diff --git a/.gitattributes b/.gitattributes index a6344aac8c09253b3b630fb776ae94478aa0275b..d3f22300b45211c32305aa7135189dc7ef2b593d 100644 --- a/.gitattributes +++ b/.gitattributes @@ -33,3 +33,10 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text *.zip filter=lfs diff=lfs merge=lfs -text *.zst filter=lfs diff=lfs merge=lfs -text *tfevents* filter=lfs diff=lfs merge=lfs -text +ggml/weights.gguf filter=lfs diff=lfs merge=lfs -text +qnn/libQnnHtp.so filter=lfs diff=lfs merge=lfs -text +qnn/libQnnHtpV75.so filter=lfs diff=lfs merge=lfs -text +qnn/libQnnHtpV75Skel.so filter=lfs diff=lfs merge=lfs -text +qnn/libQnnHtpV79.so filter=lfs diff=lfs merge=lfs -text +qnn/libQnnHtpV79Skel.so filter=lfs diff=lfs merge=lfs -text +vocab.gguf filter=lfs diff=lfs merge=lfs -text diff --git a/ggml/weights.gguf b/ggml/weights.gguf new file mode 100644 index 0000000000000000000000000000000000000000..be68df768ef5199b05b786b75a102e1a54ba6161 --- /dev/null +++ b/ggml/weights.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86580f476a9e993a024237811268d43ac869f5228da84e87c0fc3a39760e7083 +size 675709216 diff --git a/model.json b/model.json new file mode 100644 index 0000000000000000000000000000000000000000..deb69ba6f2278ada381e128a11a1a94029cb76c4 --- /dev/null +++ b/model.json @@ -0,0 +1,27 @@ +{ + "model_id": "smallthinker_500m", + "model_arch": "qwen2", + "version": 20241202, + "llm_config": { + "embed_dim": 896, + "ffn_dim": 4864, + "head_size": 64, + "kv_dim": 128, + "n_attn_heads": 14, + "n_attn_kv_heads": 2, + "n_ctx": 32768, + "n_layers": 24, + "norm_eps": 9.999999974752427e-07, + "vocab_size": 151936, + "rope_config": { + "n_rope_ctx_orig": 32768, + "rope_attn_factor": 1.0, + "rope_dim": 64, + "rope_freq_base": 1000000.0, + "rope_freq_scale": 1.0, + "rope_scale_type": "linear", + "rope_type": -1 + } + }, + "vision": {} +} \ No newline at end of file diff --git a/qnn/config.json b/qnn/config.json new file mode 100644 index 0000000000000000000000000000000000000000..2331ded69d97fda650e9880775e573d59225114e --- /dev/null +++ b/qnn/config.json @@ -0,0 +1,63 @@ +{ + "model_parameters": { + "n_layers": 24, + "vocab_size": 151936, + "embed_dim": 896, + "ffn_hidden_dim": 4864, + "head_dim": 64, + "n_kv_heads": 2, + "rope_theta": 1000000.0, + "rms_norm_eps": 1e-06, + "attention_mask_value": -50000.0, + "tie_embedding": true + }, + "qnn_parameters": { + "n_hvx_threads": 4 + }, + "graphs": [ + { + "type": "transformers", + "start_layer_id": 0, + "end_layer_id": 24, + "batch_size": 1, + "cache_size": 1920, + "context_size": 2048, + "graph_name": "batch_1", + "model_path": "smallthinker_500m_0.bin", + "kv_path_format": "kv/layer_{layer_id}_{kv_type}_{head_id}.raw", + "kv_size": 11, + "x_name": "x", + "out_name": "out" + }, + { + "type": "transformers", + "start_layer_id": 0, + "end_layer_id": 24, + "batch_size": 128, + "cache_size": 1920, + "context_size": 2048, + "graph_name": "batch_128", + "model_path": "smallthinker_500m_0.bin", + "kv_path_format": "kv/layer_{layer_id}_{kv_type}_{head_id}.raw", + "kv_size": 11, + "x_name": "x", + "out_name": "out" + } + ], + "embeddings": [ + { + "graph_name": "batch_1", + "model_path": "lm_head.bin", + "batch_size": 1, + "x_name": "x", + "out_name": "logits" + }, + { + "graph_name": "batch_128", + "model_path": "lm_head.bin", + "batch_size": 128, + "x_name": "x", + "out_name": "logits" + } + ] +} \ No newline at end of file diff --git a/qnn/kv/layer_0_key_0.raw