| program(1.3) | |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] | |
| { | |
| func main<ios18>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, state<tensor<fp16, [2, 1280, 1, 1536]>> encoder_attn_key_cache, state<tensor<fp16, [1, 1536]>> encoder_attn_key_padding_mask, state<tensor<fp16, [2, 1280, 1, 1536]>> encoder_attn_value_cache, tensor<int32, [1]> input_ids, tensor<fp16, [1, 448]> kv_cache_update_mask, state<tensor<fp16, [2, 1280, 1, 448]>> self_attn_key_cache, state<tensor<fp16, [2, 1280, 1, 448]>> self_attn_value_cache) { | |
| int32 var_22_axis_0 = const()[name = string("op_22_axis_0"), val = int32(0)]; | |
| int32 var_22_batch_dims_0 = const()[name = string("op_22_batch_dims_0"), val = int32(0)]; | |
| bool var_22_validate_indices_0 = const()[name = string("op_22_validate_indices_0"), val = bool(false)]; | |
| tensor<fp16, [51866, 1280]> embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor<fp16, [51866, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; | |
| tensor<fp16, [1, 1280]> var_22_cast_fp16 = gather(axis = var_22_axis_0, batch_dims = var_22_batch_dims_0, indices = input_ids, validate_indices = var_22_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_22_cast_fp16")]; | |
| int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; | |
| int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; | |
| bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; | |
| tensor<fp16, [448, 1280]> embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor<fp16, [448, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132777088)))]; | |
| string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; | |
| tensor<uint16, [1]> cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_43")]; | |
| tensor<fp16, [1, 1280]> var_26_cast_fp16_cast_uint16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_26_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_26_cast_fp16_cast_uint16")]; | |
| tensor<fp16, [1, 1280]> hidden_states_1_cast_fp16 = add(x = var_22_cast_fp16, y = var_26_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; | |
| tensor<int32, [1]> var_40_axes_0 = const()[name = string("op_40_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1280, 1]> var_40_cast_fp16 = expand_dims(axes = var_40_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_40_cast_fp16")]; | |
| tensor<int32, [1]> inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor<int32, [1]>([3])]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_40_cast_fp16)[name = string("inputs_1_cast_fp16")]; | |
| tensor<fp16, [2, 1280, 1, 448]> read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; | |
| tensor<int32, [2]> tile_0 = const()[name = string("tile_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 var_45_axis_0 = const()[name = string("op_45_axis_0"), val = int32(0)]; | |
| tensor<fp16, [1, 1280, 1, 448]> var_45_cast_fp16_0, tensor<fp16, [1, 1280, 1, 448]> var_45_cast_fp16_1 = split(axis = var_45_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_45_cast_fp16")]; | |
| tensor<fp16, [2, 1280, 1, 448]> read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; | |
| tensor<int32, [2]> tile_1 = const()[name = string("tile_1"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 var_50_axis_0 = const()[name = string("op_50_axis_0"), val = int32(0)]; | |
| tensor<fp16, [1, 1280, 1, 448]> var_50_cast_fp16_0, tensor<fp16, [1, 1280, 1, 448]> var_50_cast_fp16_1 = split(axis = var_50_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_50_cast_fp16")]; | |
| tensor<fp16, [2, 1280, 1, 1536]> read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; | |
| tensor<int32, [4]> obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor<int32, [4]>([1, 1280, 1, 1536])]; | |
| tensor<bool, [4]> obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<fp16, [1, 1280, 1, 1536]> obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; | |
| tensor<fp16, [2, 1280, 1, 1536]> read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; | |
| tensor<int32, [4]> obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor<int32, [4]>([1, 1280, 1, 1536])]; | |
| tensor<bool, [4]> obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<fp16, [1, 1280, 1, 1536]> obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; | |
| int32 var_68 = const()[name = string("op_68"), val = int32(3)]; | |
| tensor<int32, [1]> out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_93_to_fp16 = const()[name = string("op_93_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_93_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; | |
| tensor<fp16, [1280]> obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133924032)))]; | |
| tensor<fp16, [1280]> obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133926656)))]; | |
| tensor<fp16, [1280]> obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133929280)))]; | |
| tensor<fp16, [1280]> obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133931904)))]; | |
| fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; | |
| string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133934528)))]; | |
| tensor<fp16, [1280]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137211392)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; | |
| string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137214016)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; | |
| string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140490880)))]; | |
| tensor<fp16, [1280]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143767744)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; | |
| tensor<int32, [1]> var_128_axes_0 = const()[name = string("op_128_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1, 448]> var_128_cast_fp16 = expand_dims(axes = var_128_axes_0, x = kv_cache_update_mask)[name = string("op_128_cast_fp16")]; | |
