Upload 3 files
Browse files- pytorch_model.bin.index.json +871 -0
- tokenization_qwen.py +593 -0
- visual.py +482 -0
pytorch_model.bin.index.json
ADDED
|
@@ -0,0 +1,871 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_size": 19465979392
|
| 4 |
+
},
|
| 5 |
+
"weight_map": {
|
| 6 |
+
"lm_head.weight": "pytorch_model-00002-of-00002.bin",
|
| 7 |
+
"transformer.h.0.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 8 |
+
"transformer.h.0.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 9 |
+
"transformer.h.0.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 10 |
+
"transformer.h.0.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 11 |
+
"transformer.h.0.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 12 |
+
"transformer.h.0.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 13 |
+
"transformer.h.0.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 14 |
+
"transformer.h.0.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 15 |
+
"transformer.h.1.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 16 |
+
"transformer.h.1.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 17 |
+
"transformer.h.1.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 18 |
+
"transformer.h.1.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 19 |
+
"transformer.h.1.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 20 |
+
"transformer.h.1.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 21 |
+
"transformer.h.1.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 22 |
+
"transformer.h.1.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 23 |
+
"transformer.h.10.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 24 |
+
"transformer.h.10.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 25 |
+
"transformer.h.10.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 26 |
+
"transformer.h.10.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 27 |
+
"transformer.h.10.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 28 |
+
"transformer.h.10.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 29 |
+
"transformer.h.10.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 30 |
+
"transformer.h.10.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 31 |
+
"transformer.h.11.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 32 |
+
"transformer.h.11.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 33 |
+
"transformer.h.11.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 34 |
+
"transformer.h.11.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 35 |
+
"transformer.h.11.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 36 |
+
"transformer.h.11.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 37 |
+
"transformer.h.11.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 38 |
+
"transformer.h.11.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 39 |
+
"transformer.h.12.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 40 |
+
"transformer.h.12.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 41 |
+
"transformer.h.12.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 42 |
+
"transformer.h.12.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 43 |
+
"transformer.h.12.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 44 |
+
"transformer.h.12.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 45 |
+
"transformer.h.12.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 46 |
+
"transformer.h.12.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 47 |
+
"transformer.h.13.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 48 |
+
"transformer.h.13.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 49 |
+
"transformer.h.13.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 50 |
+
"transformer.h.13.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 51 |
+
"transformer.h.13.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 52 |
+
"transformer.h.13.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 53 |
+
"transformer.h.13.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 54 |
+
"transformer.h.13.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 55 |
+
"transformer.h.14.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 56 |
+
"transformer.h.14.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 57 |
+
"transformer.h.14.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 58 |
+
"transformer.h.14.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 59 |
+
"transformer.h.14.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 60 |
+
"transformer.h.14.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 61 |
+
"transformer.h.14.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 62 |
+
"transformer.h.14.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 63 |
+
"transformer.h.15.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 64 |
+
"transformer.h.15.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 65 |
+
"transformer.h.15.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 66 |
+
"transformer.h.15.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 67 |
+
"transformer.h.15.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 68 |
+
"transformer.h.15.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 69 |
+
"transformer.h.15.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 70 |
+
"transformer.h.15.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 71 |
+
"transformer.h.16.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 72 |
+
"transformer.h.16.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 73 |
+
"transformer.h.16.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 74 |
+
"transformer.h.16.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 75 |
+
"transformer.h.16.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 76 |
+
"transformer.h.16.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 77 |
+
"transformer.h.16.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 78 |
+
"transformer.h.16.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 79 |
+
"transformer.h.17.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 80 |
+
"transformer.h.17.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 81 |
+
"transformer.h.17.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 82 |
+
"transformer.h.17.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 83 |
+
"transformer.h.17.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 84 |
+
"transformer.h.17.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 85 |
+
"transformer.h.17.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 86 |
+
"transformer.h.17.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 87 |
+
"transformer.h.18.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 88 |
+
"transformer.h.18.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 89 |
+
"transformer.h.18.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 90 |
+
"transformer.h.18.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 91 |
+
"transformer.h.18.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 92 |
+
"transformer.h.18.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 93 |
+
"transformer.h.18.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 94 |
+
"transformer.h.18.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 95 |
+
"transformer.h.19.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 96 |
+
"transformer.h.19.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 97 |
+
"transformer.h.19.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 98 |
+
"transformer.h.19.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 99 |
+
"transformer.h.19.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 100 |
+
"transformer.h.19.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 101 |
+
"transformer.h.19.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 102 |
+
"transformer.h.19.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 103 |
+
"transformer.h.2.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 104 |
+
"transformer.h.2.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 105 |
+
"transformer.h.2.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 106 |
+
"transformer.h.2.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 107 |
+
"transformer.h.2.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 108 |
+
"transformer.h.2.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 109 |
+
"transformer.h.2.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 110 |
+
"transformer.h.2.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 111 |
+
"transformer.h.20.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 112 |
+
"transformer.h.20.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 113 |
+
"transformer.h.20.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 114 |
+
"transformer.h.20.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 115 |
+
"transformer.h.20.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 116 |
+
"transformer.h.20.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 117 |
+
"transformer.h.20.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 118 |
+
"transformer.h.20.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 119 |
+