b/qnn/kv/layer_0_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..91954383f0bc09e196330920b0a2a23c730b8ee4 Binary files /dev/null and b/qnn/kv/layer_0_key_0.raw differ diff --git a/qnn/kv/layer_0_key_1.raw b/qnn/kv/layer_0_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..dbcd8da7d86648614bd12d560ce448c583638d4e Binary files /dev/null and b/qnn/kv/layer_0_key_1.raw differ diff --git a/qnn/kv/layer_0_value_0.raw b/qnn/kv/layer_0_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..c67d15abfcbb25d24c9914f6417029e7e0c29545 Binary files /dev/null and b/qnn/kv/layer_0_value_0.raw differ diff --git a/qnn/kv/layer_0_value_1.raw b/qnn/kv/layer_0_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..620e65b2b04379bb7a7b66c4d2975fee257ef7a6 Binary files /dev/null and b/qnn/kv/layer_0_value_1.raw differ diff --git a/qnn/kv/layer_10_key_0.raw b/qnn/kv/layer_10_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..5cf2c3f396c0198a42bdd39bbb07bc5505cda14d Binary files /dev/null and b/qnn/kv/layer_10_key_0.raw differ diff --git a/qnn/kv/layer_10_key_1.raw b/qnn/kv/layer_10_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..1af54ba7df56656d364bcc3755c4b45ef79a05a9 Binary files /dev/null and b/qnn/kv/layer_10_key_1.raw differ diff --git a/qnn/kv/layer_10_value_0.raw b/qnn/kv/layer_10_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..d56e16cebeca0f9a3bb8f0a9e3adf79d1ece90a3 Binary files /dev/null and b/qnn/kv/layer_10_value_0.raw differ diff --git a/qnn/kv/layer_10_value_1.raw b/qnn/kv/layer_10_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..aa1dfd0b0ec9db927131a4eea528001737f7370e Binary files /dev/null and b/qnn/kv/layer_10_value_1.raw differ diff --git a/qnn/kv/layer_11_key_0.raw b/qnn/kv/layer_11_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..3946a45a239b94e59be36c5337e587465a26f2c8 Binary files /dev/null and b/qnn/kv/layer_11_key_0.raw differ diff --git a/qnn/kv/layer_11_key_1.raw b/qnn/kv/layer_11_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..09b47f66692bebdbc38637b8939ec4ed955a5334 Binary files /dev/null and b/qnn/kv/layer_11_key_1.raw differ diff --git a/qnn/kv/layer_11_value_0.raw b/qnn/kv/layer_11_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..4df655dfab7db052297931838a2242cd76510be7 Binary files /dev/null and b/qnn/kv/layer_11_value_0.raw differ diff --git a/qnn/kv/layer_11_value_1.raw b/qnn/kv/layer_11_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..969a025c70edef77806e00fcc7bcc601e1808dc5 Binary files /dev/null and b/qnn/kv/layer_11_value_1.raw differ diff --git a/qnn/kv/layer_12_key_0.raw b/qnn/kv/layer_12_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..4b65ae7d4ea05f1b026ac5f2cd408e50ff607683 Binary files /dev/null and b/qnn/kv/layer_12_key_0.raw differ diff --git a/qnn/kv/layer_12_key_1.raw b/qnn/kv/layer_12_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..5d5f677730245972b9f453bec6639dbe2d86e193 Binary files /dev/null and b/qnn/kv/layer_12_key_1.raw differ diff --git a/qnn/kv/layer_12_value_0.raw b/qnn/kv/layer_12_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..a3c93960e7b7d86d3740214b36ae12a69c4dfa96 Binary files /dev/null and b/qnn/kv/layer_12_value_0.raw differ diff --git a/qnn/kv/layer_12_value_1.raw b/qnn/kv/layer_12_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..0c20eee3694ef88a4e5e4a0658e20d3d9c25d8b3 Binary files /dev/null and b/qnn/kv/layer_12_value_1.raw differ diff --git a/qnn/kv/layer_13_key_0.raw