| tensor<int32, [1]> var_129_axes_0 = const()[name = string("op_129_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1, 1, 448]> var_129_cast_fp16 = expand_dims(axes = var_129_axes_0, x = var_128_cast_fp16)[name = string("op_129_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> var_131_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_129_cast_fp16)[name = string("op_131_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> key_1_cast_fp16 = add(x = var_45_cast_fp16_0, y = var_131_cast_fp16)[name = string("key_1_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> var_133_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_129_cast_fp16)[name = string("op_133_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> value_1_cast_fp16 = add(x = var_50_cast_fp16_0, y = var_133_cast_fp16)[name = string("value_1_cast_fp16")]; | |
| tensor<int32, [4]> var_136 = const()[name = string("op_136"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1]> mh_q_1_cast_fp16 = reshape(shape = var_136, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; | |
| fp16 var_138_to_fp16 = const()[name = string("op_138_to_fp16"), val = fp16(0x1p-3)]; | |
| tensor<fp16, [1, 20, 64, 1]> var_139_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_138_to_fp16)[name = string("op_139_cast_fp16")]; | |
| tensor<int32, [4]> var_140 = const()[name = string("op_140"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 448]> var_141_cast_fp16 = reshape(shape = var_140, x = key_1_cast_fp16)[name = string("op_141_cast_fp16")]; | |
| bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; | |
| bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1, 448]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_139_cast_fp16, y = var_141_cast_fp16)[name = string("mh_w_1_cast_fp16")]; | |
| tensor<int32, [1]> var_145_axes_0 = const()[name = string("op_145_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1, 448]> var_145_cast_fp16 = expand_dims(axes = var_145_axes_0, x = decoder_key_padding_mask)[name = string("op_145_cast_fp16")]; | |
| tensor<int32, [1]> var_146_axes_0 = const()[name = string("op_146_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1, 1, 448]> var_146_cast_fp16 = expand_dims(axes = var_146_axes_0, x = var_145_cast_fp16)[name = string("op_146_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_146_cast_fp16)[name = string("mh_w_3_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 448]> var_149_cast_fp16 = softmax(axis = var_68, x = mh_w_3_cast_fp16)[name = string("op_149_cast_fp16")]; | |
| tensor<int32, [4]> var_150 = const()[name = string("op_150"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 448]> var_151_cast_fp16 = reshape(shape = var_150, x = value_1_cast_fp16)[name = string("op_151_cast_fp16")]; | |
| bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; | |
| bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; | |
| tensor<fp16, [1, 20, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_151_cast_fp16, y = var_149_cast_fp16)[name = string("attn_1_cast_fp16")]; | |
| tensor<int32, [4]> var_154 = const()[name = string("op_154"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; | |
| tensor<fp16, [1, 1280, 1, 1]> input_1_cast_fp16 = reshape(shape = var_154, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; | |
| string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143770368)))]; | |
| tensor<fp16, [1280]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147047232)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; | |
| tensor<int32, [1]> out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_176_to_fp16 = const()[name = string("op_176_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_176_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; | |
| tensor<fp16, [1280]> obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147049856)))]; | |
| tensor<fp16, [1280]> obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147052480)))]; | |
| fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; | |
| string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147055104)))]; | |
| tensor<fp16, [1280]> layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331968)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; | |
| tensor<int32, [4]> var_196 = const()[name = string("op_196"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1]> mh_q_3_cast_fp16 = reshape(shape = var_196, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; | |
| fp16 var_198_to_fp16 = const()[name = string("op_198_to_fp16"), val = fp16(0x1p-3)]; | |
| tensor<fp16, [1, 20, 64, 1]> var_199_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_198_to_fp16)[name = string("op_199_cast_fp16")]; | |
| tensor<int32, [4]> var_200 = const()[name = string("op_200"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1536]> var_201_cast_fp16 = reshape(shape = var_200, x = obj_17_cast_fp16)[name = string("op_201_cast_fp16")]; | |
| bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; | |
| bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1, 1536]> mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_199_cast_fp16, y = var_201_cast_fp16)[name = string("mh_w_5_cast_fp16")]; | |
| tensor<fp16, [1, 1536]> read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; | |
| tensor<int32, [1]> var_205_axes_0 = const()[name = string("op_205_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1, 1536]> var_205_cast_fp16 = expand_dims(axes = var_205_axes_0, x = read_state_4)[name = string("op_205_cast_fp16")]; | |
| tensor<int32, [1]> var_206_axes_0 = const()[name = string("op_206_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_206_cast_fp16 = expand_dims(axes = var_206_axes_0, x = var_205_cast_fp16)[name = string("op_206_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 1536]> mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_206_cast_fp16)[name = string("mh_w_7_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 1536]> obj_23_cast_fp16 = softmax(axis = var_68, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; | |
| tensor<int32, [4]> var_210 = const()[name = string("op_210"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1536]> var_211_cast_fp16 = reshape(shape = var_210, x = obj_19_cast_fp16)[name = string("op_211_cast_fp16")]; | |
| bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; | |
| bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; | |
| tensor<fp16, [1, 20, 64, 1]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_211_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; | |
| tensor<int32, [4]> var_214 = const()[name = string("op_214"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; | |
| tensor<fp16, [1, 1280, 1, 1]> input_3_cast_fp16 = reshape(shape = var_214, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; | |
| string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334592)))]; | |
| tensor<fp16, [1280]> layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; | |
| tensor<int32, [1]> out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_232_to_fp16 = const()[name = string("op_232_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_232_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; | |
| tensor<fp16, [1280]> input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614080)))]; | |
| tensor<fp16, [1280]> input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616704)))]; | |
| fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; | |
| string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; | |
| tensor<fp16, [5120, 1280, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153619328)))]; | |
| tensor<fp16, [5120]> layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166726592)))]; | |
| tensor<fp16, [1, 5120, 1, 1]> input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; | |
| string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 5120, 1, 1]> input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; | |
| string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 5120, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166736896)))]; | |
| tensor<fp16, [1280]> layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179844160)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; | |
| tensor<int32, [4]> obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1536])]; | |
| tensor<bool, [4]> obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<fp16, [1, 1280, 1, 1536]> obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; | |
| tensor<int32, [4]> obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])]; | |
| tensor<int32, [4]> obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1536])]; | |
| tensor<bool, [4]> obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])]; | |
| tensor<fp16, [1, 1280, 1, 1536]> obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; | |
| int32 var_277 = const()[name = string("op_277"), val = int32(3)]; | |
| tensor<int32, [1]> out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_302_to_fp16 = const()[name = string("op_302_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_302_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; | |
| tensor<fp16, [1280]> obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179846784)))]; | |
| tensor<fp16, [1280]> obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179849408)))]; | |
| fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; | |
| string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179852032)))]; | |
| tensor<fp16, [1280]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128896)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; | |
| string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183131520)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_cast_fp16")]; | |
| string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186408384)))]; | |
| tensor<fp16, [1280]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189685248)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> current_value_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> var_340_cast_fp16 = mul(x = current_key_cast_fp16, y = var_129_cast_fp16)[name = string("op_340_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> key_cast_fp16 = add(x = var_45_cast_fp16_1, y = var_340_cast_fp16)[name = string("key_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> var_342_cast_fp16 = mul(x = current_value_cast_fp16, y = var_129_cast_fp16)[name = string("op_342_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 448]> value_cast_fp16 = add(x = var_50_cast_fp16_1, y = var_342_cast_fp16)[name = string("value_cast_fp16")]; | |
| tensor<int32, [4]> var_345 = const()[name = string("op_345"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1]> mh_q_5_cast_fp16 = reshape(shape = var_345, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; | |
| fp16 var_347_to_fp16 = const()[name = string("op_347_to_fp16"), val = fp16(0x1p-3)]; | |
| tensor<fp16, [1, 20, 64, 1]> var_348_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_347_to_fp16)[name = string("op_348_cast_fp16")]; | |
| tensor<int32, [4]> var_349 = const()[name = string("op_349"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 448]> var_350_cast_fp16 = reshape(shape = var_349, x = key_cast_fp16)[name = string("op_350_cast_fp16")]; | |
| bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; | |
| bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1, 448]> mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_348_cast_fp16, y = var_350_cast_fp16)[name = string("mh_w_9_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 448]> mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_146_cast_fp16)[name = string("mh_w_11_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 448]> var_358_cast_fp16 = softmax(axis = var_277, x = mh_w_11_cast_fp16)[name = string("op_358_cast_fp16")]; | |
| tensor<int32, [4]> var_359 = const()[name = string("op_359"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 448]> var_360_cast_fp16 = reshape(shape = var_359, x = value_cast_fp16)[name = string("op_360_cast_fp16")]; | |
| bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; | |
| bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; | |
| tensor<fp16, [1, 20, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_360_cast_fp16, y = var_358_cast_fp16)[name = string("attn_5_cast_fp16")]; | |