"transformer.h.21.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 120 |
+
"transformer.h.21.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 121 |
+
"transformer.h.21.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 122 |
+
"transformer.h.21.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 123 |
+
"transformer.h.21.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 124 |
+
"transformer.h.21.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 125 |
+
"transformer.h.21.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 126 |
+
"transformer.h.21.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 127 |
+
"transformer.h.22.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 128 |
+
"transformer.h.22.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 129 |
+
"transformer.h.22.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 130 |
+
"transformer.h.22.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 131 |
+
"transformer.h.22.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 132 |
+
"transformer.h.22.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 133 |
+
"transformer.h.22.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 134 |
+
"transformer.h.22.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 135 |
+
"transformer.h.23.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 136 |
+
"transformer.h.23.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 137 |
+
"transformer.h.23.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 138 |
+
"transformer.h.23.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 139 |
+
"transformer.h.23.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 140 |
+
"transformer.h.23.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 141 |
+
"transformer.h.23.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 142 |
+
"transformer.h.23.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 143 |
+
"transformer.h.24.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 144 |
+
"transformer.h.24.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 145 |
+
"transformer.h.24.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 146 |
+
"transformer.h.24.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 147 |
+
"transformer.h.24.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 148 |
+
"transformer.h.24.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 149 |
+
"transformer.h.24.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 150 |
+
"transformer.h.24.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 151 |
+
"transformer.h.25.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 152 |
+
"transformer.h.25.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 153 |
+
"transformer.h.25.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 154 |
+
"transformer.h.25.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 155 |
+
"transformer.h.25.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 156 |
+
"transformer.h.25.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 157 |
+
"transformer.h.25.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 158 |
+
"transformer.h.25.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 159 |
+
"transformer.h.26.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 160 |
+
"transformer.h.26.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 161 |
+
"transformer.h.26.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 162 |
+
"transformer.h.26.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 163 |
+
"transformer.h.26.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 164 |
+
"transformer.h.26.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 165 |
+
"transformer.h.26.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 166 |
+
"transformer.h.26.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 167 |
+
"transformer.h.27.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 168 |
+
"transformer.h.27.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 169 |
+
"transformer.h.27.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 170 |
+
"transformer.h.27.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 171 |
+
"transformer.h.27.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 172 |
+
"transformer.h.27.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 173 |
+
"transformer.h.27.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 174 |
+
"transformer.h.27.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 175 |
+
"transformer.h.28.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 176 |
+
"transformer.h.28.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 177 |
+
"transformer.h.28.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 178 |
+
"transformer.h.28.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 179 |
+
"transformer.h.28.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 180 |
+
"transformer.h.28.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 181 |
+
"transformer.h.28.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 182 |
+
"transformer.h.28.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 183 |
+
"transformer.h.29.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 184 |
+
"transformer.h.29.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 185 |
+
"transformer.h.29.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 186 |
+
"transformer.h.29.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 187 |
+
"transformer.h.29.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 188 |
+
"transformer.h.29.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 189 |
+
"transformer.h.29.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 190 |
+
"transformer.h.29.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 191 |
+
"transformer.h.3.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 192 |
+
"transformer.h.3.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 193 |
+
"transformer.h.3.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 194 |
+
"transformer.h.3.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 195 |
+
"transformer.h.3.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 196 |
+
"transformer.h.3.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 197 |
+
"transformer.h.3.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 198 |
+
"transformer.h.3.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 199 |
+
"transformer.h.30.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 200 |
+
"transformer.h.30.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 201 |
+
"transformer.h.30.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 202 |
+
"transformer.h.30.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 203 |
+
"transformer.h.30.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 204 |
+
"transformer.h.30.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 205 |
+
"transformer.h.30.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 206 |
+
"transformer.h.30.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 207 |
+
"transformer.h.31.attn.c_attn.bias": "pytorch_model-00002-of-00002.bin",
|
| 208 |
+
"transformer.h.31.attn.c_attn.weight": "pytorch_model-00002-of-00002.bin",
|
| 209 |
+
"transformer.h.31.attn.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 210 |
+
"transformer.h.31.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 211 |
+
"transformer.h.31.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 212 |
+
"transformer.h.31.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 213 |
+
"transformer.h.31.mlp.w1.weight": "pytorch_model-00002-of-00002.bin",
|
| 214 |
+
"transformer.h.31.mlp.w2.weight": "pytorch_model-00002-of-00002.bin",
|
| 215 |
+
"transformer.h.4.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 216 |
+
"transformer.h.4.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 217 |
+
"transformer.h.4.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 218 |
+
"transformer.h.4.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 219 |
+
"transformer.h.4.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 220 |
+
"transformer.h.4.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 221 |
+
"transformer.h.4.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 222 |
+
"transformer.h.4.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 223 |
+
"transformer.h.5.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 224 |
+
"transformer.h.5.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 225 |
+
"transformer.h.5.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 226 |
+
"transformer.h.5.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 227 |
+
"transformer.h.5.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 228 |
+
"transformer.h.5.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 229 |
+
"transformer.h.5.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 230 |
+
"transformer.h.5.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 231 |
+
"transformer.h.6.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 232 |
+
"transformer.h.6.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 233 |
+
"transformer.h.6.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 234 |
+
"transformer.h.6.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 235 |