b/qnn/kv/layer_13_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..2c2a9fb40dddb59a7a4a767e34a89ea7b09b5d0f Binary files /dev/null and b/qnn/kv/layer_13_key_0.raw differ diff --git a/qnn/kv/layer_13_key_1.raw b/qnn/kv/layer_13_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..4de90fa5dd8b8fe7180c6671e2fc6c67005cb0b6 Binary files /dev/null and b/qnn/kv/layer_13_key_1.raw differ diff --git a/qnn/kv/layer_13_value_0.raw b/qnn/kv/layer_13_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..c2b7e14395d5f8012e807162e7345232b8fb138d Binary files /dev/null and b/qnn/kv/layer_13_value_0.raw differ diff --git a/qnn/kv/layer_13_value_1.raw b/qnn/kv/layer_13_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..487e83eaaaf994fa399c976fffd69ad78b923ab4 Binary files /dev/null and b/qnn/kv/layer_13_value_1.raw differ diff --git a/qnn/kv/layer_14_key_0.raw b/qnn/kv/layer_14_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..b52cea12ef8691e0deb5ca2aacd8273c74537737 Binary files /dev/null and b/qnn/kv/layer_14_key_0.raw differ diff --git a/qnn/kv/layer_14_key_1.raw b/qnn/kv/layer_14_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..bab59d5c377824a8d433ac9fc17f7ec4350b01d3 Binary files /dev/null and b/qnn/kv/layer_14_key_1.raw differ diff --git a/qnn/kv/layer_14_value_0.raw b/qnn/kv/layer_14_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..80d436a43c0be51c9039afb97dc92b7230340a2f Binary files /dev/null and b/qnn/kv/layer_14_value_0.raw differ diff --git a/qnn/kv/layer_14_value_1.raw b/qnn/kv/layer_14_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..f36d8a31ec170541e6c4b6e5953b4bc9635aae80 Binary files /dev/null and b/qnn/kv/layer_14_value_1.raw differ diff --git a/qnn/kv/layer_15_key_0.raw b/qnn/kv/layer_15_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..c768a70559db1eef175fc56703d68b7a927909c0 Binary files /dev/null and b/qnn/kv/layer_15_key_0.raw differ diff --git a/qnn/kv/layer_15_key_1.raw b/qnn/kv/layer_15_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..24c8cc6540a8f7b4029fcf50d70b41d72259d8cf Binary files /dev/null and b/qnn/kv/layer_15_key_1.raw differ diff --git a/qnn/kv/layer_15_value_0.raw b/qnn/kv/layer_15_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..f9ed52dbe8bb378629ff76e9346bd1fe9aca439a Binary files /dev/null and b/qnn/kv/layer_15_value_0.raw differ diff --git a/qnn/kv/layer_15_value_1.raw b/qnn/kv/layer_15_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..1a6b31e60b5bb3d9ab824b565ef73e3f61a2495d Binary files /dev/null and b/qnn/kv/layer_15_value_1.raw differ diff --git a/qnn/kv/layer_16_key_0.raw b/qnn/kv/layer_16_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..32734e25131137e12199952a563b6c68464357cd Binary files /dev/null and b/qnn/kv/layer_16_key_0.raw differ diff --git a/qnn/kv/layer_16_key_1.raw b/qnn/kv/layer_16_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..c38451bdfd85c19909124d3661375f5d844992fd Binary files /dev/null and b/qnn/kv/layer_16_key_1.raw differ diff --git a/qnn/kv/layer_16_value_0.raw b/qnn/kv/layer_16_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..a134f0ca555232fafd0a7a72aab7644d632aa1cf Binary files /dev/null and b/qnn/kv/layer_16_value_0.raw differ diff --git a/qnn/kv/layer_16_value_1.raw b/qnn/kv/layer_16_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..e449fca6d7efba634f0326453129ece7fa0a1059 Binary files /dev/null and b/qnn/kv/layer_16_value_1.raw differ diff --git a/qnn/kv/layer_17_key_0.raw