| tensor<int32, [4]> var_363 = const()[name = string("op_363"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; | |
| tensor<fp16, [1, 1280, 1, 1]> input_11_cast_fp16 = reshape(shape = var_363, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; | |
| string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189687872)))]; | |
| tensor<fp16, [1280]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192964736)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; | |
| tensor<int32, [1]> out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_385_to_fp16 = const()[name = string("op_385_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_385_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; | |
| tensor<fp16, [1280]> obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192967360)))]; | |
| tensor<fp16, [1280]> obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192969984)))]; | |
| fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; | |
| string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> query_strides_0 = const()[name = string("query_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_pad_0 = const()[name = string("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192972608)))]; | |
| tensor<fp16, [1280]> layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196249472)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> query_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_cast_fp16")]; | |
| tensor<int32, [4]> var_405 = const()[name = string("op_405"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1]> mh_q_cast_fp16 = reshape(shape = var_405, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; | |
| fp16 var_407_to_fp16 = const()[name = string("op_407_to_fp16"), val = fp16(0x1p-3)]; | |
| tensor<fp16, [1, 20, 64, 1]> var_408_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_407_to_fp16)[name = string("op_408_cast_fp16")]; | |
| tensor<int32, [4]> var_409 = const()[name = string("op_409"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1536]> var_410_cast_fp16 = reshape(shape = var_409, x = obj_35_cast_fp16)[name = string("op_410_cast_fp16")]; | |
| bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; | |
| bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1, 1536]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_408_cast_fp16, y = var_410_cast_fp16)[name = string("mh_w_13_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 1536]> mh_w_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_206_cast_fp16)[name = string("mh_w_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1, 1536]> obj_41_cast_fp16 = softmax(axis = var_277, x = mh_w_cast_fp16)[name = string("obj_41_cast_fp16")]; | |
| tensor<int32, [4]> var_419 = const()[name = string("op_419"), val = tensor<int32, [4]>([1, 20, 64, -1])]; | |
| tensor<fp16, [1, 20, 64, 1536]> var_420_cast_fp16 = reshape(shape = var_419, x = obj_37_cast_fp16)[name = string("op_420_cast_fp16")]; | |
| bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; | |
| bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; | |
| tensor<fp16, [1, 20, 64, 1]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_420_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_cast_fp16")]; | |
| tensor<int32, [4]> var_423 = const()[name = string("op_423"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; | |
| tensor<fp16, [1, 1280, 1, 1]> input_13_cast_fp16 = reshape(shape = var_423, x = attn_cast_fp16)[name = string("input_13_cast_fp16")]; | |
| string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 1280, 1, 1]> layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196252096)))]; | |
| tensor<fp16, [1280]> layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199528960)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; | |
| tensor<int32, [1]> out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_444_to_fp16 = const()[name = string("op_444_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_444_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; | |
| tensor<fp16, [1280]> input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199531584)))]; | |
| tensor<fp16, [1280]> input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199534208)))]; | |
| fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; | |
| string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; | |
| tensor<fp16, [5120, 1280, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199536832)))]; | |
| tensor<fp16, [5120]> layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212644096)))]; | |
| tensor<fp16, [1, 5120, 1, 1]> input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; | |
| string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 5120, 1, 1]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_17_cast_fp16)[name = string("input_cast_fp16")]; | |
| string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; | |
| tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; | |
| tensor<fp16, [1280, 5120, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212654400)))]; | |
| tensor<fp16, [1280]> layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225761664)))]; | |
| tensor<fp16, [1, 1280, 1, 1]> hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; | |
| tensor<fp16, [1, 1280, 1, 1]> inputs_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_cast_fp16")]; | |
| tensor<int32, [1]> out_axes_0 = const()[name = string("out_axes_0"), val = tensor<int32, [1]>([1])]; | |
| fp16 var_487_to_fp16 = const()[name = string("op_487_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_487_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; | |
| tensor<fp16, [1280]> hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225764288)))]; | |
| tensor<fp16, [1280]> hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225766912)))]; | |
| fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1280, 1, 1]> hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; | |
| tensor<int32, [1]> var_498_axes_0 = const()[name = string("op_498_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1280, 1]> var_498_cast_fp16 = squeeze(axes = var_498_axes_0, x = hidden_states_cast_fp16)[name = string("op_498_cast_fp16")]; | |