+
"transformer.h.6.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 236 |
+
"transformer.h.6.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 237 |
+
"transformer.h.6.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 238 |
+
"transformer.h.6.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 239 |
+
"transformer.h.7.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 240 |
+
"transformer.h.7.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 241 |
+
"transformer.h.7.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 242 |
+
"transformer.h.7.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 243 |
+
"transformer.h.7.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 244 |
+
"transformer.h.7.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 245 |
+
"transformer.h.7.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 246 |
+
"transformer.h.7.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 247 |
+
"transformer.h.8.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 248 |
+
"transformer.h.8.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 249 |
+
"transformer.h.8.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 250 |
+
"transformer.h.8.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 251 |
+
"transformer.h.8.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 252 |
+
"transformer.h.8.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 253 |
+
"transformer.h.8.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 254 |
+
"transformer.h.8.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 255 |
+
"transformer.h.9.attn.c_attn.bias": "pytorch_model-00001-of-00002.bin",
|
| 256 |
+
"transformer.h.9.attn.c_attn.weight": "pytorch_model-00001-of-00002.bin",
|
| 257 |
+
"transformer.h.9.attn.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 258 |
+
"transformer.h.9.ln_1.weight": "pytorch_model-00001-of-00002.bin",
|
| 259 |
+
"transformer.h.9.ln_2.weight": "pytorch_model-00001-of-00002.bin",
|
| 260 |
+
"transformer.h.9.mlp.c_proj.weight": "pytorch_model-00001-of-00002.bin",
|
| 261 |
+
"transformer.h.9.mlp.w1.weight": "pytorch_model-00001-of-00002.bin",
|
| 262 |
+
"transformer.h.9.mlp.w2.weight": "pytorch_model-00001-of-00002.bin",
|
| 263 |
+
"transformer.ln_f.weight": "pytorch_model-00002-of-00002.bin",
|
| 264 |
+
"transformer.visual.attn_pool.attn.in_proj_bias": "pytorch_model-00002-of-00002.bin",
|
| 265 |
+
"transformer.visual.attn_pool.attn.in_proj_weight": "pytorch_model-00002-of-00002.bin",
|
| 266 |
+
"transformer.visual.attn_pool.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 267 |
+
"transformer.visual.attn_pool.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 268 |
+
"transformer.visual.attn_pool.kv_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 269 |
+
"transformer.visual.attn_pool.ln_kv.bias": "pytorch_model-00002-of-00002.bin",
|
| 270 |
+
"transformer.visual.attn_pool.ln_kv.weight": "pytorch_model-00002-of-00002.bin",
|
| 271 |
+
"transformer.visual.attn_pool.ln_q.bias": "pytorch_model-00002-of-00002.bin",
|
| 272 |
+
"transformer.visual.attn_pool.ln_q.weight": "pytorch_model-00002-of-00002.bin",
|
| 273 |
+
"transformer.visual.attn_pool.pos_embed": "pytorch_model-00002-of-00002.bin",
|
| 274 |
+
"transformer.visual.attn_pool.query": "pytorch_model-00002-of-00002.bin",
|
| 275 |
+
"transformer.visual.attn_pool2.attn.in_proj_bias": "pytorch_model-00002-of-00002.bin",
|
| 276 |
+
"transformer.visual.attn_pool2.attn.in_proj_weight": "pytorch_model-00002-of-00002.bin",
|
| 277 |
+
"transformer.visual.attn_pool2.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 278 |
+
"transformer.visual.attn_pool2.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 279 |
+
"transformer.visual.attn_pool2.kv_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 280 |
+
"transformer.visual.attn_pool2.ln_kv.bias": "pytorch_model-00002-of-00002.bin",
|
| 281 |
+
"transformer.visual.attn_pool2.ln_kv.weight": "pytorch_model-00002-of-00002.bin",
|
| 282 |
+
"transformer.visual.attn_pool2.ln_q.bias": "pytorch_model-00002-of-00002.bin",
|
| 283 |
+
"transformer.visual.attn_pool2.ln_q.weight": "pytorch_model-00002-of-00002.bin",
|
| 284 |
+
"transformer.visual.attn_pool2.pos_embed": "pytorch_model-00002-of-00002.bin",
|
| 285 |
+
"transformer.visual.attn_pool2.query": "pytorch_model-00002-of-00002.bin",
|
| 286 |
+
"transformer.visual.conv1.weight": "pytorch_model-00002-of-00002.bin",
|
| 287 |
+
"transformer.visual.ln_post.bias": "pytorch_model-00002-of-00002.bin",
|
| 288 |
+
"transformer.visual.ln_post.weight": "pytorch_model-00002-of-00002.bin",
|
| 289 |
+
"transformer.visual.ln_pre.bias": "pytorch_model-00002-of-00002.bin",
|
| 290 |
+
"transformer.visual.ln_pre.weight": "pytorch_model-00002-of-00002.bin",
|
| 291 |
+
"transformer.visual.positional_embedding": "pytorch_model-00002-of-00002.bin",
|
| 292 |
+
"transformer.visual.proj": "pytorch_model-00002-of-00002.bin",
|
| 293 |
+
"transformer.visual.transformer.resblocks.0.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 294 |
+
"transformer.visual.transformer.resblocks.0.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 295 |
+
"transformer.visual.transformer.resblocks.0.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 296 |
+
"transformer.visual.transformer.resblocks.0.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 297 |
+
"transformer.visual.transformer.resblocks.0.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 298 |
+
"transformer.visual.transformer.resblocks.0.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 299 |
+
"transformer.visual.transformer.resblocks.0.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 300 |
+
"transformer.visual.transformer.resblocks.0.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 301 |
+
"transformer.visual.transformer.resblocks.0.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 302 |
+
"transformer.visual.transformer.resblocks.0.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 303 |
+
"transformer.visual.transformer.resblocks.0.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 304 |
+
"transformer.visual.transformer.resblocks.0.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 305 |
+
"transformer.visual.transformer.resblocks.1.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 306 |
+
"transformer.visual.transformer.resblocks.1.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 307 |
+
"transformer.visual.transformer.resblocks.1.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 308 |
+
"transformer.visual.transformer.resblocks.1.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 309 |
+
"transformer.visual.transformer.resblocks.1.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 310 |
+
"transformer.visual.transformer.resblocks.1.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 311 |
+
"transformer.visual.transformer.resblocks.1.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 312 |
+
"transformer.visual.transformer.resblocks.1.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 313 |
+
"transformer.visual.transformer.resblocks.1.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 314 |
+
"transformer.visual.transformer.resblocks.1.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 315 |
+
"transformer.visual.transformer.resblocks.1.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 316 |
+
"transformer.visual.transformer.resblocks.1.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 317 |
+
"transformer.visual.transformer.resblocks.10.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 318 |
+
"transformer.visual.transformer.resblocks.10.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 319 |
+
"transformer.visual.transformer.resblocks.10.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 320 |
+
"transformer.visual.transformer.resblocks.10.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 321 |
+
"transformer.visual.transformer.resblocks.10.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 322 |
+
"transformer.visual.transformer.resblocks.10.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 323 |
+
"transformer.visual.transformer.resblocks.10.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 324 |
+
"transformer.visual.transformer.resblocks.10.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 325 |
+
"transformer.visual.transformer.resblocks.10.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 326 |
+
"transformer.visual.transformer.resblocks.10.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 327 |
+
"transformer.visual.transformer.resblocks.10.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 328 |
+
"transformer.visual.transformer.resblocks.10.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 329 |
+
"transformer.visual.transformer.resblocks.11.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 330 |
+
"transformer.visual.transformer.resblocks.11.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 331 |
+
"transformer.visual.transformer.resblocks.11.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 332 |
+
"transformer.visual.transformer.resblocks.11.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 333 |