b/qnn/kv/layer_17_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..1152e128fd840d580d7589d4066c21f0d06d8148 Binary files /dev/null and b/qnn/kv/layer_17_key_0.raw differ diff --git a/qnn/kv/layer_17_key_1.raw b/qnn/kv/layer_17_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..fe6cbaddbfb278faba6478cf4ee770a6e9a08556 Binary files /dev/null and b/qnn/kv/layer_17_key_1.raw differ diff --git a/qnn/kv/layer_17_value_0.raw b/qnn/kv/layer_17_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..ad261a06ebf96058be223c46922dcbe39f1c3be5 Binary files /dev/null and b/qnn/kv/layer_17_value_0.raw differ diff --git a/qnn/kv/layer_17_value_1.raw b/qnn/kv/layer_17_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..ac1d706fb1b02bcb1632116fb359392f5e2579cc Binary files /dev/null and b/qnn/kv/layer_17_value_1.raw differ diff --git a/qnn/kv/layer_18_key_0.raw b/qnn/kv/layer_18_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..5b9027bfd6e7e027dd4763fabf08aa0177ac10f9 Binary files /dev/null and b/qnn/kv/layer_18_key_0.raw differ diff --git a/qnn/kv/layer_18_key_1.raw b/qnn/kv/layer_18_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..dc7f4e945b8fe8c6d990d8ac3f6c8b6805c602ff Binary files /dev/null and b/qnn/kv/layer_18_key_1.raw differ diff --git a/qnn/kv/layer_18_value_0.raw b/qnn/kv/layer_18_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..48361f8bf5b7abb3f25fe99c148b3ff30fce74bb Binary files /dev/null and b/qnn/kv/layer_18_value_0.raw differ diff --git a/qnn/kv/layer_18_value_1.raw b/qnn/kv/layer_18_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..8bc3d847ea9bf03f7dcbda37e5b5b9d02e92e7e1 Binary files /dev/null and b/qnn/kv/layer_18_value_1.raw differ diff --git a/qnn/kv/layer_19_key_0.raw b/qnn/kv/layer_19_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..8cc79e14b22b0cbf0c3633d0d5d231aaeb15c688 Binary files /dev/null and b/qnn/kv/layer_19_key_0.raw differ diff --git a/qnn/kv/layer_19_key_1.raw b/qnn/kv/layer_19_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..df1d3335b03d966695250f6958493dd2499a41ea Binary files /dev/null and b/qnn/kv/layer_19_key_1.raw differ diff --git a/qnn/kv/layer_19_value_0.raw b/qnn/kv/layer_19_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..3efec7f101bf9c9e83db8aaeeb3df6ed713f01ab Binary files /dev/null and b/qnn/kv/layer_19_value_0.raw differ diff --git a/qnn/kv/layer_19_value_1.raw b/qnn/kv/layer_19_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..1309a5154b5c7662ace52e40746ef9d5b2e5115e Binary files /dev/null and b/qnn/kv/layer_19_value_1.raw differ diff --git a/qnn/kv/layer_1_key_0.raw b/qnn/kv/layer_1_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..847511820fc6c823e7356add6e4de842b02cf653 Binary files /dev/null and b/qnn/kv/layer_1_key_0.raw differ diff --git a/qnn/kv/layer_1_key_1.raw b/qnn/kv/layer_1_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..9c9a82ff537ea6a88adbdf8b1034337d62f33a37 Binary files /dev/null and b/qnn/kv/layer_1_key_1.raw differ diff --git a/qnn/kv/layer_1_value_0.raw b/qnn/kv/layer_1_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..096c6e0f87dd475786bcd929dac58ea12a807fe7 Binary files /dev/null and b/qnn/kv/layer_1_value_0.raw differ diff --git a/qnn/kv/layer_1_value_1.raw b/qnn/kv/layer_1_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..310dd5812d2168aea4ed9aa2a7e816e16271aea9 Binary files /dev/null and b/qnn/kv/layer_1_value_1.raw differ diff --git a/qnn/kv/layer_20_key_0.raw b/qnn/kv/layer_20_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..cb9c226a8186c3fb72f56bcbffb719eeb45e8a77 