dims_0, x = var_2187_cast_fp16)[name = tensor("op_2189_cast_fp16")]; tensor mean_107_cast_fp16 = real_div(x = var_2189_cast_fp16, y = v1_107_cast_fp16)[name = tensor("mean_107_cast_fp16")]; tensor var_2191_axes_0 = const()[name = tensor("op_2191_axes_0"), val = tensor([2])]; tensor var_2191_cast_fp16 = expand_dims(axes = var_2191_axes_0, x = mean_107_cast_fp16)[name = tensor("op_2191_cast_fp16")]; - tensor var_2192_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2191_cast_fp16)[name = tensor("op_2192_cast_fp16")]; - tensor dx2_107_cast_fp16 = mul(x = var_2192_cast_fp16, y = var_2192_cast_fp16)[name = tensor("dx2_107_cast_fp16")]; - tensor var_2194_cast_fp16 = mul(x = weights_217_cast_fp16, y = weights_217_cast_fp16)[name = tensor("op_2194_cast_fp16")]; + tensor var_2192_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2191_cast_fp16)[name = tensor("op_2192_cast_fp16")]; + tensor dx2_107_cast_fp16 = mul(x = var_2192_cast_fp16, y = var_2192_cast_fp16)[name = tensor("dx2_107_cast_fp16")]; + tensor var_2194_cast_fp16 = mul(x = weights_217_cast_fp16, y = weights_217_cast_fp16)[name = tensor("op_2194_cast_fp16")]; tensor v2_107_axes_0 = const()[name = tensor("v2_107_axes_0"), val = tensor([2])]; tensor v2_107_keep_dims_0 = const()[name = tensor("v2_107_keep_dims_0"), val = tensor(false)]; tensor v2_107_cast_fp16 = reduce_sum(axes = v2_107_axes_0, keep_dims = v2_107_keep_dims_0, x = var_2194_cast_fp16)[name = tensor("v2_107_cast_fp16")]; - tensor var_2197_cast_fp16 = mul(x = dx2_107_cast_fp16, y = weights_217_cast_fp16)[name = tensor("op_2197_cast_fp16")]; + tensor var_2197_cast_fp16 = mul(x = dx2_107_cast_fp16, y = weights_217_cast_fp16)[name = tensor("op_2197_cast_fp16")]; tensor var_2199_axes_0 = const()[name = tensor("op_2199_axes_0"), val = tensor([2])]; tensor var_2199_keep_dims_0 = const()[name = tensor("op_2199_keep_dims_0"), val = tensor(false)]; tensor var_2199_cast_fp16 = reduce_sum(axes = var_2199_axes_0, keep_dims = var_2199_keep_dims_0, x = var_2197_cast_fp16)[name = tensor("op_2199_cast_fp16")]; @@ -2352,31 +2352,31 @@ program(1.0) tensor var_2207_interleave_0 = const()[name = tensor("op_2207_interleave_0"), val = tensor(false)]; tensor var_2207_cast_fp16 = concat(axis = var_72, interleave = var_2207_interleave_0, values = (mean_107_cast_fp16, std_107_cast_fp16))[name = tensor("op_2207_cast_fp16")]; tensor var_2209_begin_0 = const()[name = tensor("op_2209_begin_0"), val = tensor([0, 54, 0])]; - tensor var_2209_end_0 = const()[name = tensor("op_2209_end_0"), val = tensor([1, 55, 125])]; + tensor var_2209_end_0 = const()[name = tensor("op_2209_end_0"), val = tensor([1, 55, 375])]; tensor var_2209_end_mask_0 = const()[name = tensor("op_2209_end_mask_0"), val = tensor([true, false, true])]; tensor var_2209_squeeze_mask_0 = const()[name = tensor("op_2209_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2209_cast_fp16 = slice_by_index(begin = var_2209_begin_0, end = var_2209_end_0, end_mask = var_2209_end_mask_0, squeeze_mask = var_2209_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2209_cast_fp16")]; + tensor var_2209_cast_fp16 = slice_by_index(begin = var_2209_begin_0, end = var_2209_end_0, end_mask = var_2209_end_mask_0, squeeze_mask = var_2209_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2209_cast_fp16")]; tensor weights_221_axes_0 = const()[name = tensor("weights_221_axes_0"), val = tensor([1])]; - tensor weights_221_cast_fp16 = expand_dims(axes = weights_221_axes_0, x = var_2209_cast_fp16)[name = tensor("weights_221_cast_fp16")]; + tensor weights_221_cast_fp16 = expand_dims(axes = weights_221_axes_0, x = var_2209_cast_fp16)[name = tensor("weights_221_cast_fp16")]; tensor var_2213_axes_0 = const()[name = tensor("op_2213_axes_0"), val = tensor([2])]; tensor var_2213_keep_dims_0 = const()[name = tensor("op_2213_keep_dims_0"), val = tensor(false)]; tensor var_2213_cast_fp16 = reduce_sum(axes = var_2213_axes_0, keep_dims = var_2213_keep_dims_0, x = weights_221_cast_fp16)[name = tensor("op_2213_cast_fp16")]; tensor var_2214_to_fp16 = const()[name = tensor("op_2214_to_fp16"), val = tensor(0x1p-24)]; tensor v1_109_cast_fp16 = add(x = var_2213_cast_fp16, y = var_2214_to_fp16)[name = tensor("v1_109_cast_fp16")]; - tensor var_2216_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_221_cast_fp16)[name = tensor("op_2216_cast_fp16")]; + tensor var_2216_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_221_cast_fp16)[name = tensor("op_2216_cast_fp16")]; tensor var_2218_axes_0 = const()[name = tensor("op_2218_axes_0"), val = tensor([2])]; tensor var_2218_keep_dims_0 = const()[name = tensor("op_2218_keep_dims_0"), val = tensor(false)]; tensor var_2218_cast_fp16 = reduce_sum(axes = var_2218_axes_0, keep_dims = var_2218_keep_dims_0, x = var_2216_cast_fp16)[name = tensor("op_2218_cast_fp16")]; tensor mean_109_cast_fp16 = real_div(x = var_2218_cast_fp16, y = v1_109_cast_fp16)[name = tensor("mean_109_cast_fp16")]; tensor var_2220_axes_0 = const()[name = tensor("op_2220_axes_0"), val = tensor([2])]; tensor var_2220_cast_fp16 = expand_dims(axes = var_2220_axes_0, x = mean_109_cast_fp16)[name = tensor("op_2220_cast_fp16")]; - tensor var_2221_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2220_cast_fp16)[name = tensor("op_2221_cast_fp16")]; - tensor dx2_109_cast_fp16 = mul(x = var_2221_cast_fp16, y = var_2221_cast_fp16)[name = tensor("dx2_109_cast_fp16")]; - tensor var_2223_cast_fp16 = mul(x = weights_221_cast_fp16, y = weights_221_cast_fp16)[name = tensor("op_2223_cast_fp16")]; + tensor var_2221_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2220_cast_fp16)[name = tensor("op_2221_cast_fp16")]; + tensor dx2_109_cast_fp16 = mul(x = var_2221_cast_fp16, y = var_2221_cast_fp16)[name = tensor("dx2_109_cast_fp16")]; + tensor var_2223_cast_fp16 = mul(x = weights_221_cast_fp16, y = weights_221_cast_fp16)[name = tensor("op_2223_cast_fp16")]; tensor v2_109_axes_0 = const()[name = tensor("v2_109_axes_0"), val = tensor([2])]; tensor v2_109_keep_dims_0 = const()[name = tensor("v2_109_keep_dims_0"), val = tensor(false)]; tensor v2_109_cast_fp16 = reduce_sum(axes = v2_109_axes_0, keep_dims = v2_109_keep_dims_0, x = var_2223_cast_fp16)[name = tensor("v2_109_cast_fp16")]; - tensor var_2226_cast_fp16 = mul(x = dx2_109_cast_fp16, y = weights_221_cast_fp16)[name = tensor("op_2226_cast_fp16")]; + tensor var_2226_cast_fp16 = mul(x = dx2_109_cast_fp16, y = weights_221_cast_fp16)[name = tensor("op_2226_cast_fp16")]; tensor var_2228_axes_0 = const()[name = tensor("op_2228_axes_0"), val = tensor([2])]; tensor var_2228_keep_dims_0 = const()[name = tensor("op_2228_keep_dims_0"), val = tensor(false)]; tensor var_2228_cast_fp16 = reduce_sum(axes = var_2228_axes_0, keep_dims = var_2228_keep_dims_0, x = var_2226_cast_fp16)[name = tensor("op_2228_cast_fp16")]; @@ -2389,31 +2389,31 @@ program(1.0) tensor var_2236_interleave_0 = const()[name = tensor("op_2236_interleave_0"), val = tensor(false)]; tensor var_2236_cast_fp16 = concat(axis = var_72, interleave = var_2236_interleave_0, values = (mean_109_cast_fp16, std_109_cast_fp16))[name = tensor("op_2236_cast_fp16")]; tensor var_2238_begin_0 = const()[name = tensor("op_2238_begin_0"), val = tensor([0, 55, 0])]; - tensor var_2238_end_0 = const()[name = tensor("op_2238_end_0"), val = tensor([1, 56, 125])]; + tensor var_2238_end_0 = const()[name = tensor("op_2238_end_0"), val = tensor([1, 56, 375])]; tensor var_2238_end_mask_0 = const()[name = tensor("op_2238_end_mask_0"), val = tensor([true, false, true])]; tensor var_2238_squeeze_mask_0 = const()[name = tensor("op_2238_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2238_cast_fp16 = slice_by_index(begin = var_2238_begin_0, end = var_2238_end_0, end_mask = var_2238_end_mask_0, squeeze_mask = var_2238_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2238_cast_fp16")]; + tensor var_2238_cast_fp16 = slice_by_index(begin = var_2238_begin_0, end = var_2238_end_0, end_mask = var_2238_end_mask_0, squeeze_mask = var_2238_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2238_cast_fp16")]; tensor weights_225_axes_0 = const()[name = tensor("weights_225_axes_0"), val = tensor([1])]; - tensor weights_225_cast_fp16 = expand_dims(axes = weights_225_axes_0, x = var_2238_cast_fp16)[name = tensor("weights_225_cast_fp16")]; + tensor weights_225_cast_fp16 = expand_dims(axes = weights_225_axes_0, x = var_2238_cast_fp16)[name = tensor("weights_225_cast_fp16")]; tensor var_2242_axes_0 = const()[name = tensor("op_2242_axes_0"), val = tensor([2])]; tensor var_2242_keep_dims_0 = const()[name = tensor("op_2242_keep_dims_0"), val = tensor(false)]; tensor var_2242_cast_fp16 = reduce_sum(axes = var_2242_axes_0, keep_dims = var_2242_keep_dims_0, x = weights_225_cast_fp16)[name = tensor("op_2242_cast_fp16")]; tensor var_2243_to_fp16 = const()[name = tensor("op_2243_to_fp16"), val = tensor(0x1p-24)]; tensor v1_111_cast_fp16 = add(x = var_2242_cast_fp16, y = var_2243_to_fp16)[name = tensor("v1_111_cast_fp16")]; - tensor var_2245_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_225_cast_fp16)[name = tensor("op_2245_cast_fp16")]; + tensor var_2245_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_225_cast_fp16)[name = tensor("op_2245_cast_fp16")]; tensor var_2247_axes_0 = const()[name = tensor("op_2247_axes_0"), val = tensor([2])]; tensor var_2247_keep_dims_0 = const()[name = tensor("op_2247_keep_dims_0"), val = tensor(false)]; tensor var_2247_cast_fp16 = reduce_sum(axes = var_2247_axes_0, keep_dims = var_2247_keep_dims_0, x = var_2245_cast_fp16)[name = tensor("op_2247_cast_fp16")]; tensor mean_111_cast_fp16 = real_div(x = var_2247_cast_fp16, y = v1_111_cast_fp16)[name = tensor("mean_111_cast_fp16")]; tensor var_2249_axes_0 = const()[name = tensor("op_2249_axes_0"), val = tensor([2])]; tensor var_2249_cast_fp16 = expand_dims(axes = var_2249_axes_0, x = mean_111_cast_fp16)[name = tensor("op_2249_cast_fp16")]; - tensor var_2250_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2249_cast_fp16)[name = tensor("op_2250_cast_fp16")]; - tensor dx2_111_cast_fp16 = mul(x = var_2250_cast_fp16, y = var_2250_cast_fp16)[name = tensor("dx2_111_cast_fp16")]; - tensor var_2252_cast_fp16 = mul(x = weights_225_cast_fp16, y = weights_225_cast_fp16)[name = tensor("op_2252_cast_fp16")]; + tensor var_2250_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2249_cast_fp16)[name = tensor("op_2250_cast_fp16")]; + tensor dx2_111_cast_fp16 = mul(x = var_2250_cast_fp16, y = var_2250_cast_fp16)[name = tensor("dx2_111_cast_fp16")]; + tensor var_2252_cast_fp16 = mul(x = weights_225_cast_fp16, y = weights_225_cast_fp16)[name = tensor("op_2252_cast_fp16")]; tensor v2_111_axes_0 = const()[name = tensor("v2_111_axes_0"), val = tensor([2])]; tensor v2_111_keep_dims_0 = const()[name = tensor("v2_111_keep_dims_0"), val = tensor(false)]; tensor v2_111_cast_fp16 = reduce_sum(axes = v2_111_axes_0, keep_dims = v2_111_keep_dims_0, x = var_2252_cast_fp16)[name = tensor("v2_111_cast_fp16")]; - tensor var_2255_cast_fp16 = mul(x = dx2_111_cast_fp16, y = weights_225_cast_fp16)[name = tensor("op_2255_cast_fp16")]; + tensor var_2255_cast_fp16 = mul(x = dx2_111_cast_fp16, y = weights_225_cast_fp16)[name = tensor("op_2255_cast_fp16")]; tensor var_2257_axes_0 = const()[name = tensor("op_2257_axes_0"), val = tensor([2])]; tensor var_2257_keep_dims_0 = const()[name = tensor("op_2257_keep_dims_0"), val = tensor(false)]; tensor var_2257_cast_fp16 = reduce_sum(axes = var_2257_axes_0, keep_dims = var_2257_keep_dims_0, x = var_2255_cast_fp16)[name = tensor("op_2257_cast_fp16")]; @@ -2426,31 +2426,31 @@ program(1.0) tensor var_2265_interleave_0 = const()[name = tensor("op_2265_interleave_0"), val = tensor(false)]; tensor var_2265_cast_fp16 = concat(axis = var_72, interleave = var_2265_interleave_0, values = (mean_111_cast_fp16, std_111_cast_fp16))[name = tensor("op_2265_cast_fp16")]; tensor var_2267_begin_0 = const()[name = tensor("op_2267_begin_0"), val = tensor([0, 56, 0])]; - tensor var_2267_end_0 = const()[name = tensor("op_2267_end_0"), val = tensor([1, 57, 125])]; + tensor var_2267_end_0 = const()[name = tensor("op_2267_end_0"), val = tensor([1, 57, 375])]; tensor var_2267_end_mask_0 = const()[name = tensor("op_2267_end_mask_0"), val = tensor([true, false, true])]; tensor var_2267_squeeze_mask_0 = const()[name = tensor("op_2267_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2267_cast_fp16 = slice_by_index(begin = var_2267_begin_0, end = var_2267_end_0, end_mask = var_2267_end_mask_0, squeeze_mask = var_2267_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2267_cast_fp16")]; + tensor var_2267_cast_fp16 = slice_by_index(begin = var_2267_begin_0, end = var_2267_end_0, end_mask = var_2267_end_mask_0, squeeze_mask = var_2267_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2267_cast_fp16")]; tensor weights_229_axes_0 = const()[name = tensor("weights_229_axes_0"), val = tensor([1])]; - tensor weights_229_cast_fp16 = expand_dims(axes = weights_229_axes_0, x = var_2267_cast_fp16)[name = tensor("weights_229_cast_fp16")]; + tensor weights_229_cast_fp16 = expand_dims(axes = weights_229_axes_0, x = var_2267_cast_fp16)[name = tensor("weights_229_cast_fp16")]; tensor var_2271_axes_0 = const()[name = tensor("op_2271_axes_0"), val = tensor([2])]; tensor var_2271_keep_dims_0 = const()[name = tensor("op_2271_keep_dims_0"), val = tensor(false)]; tensor var_2271_cast_fp16 = reduce_sum(axes = var_2271_axes_0, keep_dims = var_2271_keep_dims_0, x = weights_229_cast_fp16)[name = tensor("op_2271_cast_fp16")]; tensor var_2272_to_fp16 = const()[name = tensor("op_2272_to_fp16"), val = tensor(0x1p-24)]; tensor v1_113_cast_fp16 = add(x = var_2271_cast_fp16, y = var_2272_to_fp16)[name = tensor("v1_113_cast_fp16")]; - tensor var_2274_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_229_cast_fp16)[name = tensor("op_2274_cast_fp16")]; + tensor var_2274_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_229_cast_fp16)[name = tensor("op_2274_cast_fp16")]; tensor var_2276_axes_0 = const()[name = tensor("op_2276_axes_0"), val = tensor([2])]; tensor var_2276_keep_dims_0 = const()[name = tensor("op_2276_keep_dims_0"), val = tensor(false)]; tensor var_2276_cast_fp16 = reduce_sum(axes = var_2276_axes_0, keep_dims = var_2276_keep_dims_0, x = var_2274_cast_fp16)[name = tensor("op_2276_cast_fp16")]; tensor mean_113_cast_fp16 = real_div(x = var_2276_cast_fp16, y = v1_113_cast_fp16)[name = tensor("mean_113_cast_fp16")]; tensor var_2278_axes_0 = const()[name = tensor("op_2278_axes_0"), val = tensor([2])]; tensor var_2278_cast_fp16 = expand_dims(axes = var_2278_axes_0, x = mean_113_cast_fp16)[name = tensor("op_2278_cast_fp16")]; - tensor var_2279_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2278_cast_fp16)[name = tensor("op_2279_cast_fp16")]; - tensor dx2_113_cast_fp16 = mul(x = var_2279_cast_fp16, y = var_2279_cast_fp16)[name = tensor("dx2_113_cast_fp16")]; - tensor var_2281_cast_fp16 = mul(x = weights_229_cast_fp16, y = weights_229_cast_fp16)[name = tensor("op_2281_cast_fp16")]; + tensor var_2279_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2278_cast_fp16)[name = tensor("op_2279_cast_fp16")]; + tensor dx2_113_cast_fp16 = mul(x = var_2279_cast_fp16, y = var_2279_cast_fp16)[name = tensor("dx2_113_cast_fp16")]; + tensor var_2281_cast_fp16 = mul(x = weights_229_cast_fp16, y = weights_229_cast_fp16)[name = tensor("op_2281_cast_fp16")]; tensor v2_113_axes_0 = const()[name = tensor("v2_113_axes_0"), val = tensor([2])]; tensor v2_113_keep_dims_0 = const()[name = tensor("v2_113_keep_dims_0"), val = tensor(false)]; tensor v2_113_cast_fp16 = reduce_sum(axes = v2_113_axes_0, keep_dims = v2_113_keep_dims_0, x = var_2281_cast_fp16)[name = tensor("v2_113_cast_fp16")]; - tensor var_2284_cast_fp16 = mul(x = dx2_113_cast_fp16, y = weights_229_cast_fp16)[name = tensor("op_2284_cast_fp16")]; + tensor var_2284_cast_fp16 = mul(x = dx2_113_cast_fp16, y = weights_229_cast_fp16)[name = tensor("op_2284_cast_fp16")]; tensor var_2286_axes_0 = const()[name = tensor("op_2286_axes_0"), val = tensor([2])]; tensor var_2286_keep_dims_0 = const()[name = tensor("op_2286_keep_dims_0"), val = tensor(false)]; tensor var_2286_cast_fp16 = reduce_sum(axes = var_2286_axes_0, keep_dims = var_2286_keep_dims_0, x = var_2284_cast_fp16)[name = tensor("op_2286_cast_fp16")]; @@ -2463,31 +2463,31 @@ program(1.0) tensor var_2294_interleave_0 = const()[name = tensor("op_2294_interleave_0"), val = tensor(false)]; tensor var_2294_cast_fp16 = concat(axis = var_72, interleave = var_2294_interleave_0, values = (mean_113_cast_fp16, std_113_cast_fp16))[name = tensor("op_2294_cast_fp16")]; tensor var_2296_begin_0 = const()[name = tensor("op_2296_begin_0"), val = tensor([0, 57, 0])]; - tensor var_2296_end_0 = const()[name = tensor("op_2296_end_0"), val = tensor([1, 58, 125])]; + tensor var_2296_end_0 = const()[name = tensor("op_2296_end_0"), val = tensor([1, 58, 375])]; tensor var_2296_end_mask_0 = const()[name = tensor("op_2296_end_mask_0"), val = tensor([true, false, true])]; tensor var_2296_squeeze_mask_0 = const()[name = tensor("op_2296_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2296_cast_fp16 = slice_by_index(begin = var_2296_begin_0, end = var_2296_end_0, end_mask = var_2296_end_mask_0, squeeze_mask = var_2296_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2296_cast_fp16")]; + tensor var_2296_cast_fp16 = slice_by_index(begin = var_2296_begin_0, end = var_2296_end_0, end_mask = var_2296_end_mask_0, squeeze_mask = var_2296_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2296_cast_fp16")]; tensor weights_233_axes_0 = const()[name = tensor("weights_233_axes_0"), val = tensor([1])]; - tensor weights_233_cast_fp16 = expand_dims(axes = weights_233_axes_0, x = var_2296_cast_fp16)[name = tensor("weights_233_cast_fp16")]; + tensor weights_233_cast_fp16 = expand_dims(axes = weights_233_axes_0, x = var_2296_cast_fp16)[name = tensor("weights_233_cast_fp16")]; tensor var_2300_axes_0 = const()[name = tensor("op_2300_axes_0"), val = tensor([2])]; tensor var_2300_keep_dims_0 = const()[name = tensor("op_2300_keep_dims_0"), val = tensor(false)]; tensor var_2300_cast_fp16 = reduce_sum(axes = var_2300_axes_0, keep_dims = var_2300_keep_dims_0, x = weights_233_cast_fp16)[name = tensor("op_2300_cast_fp16")]; tensor var_2301_to_fp16 = const()[name = tensor("op_2301_to_fp16"), val = tensor(0x1p-24)]; tensor v1_115_cast_fp16 = add(x = var_2300_cast_fp16, y = var_2301_to_fp16)[name = tensor("v1_115_cast_fp16")]; - tensor var_2303_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_233_cast_fp16)[name = tensor("op_2303_cast_fp16")]; + tensor var_2303_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_233_cast_fp16)[name = tensor("op_2303_cast_fp16")]; tensor var_2305_axes_0 = const()[name = tensor("op_2305_axes_0"), val = tensor([2])]; tensor var_2305_keep_dims_0 = const()[name = tensor("op_2305_keep_dims_0"), val = tensor(false)]; tensor var_2305_cast_fp16 = reduce_sum(axes = var_2305_axes_0, keep_dims = var_2305_keep_dims_0, x = var_2303_cast_fp16)[name = tensor("op_2305_cast_fp16")]; tensor mean_115_cast_fp16 = real_div(x = var_2305_cast_fp16, y = v1_115_cast_fp16)[name = tensor("mean_115_cast_fp16")]; tensor var_2307_axes_0 = const()[name = tensor("op_2307_axes_0"), val = tensor([2])]; tensor var_2307_cast_fp16 = expand_dims(axes = var_2307_axes_0, x = mean_115_cast_fp16)[name = tensor("op_2307_cast_fp16")]; - tensor var_2308_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2307_cast_fp16)[name = tensor("op_2308_cast_fp16")]; - tensor dx2_115_cast_fp16 = mul(x = var_2308_cast_fp16, y = var_2308_cast_fp16)[name = tensor("dx2_115_cast_fp16")]; - tensor var_2310_cast_fp16 = mul(x = weights_233_cast_fp16, y = weights_233_cast_fp16)[name = tensor("op_2310_cast_fp16")]; + tensor var_2308_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2307_cast_fp16)[name = tensor("op_2308_cast_fp16")]; + tensor dx2_115_cast_fp16 = mul(x = var_2308_cast_fp16, y = var_2308_cast_fp16)[name = tensor("dx2_115_cast_fp16")]; + tensor var_2310_cast_fp16 = mul(x = weights_233_cast_fp16, y = weights_233_cast_fp16)[name = tensor("op_2310_cast_fp16")]; tensor v2_115_axes_0 = const()[name = tensor("v2_115_axes_0"), val = tensor([2])]; tensor v2_115_keep_dims_0 = const()[name = tensor("v2_115_keep_dims_0"), val = tensor(false)]; tensor v2_115_cast_fp16 = reduce_sum(axes = v2_115_axes_0, keep_dims = v2_115_keep_dims_0, x = var_2310_cast_fp16)[name = tensor("v2_115_cast_fp16")]; - tensor var_2313_cast_fp16 = mul(x = dx2_115_cast_fp16, y = weights_233_cast_fp16)[name = tensor("op_2313_cast_fp16")]; + tensor var_2313_cast_fp16 = mul(x = dx2_115_cast_fp16, y = weights_233_cast_fp16)[name = tensor("op_2313_cast_fp16")]; tensor var_2315_axes_0 = const()[name = tensor("op_2315_axes_0"), val = tensor([2])]; tensor var_2315_keep_dims_0 = const()[name = tensor("op_2315_keep_dims_0"), val = tensor(false)]; tensor var_2315_cast_fp16 = reduce_sum(axes = var_2315_axes_0, keep_dims = var_2315_keep_dims_0, x = var_2313_cast_fp16)[name = tensor("op_2315_cast_fp16")]; @@ -2500,31 +2500,31 @@ program(1.0) tensor var_2323_interleave_0 = const()[name = tensor("op_2323_interleave_0"), val = tensor(false)]; tensor var_2323_cast_fp16 = concat(axis = var_72, interleave = var_2323_interleave_0, values = (mean_115_cast_fp16, std_115_cast_fp16))[name = tensor("op_2323_cast_fp16")]; tensor var_2325_begin_0 = const()[name = tensor("op_2325_begin_0"), val = tensor([0, 58, 0])]; - tensor var_2325_end_0 = const()[name = tensor("op_2325_end_0"), val = tensor([1, 59, 125])]; + tensor var_2325_end_0 = const()[name = tensor("op_2325_end_0"), val = tensor([1, 59, 375])]; tensor var_2325_end_mask_0 = const()[name = tensor("op_2325_end_mask_0"), val = tensor([true, false, true])]; tensor var_2325_squeeze_mask_0 = const()[name = tensor("op_2325_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2325_cast_fp16 = slice_by_index(begin = var_2325_begin_0, end = var_2325_end_0, end_mask = var_2325_end_mask_0, squeeze_mask = var_2325_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2325_cast_fp16")]; + tensor var_2325_cast_fp16 = slice_by_index(begin = var_2325_begin_0, end = var_2325_end_0, end_mask = var_2325_end_mask_0, squeeze_mask = var_2325_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2325_cast_fp16")]; tensor weights_237_axes_0 = const()[name = tensor("weights_237_axes_0"), val = tensor([1])]; - tensor weights_237_cast_fp16 = expand_dims(axes = weights_237_axes_0, x = var_2325_cast_fp16)[name = tensor("weights_237_cast_fp16")]; + tensor weights_237_cast_fp16 = expand_dims(axes = weights_237_axes_0, x = var_2325_cast_fp16)[name = tensor("weights_237_cast_fp16")]; tensor var_2329_axes_0 = const()[name = tensor("op_2329_axes_0"), val = tensor([2])]; tensor var_2329_keep_dims_0 = const()[name = tensor("op_2329_keep_dims_0"), val = tensor(false)]; tensor var_2329_cast_fp16 = reduce_sum(axes = var_2329_axes_0, keep_dims = var_2329_keep_dims_0, x = weights_237_cast_fp16)[name = tensor("op_2329_cast_fp16")]; tensor var_2330_to_fp16 = const()[name = tensor("op_2330_to_fp16"), val = tensor(0x1p-24)]; tensor v1_117_cast_fp16 = add(x = var_2329_cast_fp16, y = var_2330_to_fp16)[name = tensor("v1_117_cast_fp16")]; - tensor var_2332_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_237_cast_fp16)[name = tensor("op_2332_cast_fp16")]; + tensor var_2332_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_237_cast_fp16)[name = tensor("op_2332_cast_fp16")]; tensor var_2334_axes_0 = const()[name = tensor("op_2334_axes_0"), val = tensor([2])]; tensor var_2334_keep_dims_0 = const()[name = tensor("op_2334_keep_dims_0"), val = tensor(false)]; tensor var_2334_cast_fp16 = reduce_sum(axes = var_2334_axes_0, keep_dims = var_2334_keep_dims_0, x = var_2332_cast_fp16)[name = tensor("op_2334_cast_fp16")]; tensor mean_117_cast_fp16 = real_div(x = var_2334_cast_fp16, y = v1_117_cast_fp16)[name = tensor("mean_117_cast_fp16")]; tensor var_2336_axes_0 = const()[name = tensor("op_2336_axes_0"), val = tensor([2])]; tensor var_2336_cast_fp16 = expand_dims(axes = var_2336_axes_0, x = mean_117_cast_fp16)[name = tensor("op_2336_cast_fp16")]; - tensor var_2337_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2336_cast_fp16)[name = tensor("op_2337_cast_fp16")]; - tensor dx2_117_cast_fp16 = mul(x = var_2337_cast_fp16, y = var_2337_cast_fp16)[name = tensor("dx2_117_cast_fp16")]; - tensor var_2339_cast_fp16 = mul(x = weights_237_cast_fp16, y = weights_237_cast_fp16)[name = tensor("op_2339_cast_fp16")]; + tensor var_2337_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2336_cast_fp16)[name = tensor("op_2337_cast_fp16")]; + tensor dx2_117_cast_fp16 = mul(x = var_2337_cast_fp16, y = var_2337_cast_fp16)[name = tensor("dx2_117_cast_fp16")]; + tensor var_2339_cast_fp16 = mul(x = weights_237_cast_fp16, y = weights_237_cast_fp16)[name = tensor("op_2339_cast_fp16")]; tensor v2_117_axes_0 = const()[name = tensor("v2_117_axes_0"), val = tensor([2])]; tensor v2_117_keep_dims_0 = const()[name = tensor("v2_117_keep_dims_0"), val = tensor(false)]; tensor v2_117_cast_fp16 = reduce_sum(axes = v2_117_axes_0, keep_dims = v2_117_keep_dims_0, x = var_2339_cast_fp16)[name = tensor("v2_117_cast_fp16")]; - tensor var_2342_cast_fp16 = mul(x = dx2_117_cast_fp16, y = weights_237_cast_fp16)[name = tensor("op_2342_cast_fp16")]; + tensor var_2342_cast_fp16 = mul(x = dx2_117_cast_fp16, y = weights_237_cast_fp16)[name = tensor("op_2342_cast_fp16")]; tensor var_2344_axes_0 = const()[name = tensor("op_2344_axes_0"), val = tensor([2])]; tensor var_2344_keep_dims_0 = const()[name = tensor("op_2344_keep_dims_0"), val = tensor(false)]; tensor var_2344_cast_fp16 = reduce_sum(axes = var_2344_axes_0, keep_dims = var_2344_keep_dims_0, x = var_2342_cast_fp16)[name = tensor("op_2344_cast_fp16")]; @@ -2537,31 +2537,31 @@ program(1.0) tensor var_2352_interleave_0 = const()[name = tensor("op_2352_interleave_0"), val = tensor(false)]; tensor var_2352_cast_fp16 = concat(axis = var_72, interleave = var_2352_interleave_0, values = (mean_117_cast_fp16, std_117_cast_fp16))[name = tensor("op_2352_cast_fp16")]; tensor var_2354_begin_0 = const()[name = tensor("op_2354_begin_0"), val = tensor([0, 59, 0])]; - tensor var_2354_end_0 = const()[name = tensor("op_2354_end_0"), val = tensor([1, 60, 125])]; + tensor var_2354_end_0 = const()[name = tensor("op_2354_end_0"), val = tensor([1, 60, 375])]; tensor var_2354_end_mask_0 = const()[name = tensor("op_2354_end_mask_0"), val = tensor([true, false, true])]; tensor var_2354_squeeze_mask_0 = const()[name = tensor("op_2354_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2354_cast_fp16 = slice_by_index(begin = var_2354_begin_0, end = var_2354_end_0, end_mask = var_2354_end_mask_0, squeeze_mask = var_2354_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2354_cast_fp16")]; + tensor var_2354_cast_fp16 = slice_by_index(begin = var_2354_begin_0, end = var_2354_end_0, end_mask = var_2354_end_mask_0, squeeze_mask = var_2354_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2354_cast_fp16")]; tensor weights_241_axes_0 = const()[name = tensor("weights_241_axes_0"), val = tensor([1])]; - tensor weights_241_cast_fp16 = expand_dims(axes = weights_241_axes_0, x = var_2354_cast_fp16)[name = tensor("weights_241_cast_fp16")]; + tensor weights_241_cast_fp16 = expand_dims(axes = weights_241_axes_0, x = var_2354_cast_fp16)[name = tensor("weights_241_cast_fp16")]; tensor var_2358_axes_0 = const()[name = tensor("op_2358_axes_0"), val = tensor([2])]; tensor var_2358_keep_dims_0 = const()[name = tensor("op_2358_keep_dims_0"), val = tensor(false)]; tensor var_2358_cast_fp16 = reduce_sum(axes = var_2358_axes_0, keep_dims = var_2358_keep_dims_0, x = weights_241_cast_fp16)[name = tensor("op_2358_cast_fp16")]; tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(0x1p-24)]; tensor v1_119_cast_fp16 = add(x = var_2358_cast_fp16, y = var_2359_to_fp16)[name = tensor("v1_119_cast_fp16")]; - tensor var_2361_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_241_cast_fp16)[name = tensor("op_2361_cast_fp16")]; + tensor var_2361_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_241_cast_fp16)[name = tensor("op_2361_cast_fp16")]; tensor var_2363_axes_0 = const()[name = tensor("op_2363_axes_0"), val = tensor([2])]; tensor var_2363_keep_dims_0 = const()[name = tensor("op_2363_keep_dims_0"), val = tensor(false)]; tensor var_2363_cast_fp16 = reduce_sum(axes = var_2363_axes_0, keep_dims = var_2363_keep_dims_0, x = var_2361_cast_fp16)[name = tensor("op_2363_cast_fp16")]; tensor mean_119_cast_fp16 = real_div(x = var_2363_cast_fp16, y = v1_119_cast_fp16)[name = tensor("mean_119_cast_fp16")]; tensor var_2365_axes_0 = const()[name = tensor("op_2365_axes_0"), val = tensor([2])]; tensor var_2365_cast_fp16 = expand_dims(axes = var_2365_axes_0, x = mean_119_cast_fp16)[name = tensor("op_2365_cast_fp16")]; - tensor var_2366_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2365_cast_fp16)[name = tensor("op_2366_cast_fp16")]; - tensor dx2_119_cast_fp16 = mul(x = var_2366_cast_fp16, y = var_2366_cast_fp16)[name = tensor("dx2_119_cast_fp16")]; - tensor var_2368_cast_fp16 = mul(x = weights_241_cast_fp16, y = weights_241_cast_fp16)[name = tensor("op_2368_cast_fp16")]; + tensor var_2366_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2365_cast_fp16)[name = tensor("op_2366_cast_fp16")]; + tensor dx2_119_cast_fp16 = mul(x = var_2366_cast_fp16, y = var_2366_cast_fp16)[name = tensor("dx2_119_cast_fp16")]; + tensor var_2368_cast_fp16 = mul(x = weights_241_cast_fp16, y = weights_241_cast_fp16)[name = tensor("op_2368_cast_fp16")]; tensor v2_119_axes_0 = const()[name = tensor("v2_119_axes_0"), val = tensor([2])]; tensor v2_119_keep_dims_0 = const()[name = tensor("v2_119_keep_dims_0"), val = tensor(false)]; tensor v2_119_cast_fp16 = reduce_sum(axes = v2_119_axes_0, keep_dims = v2_119_keep_dims_0, x = var_2368_cast_fp16)[name = tensor("v2_119_cast_fp16")]; - tensor var_2371_cast_fp16 = mul(x = dx2_119_cast_fp16, y = weights_241_cast_fp16)[name = tensor("op_2371_cast_fp16")]; + tensor var_2371_cast_fp16 = mul(x = dx2_119_cast_fp16, y = weights_241_cast_fp16)[name = tensor("op_2371_cast_fp16")]; tensor var_2373_axes_0 = const()[name = tensor("op_2373_axes_0"), val = tensor([2])]; tensor var_2373_keep_dims_0 = const()[name = tensor("op_2373_keep_dims_0"), val = tensor(false)]; tensor var_2373_cast_fp16 = reduce_sum(axes = var_2373_axes_0, keep_dims = var_2373_keep_dims_0, x = var_2371_cast_fp16)[name = tensor("op_2373_cast_fp16")]; @@ -2574,31 +2574,31 @@ program(1.0) tensor var_2381_interleave_0 = const()[name = tensor("op_2381_interleave_0"), val = tensor(false)]; tensor var_2381_cast_fp16 = concat(axis = var_72, interleave = var_2381_interleave_0, values = (mean_119_cast_fp16, std_119_cast_fp16))[name = tensor("op_2381_cast_fp16")]; tensor var_2383_begin_0 = const()[name = tensor("op_2383_begin_0"), val = tensor([0, 60, 0])]; - tensor var_2383_end_0 = const()[name = tensor("op_2383_end_0"), val = tensor([1, 61, 125])]; + tensor var_2383_end_0 = const()[name = tensor("op_2383_end_0"), val = tensor([1, 61, 375])]; tensor var_2383_end_mask_0 = const()[name = tensor("op_2383_end_mask_0"), val = tensor([true, false, true])]; tensor var_2383_squeeze_mask_0 = const()[name = tensor("op_2383_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, squeeze_mask = var_2383_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2383_cast_fp16")]; + tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, squeeze_mask = var_2383_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2383_cast_fp16")]; tensor weights_245_axes_0 = const()[name = tensor("weights_245_axes_0"), val = tensor([1])]; - tensor weights_245_cast_fp16 = expand_dims(axes = weights_245_axes_0, x = var_2383_cast_fp16)[name = tensor("weights_245_cast_fp16")]; + tensor weights_245_cast_fp16 = expand_dims(axes = weights_245_axes_0, x = var_2383_cast_fp16)[name = tensor("weights_245_cast_fp16")]; tensor var_2387_axes_0 = const()[name = tensor("op_2387_axes_0"), val = tensor([2])]; tensor var_2387_keep_dims_0 = const()[name = tensor("op_2387_keep_dims_0"), val = tensor(false)]; tensor var_2387_cast_fp16 = reduce_sum(axes = var_2387_axes_0, keep_dims = var_2387_keep_dims_0, x = weights_245_cast_fp16)[name = tensor("op_2387_cast_fp16")]; tensor var_2388_to_fp16 = const()[name = tensor("op_2388_to_fp16"), val = tensor(0x1p-24)]; tensor v1_121_cast_fp16 = add(x = var_2387_cast_fp16, y = var_2388_to_fp16)[name = tensor("v1_121_cast_fp16")]; - tensor var_2390_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_245_cast_fp16)[name = tensor("op_2390_cast_fp16")]; + tensor var_2390_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_245_cast_fp16)[name = tensor("op_2390_cast_fp16")]; tensor var_2392_axes_0 = const()[name = tensor("op_2392_axes_0"), val = tensor([2])]; tensor var_2392_keep_dims_0 = const()[name = tensor("op_2392_keep_dims_0"), val = tensor(false)]; tensor var_2392_cast_fp16 = reduce_sum(axes = var_2392_axes_0, keep_dims = var_2392_keep_dims_0, x = var_2390_cast_fp16)[name = tensor("op_2392_cast_fp16")]; tensor mean_121_cast_fp16 = real_div(x = var_2392_cast_fp16, y = v1_121_cast_fp16)[name = tensor("mean_121_cast_fp16")]; tensor var_2394_axes_0 = const()[name = tensor("op_2394_axes_0"), val = tensor([2])]; tensor var_2394_cast_fp16 = expand_dims(axes = var_2394_axes_0, x = mean_121_cast_fp16)[name = tensor("op_2394_cast_fp16")]; - tensor var_2395_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2394_cast_fp16)[name = tensor("op_2395_cast_fp16")]; - tensor dx2_121_cast_fp16 = mul(x = var_2395_cast_fp16, y = var_2395_cast_fp16)[name = tensor("dx2_121_cast_fp16")]; - tensor var_2397_cast_fp16 = mul(x = weights_245_cast_fp16, y = weights_245_cast_fp16)[name = tensor("op_2397_cast_fp16")]; + tensor var_2395_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2394_cast_fp16)[name = tensor("op_2395_cast_fp16")]; + tensor dx2_121_cast_fp16 = mul(x = var_2395_cast_fp16, y = var_2395_cast_fp16)[name = tensor("dx2_121_cast_fp16")]; + tensor var_2397_cast_fp16 = mul(x = weights_245_cast_fp16, y = weights_245_cast_fp16)[name = tensor("op_2397_cast_fp16")]; tensor v2_121_axes_0 = const()[name = tensor("v2_121_axes_0"), val = tensor([2])]; tensor v2_121_keep_dims_0 = const()[name = tensor("v2_121_keep_dims_0"), val = tensor(false)]; tensor v2_121_cast_fp16 = reduce_sum(axes = v2_121_axes_0, keep_dims = v2_121_keep_dims_0, x = var_2397_cast_fp16)[name = tensor("v2_121_cast_fp16")]; - tensor var_2400_cast_fp16 = mul(x = dx2_121_cast_fp16, y = weights_245_cast_fp16)[name = tensor("op_2400_cast_fp16")]; + tensor var_2400_cast_fp16 = mul(x = dx2_121_cast_fp16, y = weights_245_cast_fp16)[name = tensor("op_2400_cast_fp16")]; tensor var_2402_axes_0 = const()[name = tensor("op_2402_axes_0"), val = tensor([2])]; tensor var_2402_keep_dims_0 = const()[name = tensor("op_2402_keep_dims_0"), val = tensor(false)]; tensor var_2402_cast_fp16 = reduce_sum(axes = var_2402_axes_0, keep_dims = var_2402_keep_dims_0, x = var_2400_cast_fp16)[name = tensor("op_2402_cast_fp16")]; @@ -2611,31 +2611,31 @@ program(1.0) tensor var_2410_interleave_0 = const()[name = tensor("op_2410_interleave_0"), val = tensor(false)]; tensor var_2410_cast_fp16 = concat(axis = var_72, interleave = var_2410_interleave_0, values = (mean_121_cast_fp16, std_121_cast_fp16))[name = tensor("op_2410_cast_fp16")]; tensor var_2412_begin_0 = const()[name = tensor("op_2412_begin_0"), val = tensor([0, 61, 0])]; - tensor var_2412_end_0 = const()[name = tensor("op_2412_end_0"), val = tensor([1, 62, 125])]; + tensor var_2412_end_0 = const()[name = tensor("op_2412_end_0"), val = tensor([1, 62, 375])]; tensor var_2412_end_mask_0 = const()[name = tensor("op_2412_end_mask_0"), val = tensor([true, false, true])]; tensor var_2412_squeeze_mask_0 = const()[name = tensor("op_2412_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2412_cast_fp16 = slice_by_index(begin = var_2412_begin_0, end = var_2412_end_0, end_mask = var_2412_end_mask_0, squeeze_mask = var_2412_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2412_cast_fp16")]; + tensor var_2412_cast_fp16 = slice_by_index(begin = var_2412_begin_0, end = var_2412_end_0, end_mask = var_2412_end_mask_0, squeeze_mask = var_2412_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2412_cast_fp16")]; tensor weights_249_axes_0 = const()[name = tensor("weights_249_axes_0"), val = tensor([1])]; - tensor weights_249_cast_fp16 = expand_dims(axes = weights_249_axes_0, x = var_2412_cast_fp16)[name = tensor("weights_249_cast_fp16")]; + tensor weights_249_cast_fp16 = expand_dims(axes = weights_249_axes_0, x = var_2412_cast_fp16)[name = tensor("weights_249_cast_fp16")]; tensor var_2416_axes_0 = const()[name = tensor("op_2416_axes_0"), val = tensor([2])]; tensor var_2416_keep_dims_0 = const()[name = tensor("op_2416_keep_dims_0"), val = tensor(false)]; tensor