+
"transformer.visual.transformer.resblocks.11.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 334 |
+
"transformer.visual.transformer.resblocks.11.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 335 |
+
"transformer.visual.transformer.resblocks.11.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 336 |
+
"transformer.visual.transformer.resblocks.11.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 337 |
+
"transformer.visual.transformer.resblocks.11.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 338 |
+
"transformer.visual.transformer.resblocks.11.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 339 |
+
"transformer.visual.transformer.resblocks.11.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 340 |
+
"transformer.visual.transformer.resblocks.11.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 341 |
+
"transformer.visual.transformer.resblocks.12.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 342 |
+
"transformer.visual.transformer.resblocks.12.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 343 |
+
"transformer.visual.transformer.resblocks.12.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 344 |
+
"transformer.visual.transformer.resblocks.12.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 345 |
+
"transformer.visual.transformer.resblocks.12.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 346 |
+
"transformer.visual.transformer.resblocks.12.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 347 |
+
"transformer.visual.transformer.resblocks.12.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 348 |
+
"transformer.visual.transformer.resblocks.12.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 349 |
+
"transformer.visual.transformer.resblocks.12.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 350 |
+
"transformer.visual.transformer.resblocks.12.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 351 |
+
"transformer.visual.transformer.resblocks.12.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 352 |
+
"transformer.visual.transformer.resblocks.12.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 353 |
+
"transformer.visual.transformer.resblocks.13.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 354 |
+
"transformer.visual.transformer.resblocks.13.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 355 |
+
"transformer.visual.transformer.resblocks.13.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 356 |
+
"transformer.visual.transformer.resblocks.13.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 357 |
+
"transformer.visual.transformer.resblocks.13.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 358 |
+
"transformer.visual.transformer.resblocks.13.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 359 |
+
"transformer.visual.transformer.resblocks.13.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 360 |
+
"transformer.visual.transformer.resblocks.13.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 361 |
+
"transformer.visual.transformer.resblocks.13.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 362 |
+
"transformer.visual.transformer.resblocks.13.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 363 |
+
"transformer.visual.transformer.resblocks.13.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 364 |
+
"transformer.visual.transformer.resblocks.13.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 365 |
+
"transformer.visual.transformer.resblocks.14.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 366 |
+
"transformer.visual.transformer.resblocks.14.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 367 |
+
"transformer.visual.transformer.resblocks.14.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 368 |
+
"transformer.visual.transformer.resblocks.14.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 369 |
+
"transformer.visual.transformer.resblocks.14.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 370 |
+
"transformer.visual.transformer.resblocks.14.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 371 |
+
"transformer.visual.transformer.resblocks.14.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 372 |
+
"transformer.visual.transformer.resblocks.14.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 373 |
+
"transformer.visual.transformer.resblocks.14.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 374 |
+
"transformer.visual.transformer.resblocks.14.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 375 |
+
"transformer.visual.transformer.resblocks.14.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 376 |
+
"transformer.visual.transformer.resblocks.14.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 377 |
+
"transformer.visual.transformer.resblocks.15.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 378 |
+
"transformer.visual.transformer.resblocks.15.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 379 |
+
"transformer.visual.transformer.resblocks.15.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 380 |
+
"transformer.visual.transformer.resblocks.15.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 381 |
+
"transformer.visual.transformer.resblocks.15.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 382 |
+
"transformer.visual.transformer.resblocks.15.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 383 |
+
"transformer.visual.transformer.resblocks.15.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 384 |
+
"transformer.visual.transformer.resblocks.15.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 385 |
+
"transformer.visual.transformer.resblocks.15.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 386 |
+
"transformer.visual.transformer.resblocks.15.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 387 |
+
"transformer.visual.transformer.resblocks.15.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 388 |
+
"transformer.visual.transformer.resblocks.15.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 389 |
+
"transformer.visual.transformer.resblocks.16.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 390 |
+
"transformer.visual.transformer.resblocks.16.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 391 |
+
"transformer.visual.transformer.resblocks.16.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 392 |
+
"transformer.visual.transformer.resblocks.16.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 393 |
+
"transformer.visual.transformer.resblocks.16.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 394 |
+
"transformer.visual.transformer.resblocks.16.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 395 |
+
"transformer.visual.transformer.resblocks.16.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 396 |
+
"transformer.visual.transformer.resblocks.16.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 397 |
+
"transformer.visual.transformer.resblocks.16.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 398 |
+
"transformer.visual.transformer.resblocks.16.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 399 |
+
"transformer.visual.transformer.resblocks.16.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 400 |
+
"transformer.visual.transformer.resblocks.16.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 401 |
+
"transformer.visual.transformer.resblocks.17.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 402 |
+
"transformer.visual.transformer.resblocks.17.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 403 |
+
"transformer.visual.transformer.resblocks.17.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 404 |
+
"transformer.visual.transformer.resblocks.17.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 405 |
+
"transformer.visual.transformer.resblocks.17.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 406 |
+
"transformer.visual.transformer.resblocks.17.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 407 |
+
"transformer.visual.transformer.resblocks.17.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 408 |
+
"transformer.visual.transformer.resblocks.17.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 409 |
+
"transformer.visual.transformer.resblocks.17.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 410 |
+
"transformer.visual.transformer.resblocks.17.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 411 |
+
"transformer.visual.transformer.resblocks.17.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 412 |
+
"transformer.visual.transformer.resblocks.17.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 413 |
+
"transformer.visual.transformer.resblocks.18.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 414 |
+
"transformer.visual.transformer.resblocks.18.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 415 |
+
"transformer.visual.transformer.resblocks.18.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 416 |
+
"transformer.visual.transformer.resblocks.18.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 417 |
+
"transformer.visual.transformer.resblocks.18.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 418 |
+
"transformer.visual.transformer.resblocks.18.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 419 |
+