Binary files /dev/null and b/qnn/kv/layer_20_key_0.raw differ diff --git a/qnn/kv/layer_20_key_1.raw b/qnn/kv/layer_20_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..d6a5c888f62218c9766a43f467847727966cd854 Binary files /dev/null and b/qnn/kv/layer_20_key_1.raw differ diff --git a/qnn/kv/layer_20_value_0.raw b/qnn/kv/layer_20_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..446d9a821efbab55b7dca7bcf4f065c17cfa8f52 Binary files /dev/null and b/qnn/kv/layer_20_value_0.raw differ diff --git a/qnn/kv/layer_20_value_1.raw b/qnn/kv/layer_20_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..bac708f2dbda536e79be4688d42724c94420bf8d Binary files /dev/null and b/qnn/kv/layer_20_value_1.raw differ diff --git a/qnn/kv/layer_21_key_0.raw b/qnn/kv/layer_21_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..f0f689c1faf6101d5f62ee39980263d8e85bc740 Binary files /dev/null and b/qnn/kv/layer_21_key_0.raw differ diff --git a/qnn/kv/layer_21_key_1.raw b/qnn/kv/layer_21_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..2150927fa280f1723bf14d6e9e4d2218f5b97ce3 Binary files /dev/null and b/qnn/kv/layer_21_key_1.raw differ diff --git a/qnn/kv/layer_21_value_0.raw b/qnn/kv/layer_21_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..92b07f318cdc0a4c890ff86c8906fb6078a386dd Binary files /dev/null and b/qnn/kv/layer_21_value_0.raw differ diff --git a/qnn/kv/layer_21_value_1.raw b/qnn/kv/layer_21_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..ade47f6952901c1905e53b47f32dd4f3d3db18c6 Binary files /dev/null and b/qnn/kv/layer_21_value_1.raw differ diff --git a/qnn/kv/layer_22_key_0.raw b/qnn/kv/layer_22_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..c466499e53d6d9a8b4f420c4f057059e646317c9 Binary files /dev/null and b/qnn/kv/layer_22_key_0.raw differ diff --git a/qnn/kv/layer_22_key_1.raw b/qnn/kv/layer_22_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..eae10714600ce82be12e33ce02ebc384d86c8761 Binary files /dev/null and b/qnn/kv/layer_22_key_1.raw differ diff --git a/qnn/kv/layer_22_value_0.raw b/qnn/kv/layer_22_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..6845c7901ca29c51a29ca0070ee8a2bd2a2af87a Binary files /dev/null and b/qnn/kv/layer_22_value_0.raw differ diff --git a/qnn/kv/layer_22_value_1.raw b/qnn/kv/layer_22_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..fd88e4fdcc42d3e36de8d5283ebc3170ade268ce Binary files /dev/null and b/qnn/kv/layer_22_value_1.raw differ diff --git a/qnn/kv/layer_23_key_0.raw b/qnn/kv/layer_23_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..aa77bfed0e7679cebb4a400774648bd2c75064ac Binary files /dev/null and b/qnn/kv/layer_23_key_0.raw differ diff --git a/qnn/kv/layer_23_key_1.raw b/qnn/kv/layer_23_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..4cb4475d0f72781f576ec6f7ba32d48aeff95544 Binary files /dev/null and b/qnn/kv/layer_23_key_1.raw differ diff --git a/qnn/kv/layer_23_value_0.raw b/qnn/kv/layer_23_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..74b868e15ae5ba2dcb6fbb6fa6dcf8dd791fd85f Binary files /dev/null and b/qnn/kv/layer_23_value_0.raw differ diff --git a/qnn/kv/layer_23_value_1.raw b/qnn/kv/layer_23_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..c7ebe5ac48c1ea4221faa3d0c8208768d8cadcbc Binary files /dev/null and b/qnn/kv/layer_23_value_1.raw differ diff --git a/qnn/kv/layer_2_key_0.raw b/qnn/kv/layer_2_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..5f6fc59e7e2aa58c22cb8ee7b22147cd4141c156 