| tensor<int32, [3]> var_501_perm_0 = const()[name = string("op_501_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [51866]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [51866]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225769536)))]; | |
| tensor<fp16, [1, 1, 1280]> var_501_cast_fp16 = transpose(perm = var_501_perm_0, x = var_498_cast_fp16)[name = string("transpose_0")]; | |
| tensor<fp16, [1, 1, 51866]> logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_501_cast_fp16)[name = string("linear_0_cast_fp16")]; | |
| int32 var_505 = const()[name = string("op_505"), val = int32(1)]; | |
| bool obj_45_interleave_0 = const()[name = string("obj_45_interleave_0"), val = bool(false)]; | |
| tensor<fp16, [1, 2560, 1, 1]> key_cache_updates = concat(axis = var_505, interleave = obj_45_interleave_0, values = (current_key_1_cast_fp16, current_key_cast_fp16))[name = string("obj_45_cast_fp16")]; | |
| int32 var_508 = const()[name = string("op_508"), val = int32(1)]; | |
| bool obj_47_interleave_0 = const()[name = string("obj_47_interleave_0"), val = bool(false)]; | |
| tensor<fp16, [1, 2560, 1, 1]> value_cache_updates = concat(axis = var_508, interleave = obj_47_interleave_0, values = (current_value_1_cast_fp16, current_value_cast_fp16))[name = string("obj_47_cast_fp16")]; | |
| tensor<int32, [4]> var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = obj_41_cast_fp16)[name = string("op_519_cast_fp16")]; | |
| tensor<int32, [4]> var_522_begin_0 = const()[name = string("op_522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_522_end_0 = const()[name = string("op_522_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_522_end_mask_0 = const()[name = string("op_522_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_522_squeeze_mask_0 = const()[name = string("op_522_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_522_cast_fp16 = slice_by_index(begin = var_522_begin_0, end = var_522_end_0, end_mask = var_522_end_mask_0, squeeze_mask = var_522_squeeze_mask_0, x = var_519_cast_fp16)[name = string("op_522_cast_fp16")]; | |
| tensor<int32, [4]> var_537_begin_0 = const()[name = string("op_537_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])]; | |
| tensor<int32, [4]> var_537_end_0 = const()[name = string("op_537_end_0"), val = tensor<int32, [4]>([1, 2, 1, 1536])]; | |
| tensor<bool, [4]> var_537_end_mask_0 = const()[name = string("op_537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_537_cast_fp16 = slice_by_index(begin = var_537_begin_0, end = var_537_end_0, end_mask = var_537_end_mask_0, x = obj_41_cast_fp16)[name = string("op_537_cast_fp16")]; | |
| tensor<int32, [4]> var_540_begin_0 = const()[name = string("op_540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_540_end_0 = const()[name = string("op_540_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_540_end_mask_0 = const()[name = string("op_540_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_540_squeeze_mask_0 = const()[name = string("op_540_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_540_cast_fp16 = slice_by_index(begin = var_540_begin_0, end = var_540_end_0, end_mask = var_540_end_mask_0, squeeze_mask = var_540_squeeze_mask_0, x = var_537_cast_fp16)[name = string("op_540_cast_fp16")]; | |
| tensor<int32, [4]> var_555_begin_0 = const()[name = string("op_555_begin_0"), val = tensor<int32, [4]>([0, 2, 0, 0])]; | |
| tensor<int32, [4]> var_555_end_0 = const()[name = string("op_555_end_0"), val = tensor<int32, [4]>([1, 3, 1, 1536])]; | |
| tensor<bool, [4]> var_555_end_mask_0 = const()[name = string("op_555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_555_cast_fp16 = slice_by_index(begin = var_555_begin_0, end = var_555_end_0, end_mask = var_555_end_mask_0, x = obj_41_cast_fp16)[name = string("op_555_cast_fp16")]; | |
| tensor<int32, [4]> var_558_begin_0 = const()[name = string("op_558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_558_end_0 = const()[name = string("op_558_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_558_end_mask_0 = const()[name = string("op_558_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_558_squeeze_mask_0 = const()[name = string("op_558_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_558_cast_fp16 = slice_by_index(begin = var_558_begin_0, end = var_558_end_0, end_mask = var_558_end_mask_0, squeeze_mask = var_558_squeeze_mask_0, x = var_555_cast_fp16)[name = string("op_558_cast_fp16")]; | |
| tensor<int32, [4]> var_573_begin_0 = const()[name = string("op_573_begin_0"), val = tensor<int32, [4]>([0, 3, 0, 0])]; | |
| tensor<int32, [4]> var_573_end_0 = const()[name = string("op_573_end_0"), val = tensor<int32, [4]>([1, 4, 1, 1536])]; | |
| tensor<bool, [4]> var_573_end_mask_0 = const()[name = string("op_573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_573_cast_fp16 = slice_by_index(begin = var_573_begin_0, end = var_573_end_0, end_mask = var_573_end_mask_0, x = obj_41_cast_fp16)[name = string("op_573_cast_fp16")]; | |
| tensor<int32, [4]> var_576_begin_0 = const()[name = string("op_576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_576_end_0 = const()[name = string("op_576_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_576_end_mask_0 = const()[name = string("op_576_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_576_squeeze_mask_0 = const()[name = string("op_576_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_576_cast_fp16 = slice_by_index(begin = var_576_begin_0, end = var_576_end_0, end_mask = var_576_end_mask_0, squeeze_mask = var_576_squeeze_mask_0, x = var_573_cast_fp16)[name = string("op_576_cast_fp16")]; | |
| tensor<int32, [4]> var_591_begin_0 = const()[name = string("op_591_begin_0"), val = tensor<int32, [4]>([0, 4, 0, 0])]; | |
| tensor<int32, [4]> var_591_end_0 = const()[name = string("op_591_end_0"), val = tensor<int32, [4]>([1, 5, 1, 1536])]; | |
| tensor<bool, [4]> var_591_end_mask_0 = const()[name = string("op_591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_591_cast_fp16 = slice_by_index(begin = var_591_begin_0, end = var_591_end_0, end_mask = var_591_end_mask_0, x = obj_41_cast_fp16)[name = string("op_591_cast_fp16")]; | |