"transformer.visual.transformer.resblocks.18.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 420 |
+
"transformer.visual.transformer.resblocks.18.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 421 |
+
"transformer.visual.transformer.resblocks.18.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 422 |
+
"transformer.visual.transformer.resblocks.18.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 423 |
+
"transformer.visual.transformer.resblocks.18.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 424 |
+
"transformer.visual.transformer.resblocks.18.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 425 |
+
"transformer.visual.transformer.resblocks.19.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 426 |
+
"transformer.visual.transformer.resblocks.19.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 427 |
+
"transformer.visual.transformer.resblocks.19.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 428 |
+
"transformer.visual.transformer.resblocks.19.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 429 |
+
"transformer.visual.transformer.resblocks.19.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 430 |
+
"transformer.visual.transformer.resblocks.19.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 431 |
+
"transformer.visual.transformer.resblocks.19.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 432 |
+
"transformer.visual.transformer.resblocks.19.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 433 |
+
"transformer.visual.transformer.resblocks.19.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 434 |
+
"transformer.visual.transformer.resblocks.19.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 435 |
+
"transformer.visual.transformer.resblocks.19.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 436 |
+
"transformer.visual.transformer.resblocks.19.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 437 |
+
"transformer.visual.transformer.resblocks.2.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 438 |
+
"transformer.visual.transformer.resblocks.2.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 439 |
+
"transformer.visual.transformer.resblocks.2.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 440 |
+
"transformer.visual.transformer.resblocks.2.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 441 |
+
"transformer.visual.transformer.resblocks.2.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 442 |
+
"transformer.visual.transformer.resblocks.2.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 443 |
+
"transformer.visual.transformer.resblocks.2.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 444 |
+
"transformer.visual.transformer.resblocks.2.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 445 |
+
"transformer.visual.transformer.resblocks.2.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 446 |
+
"transformer.visual.transformer.resblocks.2.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 447 |
+
"transformer.visual.transformer.resblocks.2.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 448 |
+
"transformer.visual.transformer.resblocks.2.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 449 |
+
"transformer.visual.transformer.resblocks.20.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 450 |
+
"transformer.visual.transformer.resblocks.20.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 451 |
+
"transformer.visual.transformer.resblocks.20.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 452 |
+
"transformer.visual.transformer.resblocks.20.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 453 |
+
"transformer.visual.transformer.resblocks.20.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 454 |
+
"transformer.visual.transformer.resblocks.20.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 455 |
+
"transformer.visual.transformer.resblocks.20.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 456 |
+
"transformer.visual.transformer.resblocks.20.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 457 |
+
"transformer.visual.transformer.resblocks.20.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 458 |
+
"transformer.visual.transformer.resblocks.20.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 459 |
+
"transformer.visual.transformer.resblocks.20.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 460 |
+
"transformer.visual.transformer.resblocks.20.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 461 |
+
"transformer.visual.transformer.resblocks.21.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 462 |
+
"transformer.visual.transformer.resblocks.21.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 463 |
+
"transformer.visual.transformer.resblocks.21.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 464 |
+
"transformer.visual.transformer.resblocks.21.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 465 |
+
"transformer.visual.transformer.resblocks.21.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 466 |
+
"transformer.visual.transformer.resblocks.21.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 467 |
+
"transformer.visual.transformer.resblocks.21.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 468 |
+
"transformer.visual.transformer.resblocks.21.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 469 |
+
"transformer.visual.transformer.resblocks.21.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 470 |
+
"transformer.visual.transformer.resblocks.21.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 471 |
+
"transformer.visual.transformer.resblocks.21.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 472 |
+
"transformer.visual.transformer.resblocks.21.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 473 |
+
"transformer.visual.transformer.resblocks.22.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 474 |
+
"transformer.visual.transformer.resblocks.22.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 475 |
+
"transformer.visual.transformer.resblocks.22.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 476 |
+
"transformer.visual.transformer.resblocks.22.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 477 |
+
"transformer.visual.transformer.resblocks.22.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 478 |
+
"transformer.visual.transformer.resblocks.22.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 479 |
+
"transformer.visual.transformer.resblocks.22.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 480 |
+
"transformer.visual.transformer.resblocks.22.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 481 |
+
"transformer.visual.transformer.resblocks.22.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 482 |
+
"transformer.visual.transformer.resblocks.22.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 483 |
+
"transformer.visual.transformer.resblocks.22.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 484 |
+
"transformer.visual.transformer.resblocks.22.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 485 |
+
"transformer.visual.transformer.resblocks.23.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 486 |
+
"transformer.visual.transformer.resblocks.23.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 487 |
+
"transformer.visual.transformer.resblocks.23.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 488 |
+
"transformer.visual.transformer.resblocks.23.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 489 |
+
"transformer.visual.transformer.resblocks.23.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 490 |
+
"transformer.visual.transformer.resblocks.23.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 491 |
+
"transformer.visual.transformer.resblocks.23.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 492 |
+
"transformer.visual.transformer.resblocks.23.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 493 |
+
"transformer.visual.transformer.resblocks.23.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 494 |
+
"transformer.visual.transformer.resblocks.23.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 495 |
+
"transformer.visual.transformer.resblocks.23.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 496 |
+
"transformer.visual.transformer.resblocks.23.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 497 |
+
"transformer.visual.transformer.resblocks.24.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 498 |
+
"transformer.visual.transformer.resblocks.24.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 499 |
+
"transformer.visual.transformer.resblocks.24.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 500 |
+
"transformer.visual.transformer.resblocks.24.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 501 |
+
"transformer.visual.transformer.resblocks.24.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 502 |
+
"transformer.visual.transformer.resblocks.24.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 503 |
+
"transformer.visual.transformer.resblocks.24.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 504 |
+
"transformer.visual.transformer.resblocks.24.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 505 |
+