Binary files /dev/null and b/qnn/kv/layer_2_key_0.raw differ diff --git a/qnn/kv/layer_2_key_1.raw b/qnn/kv/layer_2_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..ef352c2ca730e07e0c39e150dc21e0731dbcbdac Binary files /dev/null and b/qnn/kv/layer_2_key_1.raw differ diff --git a/qnn/kv/layer_2_value_0.raw b/qnn/kv/layer_2_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..f52afd51fae504c354830062f67712f0fa76f107 Binary files /dev/null and b/qnn/kv/layer_2_value_0.raw differ diff --git a/qnn/kv/layer_2_value_1.raw b/qnn/kv/layer_2_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..8290ad2d713d0542d7abc8ce953ae6fe4228c43c Binary files /dev/null and b/qnn/kv/layer_2_value_1.raw differ diff --git a/qnn/kv/layer_3_key_0.raw b/qnn/kv/layer_3_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..1db5d8884bc9a9a2cc77d8e3948220df1f91bac4 Binary files /dev/null and b/qnn/kv/layer_3_key_0.raw differ diff --git a/qnn/kv/layer_3_key_1.raw b/qnn/kv/layer_3_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..071b91952d047f128d4dce32959fd343f81483e9 Binary files /dev/null and b/qnn/kv/layer_3_key_1.raw differ diff --git a/qnn/kv/layer_3_value_0.raw b/qnn/kv/layer_3_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..e3b627d203a0da5b8578ea15156567ed7da7e84d Binary files /dev/null and b/qnn/kv/layer_3_value_0.raw differ diff --git a/qnn/kv/layer_3_value_1.raw b/qnn/kv/layer_3_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..714b8fb928733e72103ef61c9b8b20843b8e7603 Binary files /dev/null and b/qnn/kv/layer_3_value_1.raw differ diff --git a/qnn/kv/layer_4_key_0.raw b/qnn/kv/layer_4_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..b90f33b4be898c1d0a49b0302ca52227c3fc87ec Binary files /dev/null and b/qnn/kv/layer_4_key_0.raw differ diff --git a/qnn/kv/layer_4_key_1.raw b/qnn/kv/layer_4_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..a4394280ea10e28841ea5fa5860174a60885489e Binary files /dev/null and b/qnn/kv/layer_4_key_1.raw differ diff --git a/qnn/kv/layer_4_value_0.raw b/qnn/kv/layer_4_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..53c88f0f70b5ad8e90ed328f0f186bd5a8770df3 Binary files /dev/null and b/qnn/kv/layer_4_value_0.raw differ diff --git a/qnn/kv/layer_4_value_1.raw b/qnn/kv/layer_4_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..fa28c5fa2ca1db5282145fc571b8922e96a4aef6 Binary files /dev/null and b/qnn/kv/layer_4_value_1.raw differ diff --git a/qnn/kv/layer_5_key_0.raw b/qnn/kv/layer_5_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..7b4c4650c12830ee0014fb064712a1e39047bcf8 Binary files /dev/null and b/qnn/kv/layer_5_key_0.raw differ diff --git a/qnn/kv/layer_5_key_1.raw b/qnn/kv/layer_5_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..59e474876102240c1d46a035475bf2707320467d Binary files /dev/null and b/qnn/kv/layer_5_key_1.raw differ diff --git a/qnn/kv/layer_5_value_0.raw b/qnn/kv/layer_5_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..98684ab88c2289698697c58e283f75c5a2c8f3dc Binary files /dev/null and b/qnn/kv/layer_5_value_0.raw differ diff --git a/qnn/kv/layer_5_value_1.raw b/qnn/kv/layer_5_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..3fdc37ed259006a85f90686ee2fc94ba1863112a Binary files /dev/null and b/qnn/kv/layer_5_value_1.raw differ diff --git a/qnn/kv/layer_6_key_0.raw b/qnn/kv/layer_6_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..052fe5309d6ccbdc610c65d619d16f16bbeb303e