| tensor<int32, [4]> var_594_begin_0 = const()[name = string("op_594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_594_end_0 = const()[name = string("op_594_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_594_end_mask_0 = const()[name = string("op_594_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_594_squeeze_mask_0 = const()[name = string("op_594_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_594_cast_fp16 = slice_by_index(begin = var_594_begin_0, end = var_594_end_0, end_mask = var_594_end_mask_0, squeeze_mask = var_594_squeeze_mask_0, x = var_591_cast_fp16)[name = string("op_594_cast_fp16")]; | |
| tensor<int32, [4]> var_609_begin_0 = const()[name = string("op_609_begin_0"), val = tensor<int32, [4]>([0, 5, 0, 0])]; | |
| tensor<int32, [4]> var_609_end_0 = const()[name = string("op_609_end_0"), val = tensor<int32, [4]>([1, 6, 1, 1536])]; | |
| tensor<bool, [4]> var_609_end_mask_0 = const()[name = string("op_609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_609_cast_fp16 = slice_by_index(begin = var_609_begin_0, end = var_609_end_0, end_mask = var_609_end_mask_0, x = obj_41_cast_fp16)[name = string("op_609_cast_fp16")]; | |
| tensor<int32, [4]> var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_612_squeeze_mask_0 = const()[name = string("op_612_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, squeeze_mask = var_612_squeeze_mask_0, x = var_609_cast_fp16)[name = string("op_612_cast_fp16")]; | |
| tensor<int32, [4]> var_627_begin_0 = const()[name = string("op_627_begin_0"), val = tensor<int32, [4]>([0, 6, 0, 0])]; | |
| tensor<int32, [4]> var_627_end_0 = const()[name = string("op_627_end_0"), val = tensor<int32, [4]>([1, 7, 1, 1536])]; | |
| tensor<bool, [4]> var_627_end_mask_0 = const()[name = string("op_627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_627_cast_fp16 = slice_by_index(begin = var_627_begin_0, end = var_627_end_0, end_mask = var_627_end_mask_0, x = obj_41_cast_fp16)[name = string("op_627_cast_fp16")]; | |
| tensor<int32, [4]> var_630_begin_0 = const()[name = string("op_630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_630_end_0 = const()[name = string("op_630_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_630_end_mask_0 = const()[name = string("op_630_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_630_squeeze_mask_0 = const()[name = string("op_630_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_630_cast_fp16 = slice_by_index(begin = var_630_begin_0, end = var_630_end_0, end_mask = var_630_end_mask_0, squeeze_mask = var_630_squeeze_mask_0, x = var_627_cast_fp16)[name = string("op_630_cast_fp16")]; | |
| tensor<int32, [4]> var_645_begin_0 = const()[name = string("op_645_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; | |
| tensor<int32, [4]> var_645_end_0 = const()[name = string("op_645_end_0"), val = tensor<int32, [4]>([1, 8, 1, 1536])]; | |
| tensor<bool, [4]> var_645_end_mask_0 = const()[name = string("op_645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = obj_41_cast_fp16)[name = string("op_645_cast_fp16")]; | |
| tensor<int32, [4]> var_648_begin_0 = const()[name = string("op_648_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_648_end_0 = const()[name = string("op_648_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_648_end_mask_0 = const()[name = string("op_648_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_648_squeeze_mask_0 = const()[name = string("op_648_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_648_cast_fp16 = slice_by_index(begin = var_648_begin_0, end = var_648_end_0, end_mask = var_648_end_mask_0, squeeze_mask = var_648_squeeze_mask_0, x = var_645_cast_fp16)[name = string("op_648_cast_fp16")]; | |
| tensor<int32, [4]> var_663_begin_0 = const()[name = string("op_663_begin_0"), val = tensor<int32, [4]>([0, 8, 0, 0])]; | |
| tensor<int32, [4]> var_663_end_0 = const()[name = string("op_663_end_0"), val = tensor<int32, [4]>([1, 9, 1, 1536])]; | |
| tensor<bool, [4]> var_663_end_mask_0 = const()[name = string("op_663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = obj_41_cast_fp16)[name = string("op_663_cast_fp16")]; | |
| tensor<int32, [4]> var_666_begin_0 = const()[name = string("op_666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_666_end_0 = const()[name = string("op_666_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_666_end_mask_0 = const()[name = string("op_666_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_666_squeeze_mask_0 = const()[name = string("op_666_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_666_cast_fp16 = slice_by_index(begin = var_666_begin_0, end = var_666_end_0, end_mask = var_666_end_mask_0, squeeze_mask = var_666_squeeze_mask_0, x = var_663_cast_fp16)[name = string("op_666_cast_fp16")]; | |
| tensor<int32, [4]> var_681_begin_0 = const()[name = string("op_681_begin_0"), val = tensor<int32, [4]>([0, 9, 0, 0])]; | |
| tensor<int32, [4]> var_681_end_0 = const()[name = string("op_681_end_0"), val = tensor<int32, [4]>([1, 10, 1, 1536])]; | |
| tensor<bool, [4]> var_681_end_mask_0 = const()[name = string("op_681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_681_cast_fp16 = slice_by_index(begin = var_681_begin_0, end = var_681_end_0, end_mask = var_681_end_mask_0, x = obj_41_cast_fp16)[name = string("op_681_cast_fp16")]; | |
| tensor<int32, [4]> var_684_begin_0 = const()[name = string("op_684_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_684_end_0 = const()[name = string("op_684_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_684_end_mask_0 = const()[name = string("op_684_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_684_squeeze_mask_0 = const()[name = string("op_684_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_684_cast_fp16 = slice_by_index(begin = var_684_begin_0, end = var_684_end_0, end_mask = var_684_end_mask_0, squeeze_mask = var_684_squeeze_mask_0, x = var_681_cast_fp16)[name = string("op_684_cast_fp16")]; | |