"transformer.visual.transformer.resblocks.24.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 506 |
+
"transformer.visual.transformer.resblocks.24.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 507 |
+
"transformer.visual.transformer.resblocks.24.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 508 |
+
"transformer.visual.transformer.resblocks.24.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 509 |
+
"transformer.visual.transformer.resblocks.25.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 510 |
+
"transformer.visual.transformer.resblocks.25.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 511 |
+
"transformer.visual.transformer.resblocks.25.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 512 |
+
"transformer.visual.transformer.resblocks.25.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 513 |
+
"transformer.visual.transformer.resblocks.25.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 514 |
+
"transformer.visual.transformer.resblocks.25.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 515 |
+
"transformer.visual.transformer.resblocks.25.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 516 |
+
"transformer.visual.transformer.resblocks.25.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 517 |
+
"transformer.visual.transformer.resblocks.25.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 518 |
+
"transformer.visual.transformer.resblocks.25.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 519 |
+
"transformer.visual.transformer.resblocks.25.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 520 |
+
"transformer.visual.transformer.resblocks.25.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 521 |
+
"transformer.visual.transformer.resblocks.26.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 522 |
+
"transformer.visual.transformer.resblocks.26.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 523 |
+
"transformer.visual.transformer.resblocks.26.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 524 |
+
"transformer.visual.transformer.resblocks.26.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 525 |
+
"transformer.visual.transformer.resblocks.26.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 526 |
+
"transformer.visual.transformer.resblocks.26.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 527 |
+
"transformer.visual.transformer.resblocks.26.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 528 |
+
"transformer.visual.transformer.resblocks.26.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 529 |
+
"transformer.visual.transformer.resblocks.26.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 530 |
+
"transformer.visual.transformer.resblocks.26.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 531 |
+
"transformer.visual.transformer.resblocks.26.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 532 |
+
"transformer.visual.transformer.resblocks.26.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 533 |
+
"transformer.visual.transformer.resblocks.27.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 534 |
+
"transformer.visual.transformer.resblocks.27.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 535 |
+
"transformer.visual.transformer.resblocks.27.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 536 |
+
"transformer.visual.transformer.resblocks.27.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 537 |
+
"transformer.visual.transformer.resblocks.27.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 538 |
+
"transformer.visual.transformer.resblocks.27.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 539 |
+
"transformer.visual.transformer.resblocks.27.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 540 |
+
"transformer.visual.transformer.resblocks.27.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 541 |
+
"transformer.visual.transformer.resblocks.27.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 542 |
+
"transformer.visual.transformer.resblocks.27.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 543 |
+
"transformer.visual.transformer.resblocks.27.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 544 |
+
"transformer.visual.transformer.resblocks.27.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 545 |
+
"transformer.visual.transformer.resblocks.28.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 546 |
+
"transformer.visual.transformer.resblocks.28.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 547 |
+
"transformer.visual.transformer.resblocks.28.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 548 |
+
"transformer.visual.transformer.resblocks.28.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 549 |
+
"transformer.visual.transformer.resblocks.28.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 550 |
+
"transformer.visual.transformer.resblocks.28.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 551 |
+
"transformer.visual.transformer.resblocks.28.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 552 |
+
"transformer.visual.transformer.resblocks.28.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 553 |
+
"transformer.visual.transformer.resblocks.28.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 554 |
+
"transformer.visual.transformer.resblocks.28.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 555 |
+
"transformer.visual.transformer.resblocks.28.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 556 |
+
"transformer.visual.transformer.resblocks.28.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 557 |
+
"transformer.visual.transformer.resblocks.29.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 558 |
+
"transformer.visual.transformer.resblocks.29.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 559 |
+
"transformer.visual.transformer.resblocks.29.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 560 |
+
"transformer.visual.transformer.resblocks.29.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 561 |
+
"transformer.visual.transformer.resblocks.29.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 562 |
+
"transformer.visual.transformer.resblocks.29.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 563 |
+
"transformer.visual.transformer.resblocks.29.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 564 |
+
"transformer.visual.transformer.resblocks.29.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 565 |
+
"transformer.visual.transformer.resblocks.29.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 566 |
+
"transformer.visual.transformer.resblocks.29.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 567 |
+
"transformer.visual.transformer.resblocks.29.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 568 |
+
"transformer.visual.transformer.resblocks.29.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 569 |
+
"transformer.visual.transformer.resblocks.3.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 570 |
+
"transformer.visual.transformer.resblocks.3.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 571 |
+
"transformer.visual.transformer.resblocks.3.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 572 |
+
"transformer.visual.transformer.resblocks.3.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 573 |
+
"transformer.visual.transformer.resblocks.3.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 574 |
+
"transformer.visual.transformer.resblocks.3.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 575 |
+
"transformer.visual.transformer.resblocks.3.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 576 |
+
"transformer.visual.transformer.resblocks.3.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 577 |
+
"transformer.visual.transformer.resblocks.3.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 578 |
+
"transformer.visual.transformer.resblocks.3.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 579 |
+
"transformer.visual.transformer.resblocks.3.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 580 |
+
"transformer.visual.transformer.resblocks.3.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 581 |
+
"transformer.visual.transformer.resblocks.30.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 582 |
+
"transformer.visual.transformer.resblocks.30.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 583 |
+
"transformer.visual.transformer.resblocks.30.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 584 |
+
"transformer.visual.transformer.resblocks.30.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 585 |
+
"transformer.visual.transformer.resblocks.30.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 586 |
+
"transformer.visual.transformer.resblocks.30.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 587 |
+
"transformer.visual.transformer.resblocks.30.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 588 |
+
"transformer.visual.transformer.resblocks.30.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 589 |
+
"transformer.visual.transformer.resblocks.30.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 590 |
+
"transformer.visual.transformer.resblocks.30.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 591 |
+
"transformer.visual.transformer.resblocks.30.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 592 |