Binary files /dev/null and b/qnn/kv/layer_6_key_0.raw differ diff --git a/qnn/kv/layer_6_key_1.raw b/qnn/kv/layer_6_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..8542dd4f8627f171dd0444f2663eccf7ca5642ca Binary files /dev/null and b/qnn/kv/layer_6_key_1.raw differ diff --git a/qnn/kv/layer_6_value_0.raw b/qnn/kv/layer_6_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..9d7232aaa1e9d9cc08ab5e09e00091d98f996100 Binary files /dev/null and b/qnn/kv/layer_6_value_0.raw differ diff --git a/qnn/kv/layer_6_value_1.raw b/qnn/kv/layer_6_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..3f8c3af77d89e26b58fe2a1865c05ff7e16bc3fc Binary files /dev/null and b/qnn/kv/layer_6_value_1.raw differ diff --git a/qnn/kv/layer_7_key_0.raw b/qnn/kv/layer_7_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..38e470c541c4289c537ad30f9606fe1f6bb2d8c2 Binary files /dev/null and b/qnn/kv/layer_7_key_0.raw differ diff --git a/qnn/kv/layer_7_key_1.raw b/qnn/kv/layer_7_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..75f012ad98516a20f914a0f1a7b1ea4408c640f3 Binary files /dev/null and b/qnn/kv/layer_7_key_1.raw differ diff --git a/qnn/kv/layer_7_value_0.raw b/qnn/kv/layer_7_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..fdc8f04f2b542249228a6e333e4d1d3fbc60e470 Binary files /dev/null and b/qnn/kv/layer_7_value_0.raw differ diff --git a/qnn/kv/layer_7_value_1.raw b/qnn/kv/layer_7_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..418c52c481aa271de158eedc94f9544b61d05041 Binary files /dev/null and b/qnn/kv/layer_7_value_1.raw differ diff --git a/qnn/kv/layer_8_key_0.raw b/qnn/kv/layer_8_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..84f3c024ba09435644dfc3e083cba72af9814c13 Binary files /dev/null and b/qnn/kv/layer_8_key_0.raw differ diff --git a/qnn/kv/layer_8_key_1.raw b/qnn/kv/layer_8_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..4fb1bc8f9be2933caa4d5db00c162caabe3fae50 Binary files /dev/null and b/qnn/kv/layer_8_key_1.raw differ diff --git a/qnn/kv/layer_8_value_0.raw b/qnn/kv/layer_8_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..c179761a1ada405e80e7080e6594d22c84442340 Binary files /dev/null and b/qnn/kv/layer_8_value_0.raw differ diff --git a/qnn/kv/layer_8_value_1.raw b/qnn/kv/layer_8_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..247b02fdc19c14a115ba6f1444383652903d947c Binary files /dev/null and b/qnn/kv/layer_8_value_1.raw differ diff --git a/qnn/kv/layer_9_key_0.raw b/qnn/kv/layer_9_key_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..e949e1d939c7423914e42fcb2f82bd67c2a302be Binary files /dev/null and b/qnn/kv/layer_9_key_0.raw differ diff --git a/qnn/kv/layer_9_key_1.raw b/qnn/kv/layer_9_key_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..6c073a585347ee2b01e0f689f95c0a4cec160ec7 Binary files /dev/null and b/qnn/kv/layer_9_key_1.raw differ diff --git a/qnn/kv/layer_9_value_0.raw b/qnn/kv/layer_9_value_0.raw new file mode 100644 index 0000000000000000000000000000000000000000..114de67f949c399edaaf187db9fb7d6eab899b3d Binary files /dev/null and b/qnn/kv/layer_9_value_0.raw differ diff --git a/qnn/kv/layer_9_value_1.raw b/qnn/kv/layer_9_value_1.raw new file mode 100644 index 0000000000000000000000000000000000000000..fbb1ee10c1550d138c76396ff3eaa422d4868deb Binary files /dev/null and b/qnn/kv/layer_9_value_1.raw differ diff --git a/qnn/libQnnHexagonSkel_dspApp.so b/qnn/libQnnHexagonSkel_dspApp.so new file mode 100644 index 0000000000000000000000000000000000000000..0dd8874e9ca678a88f0500a09a95cade7bc7e7a5 Binary files /dev/null and b/qnn/libQnnHexagonSkel_dspApp.so differ diff --git a/qnn/libQnnHtp.so b/qnn/libQnnHtp.so new file mode 100644 index 0000000000000000000000000000000000000000..4c43b59cf9beaf71ed06d6669209c1102a802c38 --- /dev/null +++ b/qnn/libQnnHtp.