| tensor<int32, [4]> var_699_begin_0 = const()[name = string("op_699_begin_0"), val = tensor<int32, [4]>([0, 10, 0, 0])]; | |
| tensor<int32, [4]> var_699_end_0 = const()[name = string("op_699_end_0"), val = tensor<int32, [4]>([1, 11, 1, 1536])]; | |
| tensor<bool, [4]> var_699_end_mask_0 = const()[name = string("op_699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_699_cast_fp16 = slice_by_index(begin = var_699_begin_0, end = var_699_end_0, end_mask = var_699_end_mask_0, x = obj_41_cast_fp16)[name = string("op_699_cast_fp16")]; | |
| tensor<int32, [4]> var_702_begin_0 = const()[name = string("op_702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_702_end_0 = const()[name = string("op_702_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_702_end_mask_0 = const()[name = string("op_702_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_702_squeeze_mask_0 = const()[name = string("op_702_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_702_cast_fp16 = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, squeeze_mask = var_702_squeeze_mask_0, x = var_699_cast_fp16)[name = string("op_702_cast_fp16")]; | |
| tensor<int32, [4]> var_717_begin_0 = const()[name = string("op_717_begin_0"), val = tensor<int32, [4]>([0, 11, 0, 0])]; | |
| tensor<int32, [4]> var_717_end_0 = const()[name = string("op_717_end_0"), val = tensor<int32, [4]>([1, 12, 1, 1536])]; | |
| tensor<bool, [4]> var_717_end_mask_0 = const()[name = string("op_717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_717_cast_fp16 = slice_by_index(begin = var_717_begin_0, end = var_717_end_0, end_mask = var_717_end_mask_0, x = obj_41_cast_fp16)[name = string("op_717_cast_fp16")]; | |
| tensor<int32, [4]> var_720_begin_0 = const()[name = string("op_720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_720_end_0 = const()[name = string("op_720_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_720_end_mask_0 = const()[name = string("op_720_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_720_squeeze_mask_0 = const()[name = string("op_720_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_720_cast_fp16 = slice_by_index(begin = var_720_begin_0, end = var_720_end_0, end_mask = var_720_end_mask_0, squeeze_mask = var_720_squeeze_mask_0, x = var_717_cast_fp16)[name = string("op_720_cast_fp16")]; | |
| tensor<int32, [4]> var_735_begin_0 = const()[name = string("op_735_begin_0"), val = tensor<int32, [4]>([0, 12, 0, 0])]; | |
| tensor<int32, [4]> var_735_end_0 = const()[name = string("op_735_end_0"), val = tensor<int32, [4]>([1, 13, 1, 1536])]; | |
| tensor<bool, [4]> var_735_end_mask_0 = const()[name = string("op_735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_735_cast_fp16 = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = obj_41_cast_fp16)[name = string("op_735_cast_fp16")]; | |
| tensor<int32, [4]> var_738_begin_0 = const()[name = string("op_738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_738_end_0 = const()[name = string("op_738_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_738_end_mask_0 = const()[name = string("op_738_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_738_squeeze_mask_0 = const()[name = string("op_738_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_738_cast_fp16 = slice_by_index(begin = var_738_begin_0, end = var_738_end_0, end_mask = var_738_end_mask_0, squeeze_mask = var_738_squeeze_mask_0, x = var_735_cast_fp16)[name = string("op_738_cast_fp16")]; | |
| tensor<int32, [4]> var_753_begin_0 = const()[name = string("op_753_begin_0"), val = tensor<int32, [4]>([0, 13, 0, 0])]; | |
| tensor<int32, [4]> var_753_end_0 = const()[name = string("op_753_end_0"), val = tensor<int32, [4]>([1, 14, 1, 1536])]; | |
| tensor<bool, [4]> var_753_end_mask_0 = const()[name = string("op_753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = obj_41_cast_fp16)[name = string("op_753_cast_fp16")]; | |
| tensor<int32, [4]> var_756_begin_0 = const()[name = string("op_756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_756_end_0 = const()[name = string("op_756_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_756_end_mask_0 = const()[name = string("op_756_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_756_squeeze_mask_0 = const()[name = string("op_756_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_756_cast_fp16 = slice_by_index(begin = var_756_begin_0, end = var_756_end_0, end_mask = var_756_end_mask_0, squeeze_mask = var_756_squeeze_mask_0, x = var_753_cast_fp16)[name = string("op_756_cast_fp16")]; | |
| tensor<int32, [4]> var_771_begin_0 = const()[name = string("op_771_begin_0"), val = tensor<int32, [4]>([0, 14, 0, 0])]; | |
| tensor<int32, [4]> var_771_end_0 = const()[name = string("op_771_end_0"), val = tensor<int32, [4]>([1, 15, 1, 1536])]; | |
| tensor<bool, [4]> var_771_end_mask_0 = const()[name = string("op_771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_771_cast_fp16 = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = obj_41_cast_fp16)[name = string("op_771_cast_fp16")]; | |
| tensor<int32, [4]> var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_774_squeeze_mask_0 = const()[name = string("op_774_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, squeeze_mask = var_774_squeeze_mask_0, x = var_771_cast_fp16)[name = string("op_774_cast_fp16")]; | |
| tensor<int32, [4]> var_789_begin_0 = const()[name = string("op_789_begin_0"), val = tensor<int32, [4]>([0, 15, 0, 0])]; | |
| tensor<int32, [4]> var_789_end_0 = const()[name = string("op_789_end_0"), val = tensor<int32, [4]>([1, 16, 1, 1536])]; | |
| tensor<bool, [4]> var_789_end_mask_0 = const()[name = string("op_789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_789_cast_fp16 = slice_by_index(begin = var_789_begin_0, end = var_789_end_0, end_mask = var_789_end_mask_0, x = obj_41_cast_fp16)[name = string("op_789_cast_fp16")]; | |