+
"transformer.visual.transformer.resblocks.30.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 593 |
+
"transformer.visual.transformer.resblocks.31.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 594 |
+
"transformer.visual.transformer.resblocks.31.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 595 |
+
"transformer.visual.transformer.resblocks.31.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 596 |
+
"transformer.visual.transformer.resblocks.31.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 597 |
+
"transformer.visual.transformer.resblocks.31.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 598 |
+
"transformer.visual.transformer.resblocks.31.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 599 |
+
"transformer.visual.transformer.resblocks.31.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 600 |
+
"transformer.visual.transformer.resblocks.31.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 601 |
+
"transformer.visual.transformer.resblocks.31.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 602 |
+
"transformer.visual.transformer.resblocks.31.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 603 |
+
"transformer.visual.transformer.resblocks.31.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 604 |
+
"transformer.visual.transformer.resblocks.31.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 605 |
+
"transformer.visual.transformer.resblocks.32.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 606 |
+
"transformer.visual.transformer.resblocks.32.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 607 |
+
"transformer.visual.transformer.resblocks.32.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 608 |
+
"transformer.visual.transformer.resblocks.32.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 609 |
+
"transformer.visual.transformer.resblocks.32.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 610 |
+
"transformer.visual.transformer.resblocks.32.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 611 |
+
"transformer.visual.transformer.resblocks.32.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 612 |
+
"transformer.visual.transformer.resblocks.32.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 613 |
+
"transformer.visual.transformer.resblocks.32.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 614 |
+
"transformer.visual.transformer.resblocks.32.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 615 |
+
"transformer.visual.transformer.resblocks.32.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 616 |
+
"transformer.visual.transformer.resblocks.32.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 617 |
+
"transformer.visual.transformer.resblocks.33.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 618 |
+
"transformer.visual.transformer.resblocks.33.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 619 |
+
"transformer.visual.transformer.resblocks.33.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 620 |
+
"transformer.visual.transformer.resblocks.33.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 621 |
+
"transformer.visual.transformer.resblocks.33.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 622 |
+
"transformer.visual.transformer.resblocks.33.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 623 |
+
"transformer.visual.transformer.resblocks.33.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 624 |
+
"transformer.visual.transformer.resblocks.33.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 625 |
+
"transformer.visual.transformer.resblocks.33.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 626 |
+
"transformer.visual.transformer.resblocks.33.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 627 |
+
"transformer.visual.transformer.resblocks.33.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 628 |
+
"transformer.visual.transformer.resblocks.33.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 629 |
+
"transformer.visual.transformer.resblocks.34.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 630 |
+
"transformer.visual.transformer.resblocks.34.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 631 |
+
"transformer.visual.transformer.resblocks.34.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 632 |
+
"transformer.visual.transformer.resblocks.34.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 633 |
+
"transformer.visual.transformer.resblocks.34.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 634 |
+
"transformer.visual.transformer.resblocks.34.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 635 |
+
"transformer.visual.transformer.resblocks.34.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 636 |
+
"transformer.visual.transformer.resblocks.34.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 637 |
+
"transformer.visual.transformer.resblocks.34.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 638 |
+
"transformer.visual.transformer.resblocks.34.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 639 |
+
"transformer.visual.transformer.resblocks.34.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 640 |
+
"transformer.visual.transformer.resblocks.34.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 641 |
+
"transformer.visual.transformer.resblocks.35.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 642 |
+
"transformer.visual.transformer.resblocks.35.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 643 |
+
"transformer.visual.transformer.resblocks.35.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 644 |
+
"transformer.visual.transformer.resblocks.35.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 645 |
+
"transformer.visual.transformer.resblocks.35.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 646 |
+
"transformer.visual.transformer.resblocks.35.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 647 |
+
"transformer.visual.transformer.resblocks.35.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 648 |
+
"transformer.visual.transformer.resblocks.35.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 649 |
+
"transformer.visual.transformer.resblocks.35.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 650 |
+
"transformer.visual.transformer.resblocks.35.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 651 |
+
"transformer.visual.transformer.resblocks.35.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 652 |
+
"transformer.visual.transformer.resblocks.35.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 653 |
+
"transformer.visual.transformer.resblocks.36.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 654 |
+
"transformer.visual.transformer.resblocks.36.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 655 |
+
"transformer.visual.transformer.resblocks.36.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 656 |
+
"transformer.visual.transformer.resblocks.36.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 657 |
+
"transformer.visual.transformer.resblocks.36.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 658 |
+
"transformer.visual.transformer.resblocks.36.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 659 |
+
"transformer.visual.transformer.resblocks.36.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 660 |
+
"transformer.visual.transformer.resblocks.36.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 661 |
+
"transformer.visual.transformer.resblocks.36.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 662 |
+
"transformer.visual.transformer.resblocks.36.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 663 |
+
"transformer.visual.transformer.resblocks.36.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 664 |
+
"transformer.visual.transformer.resblocks.36.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 665 |
+
"transformer.visual.transformer.resblocks.37.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 666 |
+
"transformer.visual.transformer.resblocks.37.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 667 |
+
"transformer.visual.transformer.resblocks.37.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 668 |
+
"transformer.visual.transformer.resblocks.37.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 669 |
+
"transformer.visual.transformer.resblocks.37.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 670 |
+
"transformer.visual.transformer.resblocks.37.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 671 |
+
"transformer.visual.transformer.resblocks.37.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 672 |
+
"transformer.visual.transformer.resblocks.37.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 673 |
+
"transformer.visual.transformer.resblocks.37.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 674 |
+
"transformer.visual.transformer.resblocks.37.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 675 |
+
"transformer.visual.transformer.resblocks.37.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 676 |
+
"transformer.visual.transformer.resblocks.37.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 677 |
+
"transformer.visual.transformer.resblocks.38.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 678 |