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c869e376e25580cbd9b116b7d18b662dffddcb48f41b1e67e7c7f89e55f1fc8a +size 2014096 diff --git a/qnn/libQnnHtpV75.so b/qnn/libQnnHtpV75.so new file mode 100644 index 0000000000000000000000000000000000000000..e9113aa1a90432b8171eeb37d841822516a3ac9e --- /dev/null +++ b/qnn/libQnnHtpV75.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:778633b0c0a36b1218c18386fbba4bc77cb28f303b33a988054a7d2ad404365e +size 10746336 diff --git a/qnn/libQnnHtpV75Skel.so b/qnn/libQnnHtpV75Skel.so new file mode 100644 index 0000000000000000000000000000000000000000..75114ac0cd7f7205d6abd4f82992eb6840f5e0c2 --- /dev/null +++ b/qnn/libQnnHtpV75Skel.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21399f3ebb637d0546f00fa90a89373019c283a059738dda2695de172a4fbd73 +size 8400052 diff --git a/qnn/libQnnHtpV75Stub.so b/qnn/libQnnHtpV75Stub.so new file mode 100644 index 0000000000000000000000000000000000000000..71154c1cd4a92877cee2ddb43c6ffa92072e89ba Binary files /dev/null and b/qnn/libQnnHtpV75Stub.so differ diff --git a/qnn/libQnnHtpV79.so b/qnn/libQnnHtpV79.so new file mode 100644 index 0000000000000000000000000000000000000000..a233f00bec763ca29da5ddae0e865e3789e73dc3 --- /dev/null +++ b/qnn/libQnnHtpV79.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bfb62d3208eff8f10d43cfcda62169d892cf6c7937a2be5630853b8316dc4785 +size 11221468 diff --git a/qnn/libQnnHtpV79Skel.so b/qnn/libQnnHtpV79Skel.so new file mode 100644 index 0000000000000000000000000000000000000000..8d80530489306f6fe5a8e00be7f2e42b4109b2ca --- /dev/null +++ b/qnn/libQnnHtpV79Skel.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9cd8ca94206232c98d920cea911e9af660a3eb49425e752374b7dccb57706cf1 +size 8596664 diff --git a/qnn/libQnnHtpV79Stub.so b/qnn/libQnnHtpV79Stub.so new file mode 100644 index 0000000000000000000000000000000000000000..dd9d368334bd785a602eedf820a3b6ec66617624 Binary files /dev/null and b/qnn/libQnnHtpV79Stub.so differ diff --git a/qnn/libQnnSystem.so b/qnn/libQnnSystem.so new file mode 100644 index 0000000000000000000000000000000000000000..81b29f9db4a5fdaff9763ef8a0e9d50cc2880c7b Binary files /dev/null and b/qnn/libQnnSystem.so differ diff --git a/qnn/lm_head.bin b/qnn/lm_head.bin new file mode 100644 index 0000000000000000000000000000000000000000..0cb145ca10b872fda39b90140de1aaeededf18cd --- /dev/null +++ b/qnn/lm_head.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2676cf7b04415ce18713493c454ea486aa7e117ecc8db90cd77224bd24cbe49d +size 76523904 diff --git a/qnn/smallthinker_500m_0.bin b/qnn/smallthinker_500m_0.bin new file mode 100644 index 0000000000000000000000000000000000000000..5bbbd42b6f71849ff492b2ae6b51befe50ef9d64 --- /dev/null +++ b/qnn/smallthinker_500m_0.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a91feef38830b796a59674a731c6845f321220daccd1830abb44c107c529e888 +size 321768408 diff --git a/vocab.gguf b/vocab.gguf new file mode 100644 index 0000000000000000000000000000000000000000..64619f4ee6bd9fa0a050621cc66b5b2cd59af438 --- /dev/null +++ b/vocab.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddf42ea321c52c34043c8afd049707722823d26c27a15efedc9452be6c6a4f57 +size 5929526