| tensor<int32, [4]> var_792_begin_0 = const()[name = string("op_792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_792_end_0 = const()[name = string("op_792_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_792_end_mask_0 = const()[name = string("op_792_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_792_squeeze_mask_0 = const()[name = string("op_792_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_792_cast_fp16 = slice_by_index(begin = var_792_begin_0, end = var_792_end_0, end_mask = var_792_end_mask_0, squeeze_mask = var_792_squeeze_mask_0, x = var_789_cast_fp16)[name = string("op_792_cast_fp16")]; | |
| tensor<int32, [4]> var_807_begin_0 = const()[name = string("op_807_begin_0"), val = tensor<int32, [4]>([0, 16, 0, 0])]; | |
| tensor<int32, [4]> var_807_end_0 = const()[name = string("op_807_end_0"), val = tensor<int32, [4]>([1, 17, 1, 1536])]; | |
| tensor<bool, [4]> var_807_end_mask_0 = const()[name = string("op_807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_807_cast_fp16 = slice_by_index(begin = var_807_begin_0, end = var_807_end_0, end_mask = var_807_end_mask_0, x = obj_41_cast_fp16)[name = string("op_807_cast_fp16")]; | |
| tensor<int32, [4]> var_810_begin_0 = const()[name = string("op_810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_810_end_0 = const()[name = string("op_810_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_810_end_mask_0 = const()[name = string("op_810_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_810_squeeze_mask_0 = const()[name = string("op_810_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_810_cast_fp16 = slice_by_index(begin = var_810_begin_0, end = var_810_end_0, end_mask = var_810_end_mask_0, squeeze_mask = var_810_squeeze_mask_0, x = var_807_cast_fp16)[name = string("op_810_cast_fp16")]; | |
| tensor<int32, [4]> var_825_begin_0 = const()[name = string("op_825_begin_0"), val = tensor<int32, [4]>([0, 17, 0, 0])]; | |
| tensor<int32, [4]> var_825_end_0 = const()[name = string("op_825_end_0"), val = tensor<int32, [4]>([1, 18, 1, 1536])]; | |
| tensor<bool, [4]> var_825_end_mask_0 = const()[name = string("op_825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_825_cast_fp16 = slice_by_index(begin = var_825_begin_0, end = var_825_end_0, end_mask = var_825_end_mask_0, x = obj_41_cast_fp16)[name = string("op_825_cast_fp16")]; | |
| tensor<int32, [4]> var_828_begin_0 = const()[name = string("op_828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_828_end_0 = const()[name = string("op_828_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_828_end_mask_0 = const()[name = string("op_828_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_828_squeeze_mask_0 = const()[name = string("op_828_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_828_cast_fp16 = slice_by_index(begin = var_828_begin_0, end = var_828_end_0, end_mask = var_828_end_mask_0, squeeze_mask = var_828_squeeze_mask_0, x = var_825_cast_fp16)[name = string("op_828_cast_fp16")]; | |
| tensor<int32, [4]> var_843_begin_0 = const()[name = string("op_843_begin_0"), val = tensor<int32, [4]>([0, 18, 0, 0])]; | |
| tensor<int32, [4]> var_843_end_0 = const()[name = string("op_843_end_0"), val = tensor<int32, [4]>([1, 19, 1, 1536])]; | |
| tensor<bool, [4]> var_843_end_mask_0 = const()[name = string("op_843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_843_cast_fp16 = slice_by_index(begin = var_843_begin_0, end = var_843_end_0, end_mask = var_843_end_mask_0, x = obj_41_cast_fp16)[name = string("op_843_cast_fp16")]; | |
| tensor<int32, [4]> var_846_begin_0 = const()[name = string("op_846_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_846_end_0 = const()[name = string("op_846_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_846_end_mask_0 = const()[name = string("op_846_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_846_squeeze_mask_0 = const()[name = string("op_846_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> 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 = var_843_cast_fp16)[name = string("op_846_cast_fp16")]; | |
| tensor<int32, [4]> var_861_begin_0 = const()[name = string("op_861_begin_0"), val = tensor<int32, [4]>([0, 19, 0, 0])]; | |
| tensor<int32, [4]> var_861_end_0 = const()[name = string("op_861_end_0"), val = tensor<int32, [4]>([1, 20, 1, 1536])]; | |
| tensor<bool, [4]> var_861_end_mask_0 = const()[name = string("op_861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1536]> var_861_cast_fp16 = slice_by_index(begin = var_861_begin_0, end = var_861_end_0, end_mask = var_861_end_mask_0, x = obj_41_cast_fp16)[name = string("op_861_cast_fp16")]; | |
| tensor<int32, [4]> var_864_begin_0 = const()[name = string("op_864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_864_end_0 = const()[name = string("op_864_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1536])]; | |
| tensor<bool, [4]> var_864_end_mask_0 = const()[name = string("op_864_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_864_squeeze_mask_0 = const()[name = string("op_864_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1536]> var_864_cast_fp16 = slice_by_index(begin = var_864_begin_0, end = var_864_end_0, end_mask = var_864_end_mask_0, squeeze_mask = var_864_squeeze_mask_0, x = var_861_cast_fp16)[name = string("op_864_cast_fp16")]; | |
| int32 var_871 = const()[name = string("op_871"), val = int32(1)]; | |
| bool var_872_interleave_0 = const()[name = string("op_872_interleave_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1536]> var_872_cast_fp16 = concat(axis = var_871, interleave = var_872_interleave_0, values = (var_522_cast_fp16, var_540_cast_fp16, var_558_cast_fp16, var_576_cast_fp16, var_594_cast_fp16, var_612_cast_fp16, var_630_cast_fp16, var_648_cast_fp16, var_666_cast_fp16, var_684_cast_fp16, var_702_cast_fp16, var_720_cast_fp16, var_738_cast_fp16, var_756_cast_fp16, var_774_cast_fp16, var_792_cast_fp16, var_810_cast_fp16, var_828_cast_fp16, var_846_cast_fp16, var_864_cast_fp16))[name = string("op_872_cast_fp16")]; | |
| bool var_875 = const()[name = string("op_875"), val = bool(false)]; | |
| tensor<int32, [1]> obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1536]> alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_875, x = var_872_cast_fp16)[name = string("obj_cast_fp16")]; | |
| } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); | |
| } |