var_2416_cast_fp16 = reduce_sum(axes = var_2416_axes_0, keep_dims = var_2416_keep_dims_0, x = weights_249_cast_fp16)[name = tensor("op_2416_cast_fp16")]; tensor var_2417_to_fp16 = const()[name = tensor("op_2417_to_fp16"), val = tensor(0x1p-24)]; tensor v1_123_cast_fp16 = add(x = var_2416_cast_fp16, y = var_2417_to_fp16)[name = tensor("v1_123_cast_fp16")]; - tensor var_2419_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_249_cast_fp16)[name = tensor("op_2419_cast_fp16")]; + tensor var_2419_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_249_cast_fp16)[name = tensor("op_2419_cast_fp16")]; tensor var_2421_axes_0 = const()[name = tensor("op_2421_axes_0"), val = tensor([2])]; tensor var_2421_keep_dims_0 = const()[name = tensor("op_2421_keep_dims_0"), val = tensor(false)]; tensor var_2421_cast_fp16 = reduce_sum(axes = var_2421_axes_0, keep_dims = var_2421_keep_dims_0, x = var_2419_cast_fp16)[name = tensor("op_2421_cast_fp16")]; tensor mean_123_cast_fp16 = real_div(x = var_2421_cast_fp16, y = v1_123_cast_fp16)[name = tensor("mean_123_cast_fp16")]; tensor var_2423_axes_0 = const()[name = tensor("op_2423_axes_0"), val = tensor([2])]; tensor var_2423_cast_fp16 = expand_dims(axes = var_2423_axes_0, x = mean_123_cast_fp16)[name = tensor("op_2423_cast_fp16")]; - tensor var_2424_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2423_cast_fp16)[name = tensor("op_2424_cast_fp16")]; - tensor dx2_123_cast_fp16 = mul(x = var_2424_cast_fp16, y = var_2424_cast_fp16)[name = tensor("dx2_123_cast_fp16")]; - tensor var_2426_cast_fp16 = mul(x = weights_249_cast_fp16, y = weights_249_cast_fp16)[name = tensor("op_2426_cast_fp16")]; + tensor var_2424_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2423_cast_fp16)[name = tensor("op_2424_cast_fp16")]; + tensor dx2_123_cast_fp16 = mul(x = var_2424_cast_fp16, y = var_2424_cast_fp16)[name = tensor("dx2_123_cast_fp16")]; + tensor var_2426_cast_fp16 = mul(x = weights_249_cast_fp16, y = weights_249_cast_fp16)[name = tensor("op_2426_cast_fp16")]; tensor v2_123_axes_0 = const()[name = tensor("v2_123_axes_0"), val = tensor([2])]; tensor v2_123_keep_dims_0 = const()[name = tensor("v2_123_keep_dims_0"), val = tensor(false)]; tensor v2_123_cast_fp16 = reduce_sum(axes = v2_123_axes_0, keep_dims = v2_123_keep_dims_0, x = var_2426_cast_fp16)[name = tensor("v2_123_cast_fp16")]; - tensor var_2429_cast_fp16 = mul(x = dx2_123_cast_fp16, y = weights_249_cast_fp16)[name = tensor("op_2429_cast_fp16")]; + tensor var_2429_cast_fp16 = mul(x = dx2_123_cast_fp16, y = weights_249_cast_fp16)[name = tensor("op_2429_cast_fp16")]; tensor var_2431_axes_0 = const()[name = tensor("op_2431_axes_0"), val = tensor([2])]; tensor var_2431_keep_dims_0 = const()[name = tensor("op_2431_keep_dims_0"), val = tensor(false)]; tensor var_2431_cast_fp16 = reduce_sum(axes = var_2431_axes_0, keep_dims = var_2431_keep_dims_0, x = var_2429_cast_fp16)[name = tensor("op_2431_cast_fp16")]; @@ -2648,31 +2648,31 @@ program(1.0) tensor var_2439_interleave_0 = const()[name = tensor("op_2439_interleave_0"), val = tensor(false)]; tensor var_2439_cast_fp16 = concat(axis = var_72, interleave = var_2439_interleave_0, values = (mean_123_cast_fp16, std_123_cast_fp16))[name = tensor("op_2439_cast_fp16")]; tensor var_2441_begin_0 = const()[name = tensor("op_2441_begin_0"), val = tensor([0, 62, 0])]; - tensor var_2441_end_0 = const()[name = tensor("op_2441_end_0"), val = tensor([1, 63, 125])]; + tensor var_2441_end_0 = const()[name = tensor("op_2441_end_0"), val = tensor([1, 63, 375])]; tensor var_2441_end_mask_0 = const()[name = tensor("op_2441_end_mask_0"), val = tensor([true, false, true])]; tensor var_2441_squeeze_mask_0 = const()[name = tensor("op_2441_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2441_cast_fp16 = slice_by_index(begin = var_2441_begin_0, end = var_2441_end_0, end_mask = var_2441_end_mask_0, squeeze_mask = var_2441_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2441_cast_fp16")]; + tensor var_2441_cast_fp16 = slice_by_index(begin = var_2441_begin_0, end = var_2441_end_0, end_mask = var_2441_end_mask_0, squeeze_mask = var_2441_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2441_cast_fp16")]; tensor weights_253_axes_0 = const()[name = tensor("weights_253_axes_0"), val = tensor([1])]; - tensor weights_253_cast_fp16 = expand_dims(axes = weights_253_axes_0, x = var_2441_cast_fp16)[name = tensor("weights_253_cast_fp16")]; + tensor weights_253_cast_fp16 = expand_dims(axes = weights_253_axes_0, x = var_2441_cast_fp16)[name = tensor("weights_253_cast_fp16")]; tensor var_2445_axes_0 = const()[name = tensor("op_2445_axes_0"), val = tensor([2])]; tensor var_2445_keep_dims_0 = const()[name = tensor("op_2445_keep_dims_0"), val = tensor(false)]; tensor var_2445_cast_fp16 = reduce_sum(axes = var_2445_axes_0, keep_dims = var_2445_keep_dims_0, x = weights_253_cast_fp16)[name = tensor("op_2445_cast_fp16")]; tensor var_2446_to_fp16 = const()[name = tensor("op_2446_to_fp16"), val = tensor(0x1p-24)]; tensor v1_125_cast_fp16 = add(x = var_2445_cast_fp16, y = var_2446_to_fp16)[name = tensor("v1_125_cast_fp16")]; - tensor var_2448_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_253_cast_fp16)[name = tensor("op_2448_cast_fp16")]; + tensor var_2448_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_253_cast_fp16)[name = tensor("op_2448_cast_fp16")]; tensor var_2450_axes_0 = const()[name = tensor("op_2450_axes_0"), val = tensor([2])]; tensor var_2450_keep_dims_0 = const()[name = tensor("op_2450_keep_dims_0"), val = tensor(false)]; tensor var_2450_cast_fp16 = reduce_sum(axes = var_2450_axes_0, keep_dims = var_2450_keep_dims_0, x = var_2448_cast_fp16)[name = tensor("op_2450_cast_fp16")]; tensor mean_125_cast_fp16 = real_div(x = var_2450_cast_fp16, y = v1_125_cast_fp16)[name = tensor("mean_125_cast_fp16")]; tensor var_2452_axes_0 = const()[name = tensor("op_2452_axes_0"), val = tensor([2])]; tensor var_2452_cast_fp16 = expand_dims(axes = var_2452_axes_0, x = mean_125_cast_fp16)[name = tensor("op_2452_cast_fp16")]; - tensor var_2453_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2452_cast_fp16)[name = tensor("op_2453_cast_fp16")]; - tensor dx2_125_cast_fp16 = mul(x = var_2453_cast_fp16, y = var_2453_cast_fp16)[name = tensor("dx2_125_cast_fp16")]; - tensor var_2455_cast_fp16 = mul(x = weights_253_cast_fp16, y = weights_253_cast_fp16)[name = tensor("op_2455_cast_fp16")]; + tensor var_2453_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2452_cast_fp16)[name = tensor("op_2453_cast_fp16")]; + tensor dx2_125_cast_fp16 = mul(x = var_2453_cast_fp16, y = var_2453_cast_fp16)[name = tensor("dx2_125_cast_fp16")]; + tensor var_2455_cast_fp16 = mul(x = weights_253_cast_fp16, y = weights_253_cast_fp16)[name = tensor("op_2455_cast_fp16")]; tensor v2_125_axes_0 = const()[name = tensor("v2_125_axes_0"), val = tensor([2])]; tensor v2_125_keep_dims_0 = const()[name = tensor("v2_125_keep_dims_0"), val = tensor(false)]; tensor v2_125_cast_fp16 = reduce_sum(axes = v2_125_axes_0, keep_dims = v2_125_keep_dims_0, x = var_2455_cast_fp16)[name = tensor("v2_125_cast_fp16")]; - tensor var_2458_cast_fp16 = mul(x = dx2_125_cast_fp16, y = weights_253_cast_fp16)[name = tensor("op_2458_cast_fp16")]; + tensor var_2458_cast_fp16 = mul(x = dx2_125_cast_fp16, y = weights_253_cast_fp16)[name = tensor("op_2458_cast_fp16")]; tensor var_2460_axes_0 = const()[name = tensor("op_2460_axes_0"), val = tensor([2])]; tensor var_2460_keep_dims_0 = const()[name = tensor("op_2460_keep_dims_0"), val = tensor(false)]; tensor var_2460_cast_fp16 = reduce_sum(axes = var_2460_axes_0, keep_dims = var_2460_keep_dims_0, x = var_2458_cast_fp16)[name = tensor("op_2460_cast_fp16")]; @@ -2685,31 +2685,31 @@ program(1.0) tensor var_2468_interleave_0 = const()[name = tensor("op_2468_interleave_0"), val = tensor(false)]; tensor var_2468_cast_fp16 = concat(axis = var_72, interleave = var_2468_interleave_0, values = (mean_125_cast_fp16, std_125_cast_fp16))[name = tensor("op_2468_cast_fp16")]; tensor var_2470_begin_0 = const()[name = tensor("op_2470_begin_0"), val = tensor([0, 63, 0])]; - tensor var_2470_end_0 = const()[name = tensor("op_2470_end_0"), val = tensor([1, 64, 125])]; + tensor var_2470_end_0 = const()[name = tensor("op_2470_end_0"), val = tensor([1, 64, 375])]; tensor var_2470_end_mask_0 = const()[name = tensor("op_2470_end_mask_0"), val = tensor([true, false, true])]; tensor var_2470_squeeze_mask_0 = const()[name = tensor("op_2470_squeeze_mask_0"), val = tensor([false, true, false])]; - tensor var_2470_cast_fp16 = slice_by_index(begin = var_2470_begin_0, end = var_2470_end_0, end_mask = var_2470_end_mask_0, squeeze_mask = var_2470_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2470_cast_fp16")]; + tensor var_2470_cast_fp16 = slice_by_index(begin = var_2470_begin_0, end = var_2470_end_0, end_mask = var_2470_end_mask_0, squeeze_mask = var_2470_squeeze_mask_0, x = weights_1_cast_fp16)[name = tensor("op_2470_cast_fp16")]; tensor weights_axes_0 = const()[name = tensor("weights_axes_0"), val = tensor([1])]; - tensor weights_cast_fp16 = expand_dims(axes = weights_axes_0, x = var_2470_cast_fp16)[name = tensor("weights_cast_fp16")]; + tensor weights_cast_fp16 = expand_dims(axes = weights_axes_0, x = var_2470_cast_fp16)[name = tensor("weights_cast_fp16")]; tensor var_2474_axes_0 = const()[name = tensor("op_2474_axes_0"), val = tensor([2])]; tensor var_2474_keep_dims_0 = const()[name = tensor("op_2474_keep_dims_0"), val = tensor(false)]; tensor var_2474_cast_fp16 = reduce_sum(axes = var_2474_axes_0, keep_dims = var_2474_keep_dims_0, x = weights_cast_fp16)[name = tensor("op_2474_cast_fp16")]; tensor var_2475_to_fp16 = const()[name = tensor("op_2475_to_fp16"), val = tensor(0x1p-24)]; tensor v1_cast_fp16 = add(x = var_2474_cast_fp16, y = var_2475_to_fp16)[name = tensor("v1_cast_fp16")]; - tensor var_2477_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_cast_fp16)[name = tensor("op_2477_cast_fp16")]; + tensor var_2477_cast_fp16 = mul(x = sequences_cast_fp16, y = weights_cast_fp16)[name = tensor("op_2477_cast_fp16")]; tensor var_2479_axes_0 = const()[name = tensor("op_2479_axes_0"), val = tensor([2])]; tensor var_2479_keep_dims_0 = const()[name = tensor("op_2479_keep_dims_0"), val = tensor(false)]; tensor var_2479_cast_fp16 = reduce_sum(axes = var_2479_axes_0, keep_dims = var_2479_keep_dims_0, x = var_2477_cast_fp16)[name = tensor("op_2479_cast_fp16")]; tensor mean_cast_fp16 = real_div(x = var_2479_cast_fp16, y = v1_cast_fp16)[name = tensor("mean_cast_fp16")]; tensor var_2481_axes_0 = const()[name = tensor("op_2481_axes_0"), val = tensor([2])]; tensor var_2481_cast_fp16 = expand_dims(axes = var_2481_axes_0, x = mean_cast_fp16)[name = tensor("op_2481_cast_fp16")]; - tensor var_2482_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2481_cast_fp16)[name = tensor("op_2482_cast_fp16")]; - tensor dx2_cast_fp16 = mul(x = var_2482_cast_fp16, y = var_2482_cast_fp16)[name = tensor("dx2_cast_fp16")]; - tensor var_2484_cast_fp16 = mul(x = weights_cast_fp16, y = weights_cast_fp16)[name = tensor("op_2484_cast_fp16")]; + tensor var_2482_cast_fp16 = sub(x = sequences_cast_fp16, y = var_2481_cast_fp16)[name = tensor("op_2482_cast_fp16")]; + tensor dx2_cast_fp16 = mul(x = var_2482_cast_fp16, y = var_2482_cast_fp16)[name = tensor("dx2_cast_fp16")]; + tensor var_2484_cast_fp16 = mul(x = weights_cast_fp16, y = weights_cast_fp16)[name = tensor("op_2484_cast_fp16")]; tensor v2_axes_0 = const()[name = tensor("v2_axes_0"), val = tensor([2])]; tensor v2_keep_dims_0 = const()[name = tensor("v2_keep_dims_0"), val = tensor(false)]; tensor v2_cast_fp16 = reduce_sum(axes = v2_axes_0, keep_dims = v2_keep_dims_0, x = var_2484_cast_fp16)[name = tensor("v2_cast_fp16")]; - tensor var_2487_cast_fp16 = mul(x = dx2_cast_fp16, y = weights_cast_fp16)[name = tensor("op_2487_cast_fp16")]; + tensor var_2487_cast_fp16 = mul(x = dx2_cast_fp16, y = weights_cast_fp16)[name = tensor("op_2487_cast_fp16")]; tensor var_2489_axes_0 = const()[name = tensor("op_2489_axes_0"), val = tensor([2])]; tensor var_2489_keep_dims_0 = const()[name = tensor("op_2489_keep_dims_0"), val = tensor(false)]; tensor var_2489_cast_fp16 = reduce_sum(axes = var_2489_axes_0, keep_dims = var_2489_keep_dims_0, x = var_2487_cast_fp16)[name = tensor("op_2489_cast_fp16")];