+
"transformer.visual.transformer.resblocks.38.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 679 |
+
"transformer.visual.transformer.resblocks.38.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 680 |
+
"transformer.visual.transformer.resblocks.38.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 681 |
+
"transformer.visual.transformer.resblocks.38.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 682 |
+
"transformer.visual.transformer.resblocks.38.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 683 |
+
"transformer.visual.transformer.resblocks.38.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 684 |
+
"transformer.visual.transformer.resblocks.38.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 685 |
+
"transformer.visual.transformer.resblocks.38.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 686 |
+
"transformer.visual.transformer.resblocks.38.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 687 |
+
"transformer.visual.transformer.resblocks.38.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 688 |
+
"transformer.visual.transformer.resblocks.38.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 689 |
+
"transformer.visual.transformer.resblocks.39.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 690 |
+
"transformer.visual.transformer.resblocks.39.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 691 |
+
"transformer.visual.transformer.resblocks.39.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 692 |
+
"transformer.visual.transformer.resblocks.39.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 693 |
+
"transformer.visual.transformer.resblocks.39.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 694 |
+
"transformer.visual.transformer.resblocks.39.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 695 |
+
"transformer.visual.transformer.resblocks.39.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 696 |
+
"transformer.visual.transformer.resblocks.39.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 697 |
+
"transformer.visual.transformer.resblocks.39.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 698 |
+
"transformer.visual.transformer.resblocks.39.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 699 |
+
"transformer.visual.transformer.resblocks.39.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 700 |
+
"transformer.visual.transformer.resblocks.39.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 701 |
+
"transformer.visual.transformer.resblocks.4.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 702 |
+
"transformer.visual.transformer.resblocks.4.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 703 |
+
"transformer.visual.transformer.resblocks.4.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 704 |
+
"transformer.visual.transformer.resblocks.4.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 705 |
+
"transformer.visual.transformer.resblocks.4.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 706 |
+
"transformer.visual.transformer.resblocks.4.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 707 |
+
"transformer.visual.transformer.resblocks.4.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 708 |
+
"transformer.visual.transformer.resblocks.4.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 709 |
+
"transformer.visual.transformer.resblocks.4.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 710 |
+
"transformer.visual.transformer.resblocks.4.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 711 |
+
"transformer.visual.transformer.resblocks.4.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 712 |
+
"transformer.visual.transformer.resblocks.4.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 713 |
+
"transformer.visual.transformer.resblocks.40.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 714 |
+
"transformer.visual.transformer.resblocks.40.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 715 |
+
"transformer.visual.transformer.resblocks.40.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 716 |
+
"transformer.visual.transformer.resblocks.40.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 717 |
+
"transformer.visual.transformer.resblocks.40.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 718 |
+
"transformer.visual.transformer.resblocks.40.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 719 |
+
"transformer.visual.transformer.resblocks.40.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 720 |
+
"transformer.visual.transformer.resblocks.40.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 721 |
+
"transformer.visual.transformer.resblocks.40.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 722 |
+
"transformer.visual.transformer.resblocks.40.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 723 |
+
"transformer.visual.transformer.resblocks.40.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 724 |
+
"transformer.visual.transformer.resblocks.40.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 725 |
+
"transformer.visual.transformer.resblocks.41.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 726 |
+
"transformer.visual.transformer.resblocks.41.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 727 |
+
"transformer.visual.transformer.resblocks.41.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 728 |
+
"transformer.visual.transformer.resblocks.41.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 729 |
+
"transformer.visual.transformer.resblocks.41.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 730 |
+
"transformer.visual.transformer.resblocks.41.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 731 |
+
"transformer.visual.transformer.resblocks.41.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 732 |
+
"transformer.visual.transformer.resblocks.41.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 733 |
+
"transformer.visual.transformer.resblocks.41.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 734 |
+
"transformer.visual.transformer.resblocks.41.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 735 |
+
"transformer.visual.transformer.resblocks.41.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 736 |
+
"transformer.visual.transformer.resblocks.41.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 737 |
+
"transformer.visual.transformer.resblocks.42.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 738 |
+
"transformer.visual.transformer.resblocks.42.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 739 |
+
"transformer.visual.transformer.resblocks.42.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 740 |
+
"transformer.visual.transformer.resblocks.42.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 741 |
+
"transformer.visual.transformer.resblocks.42.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 742 |
+
"transformer.visual.transformer.resblocks.42.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 743 |
+
"transformer.visual.transformer.resblocks.42.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 744 |
+
"transformer.visual.transformer.resblocks.42.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 745 |
+
"transformer.visual.transformer.resblocks.42.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 746 |
+
"transformer.visual.transformer.resblocks.42.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 747 |
+
"transformer.visual.transformer.resblocks.42.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 748 |
+
"transformer.visual.transformer.resblocks.42.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 749 |
+
"transformer.visual.transformer.resblocks.43.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 750 |
+
"transformer.visual.transformer.resblocks.43.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 751 |
+
"transformer.visual.transformer.resblocks.43.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 752 |
+
"transformer.visual.transformer.resblocks.43.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 753 |
+
"transformer.visual.transformer.resblocks.43.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 754 |
+
"transformer.visual.transformer.resblocks.43.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 755 |
+
"transformer.visual.transformer.resblocks.43.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 756 |
+
"transformer.visual.transformer.resblocks.43.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 757 |
+
"transformer.visual.transformer.resblocks.43.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 758 |
+
"transformer.visual.transformer.resblocks.43.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 759 |
+
"transformer.visual.transformer.resblocks.43.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 760 |
+
"transformer.visual.transformer.resblocks.43.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 761 |
+
"transformer.visual.transformer.resblocks.44.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 762 |
+
"transformer.visual.transformer.resblocks.44.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 763 |
+
"transformer.visual.transformer.resblocks.44.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 764 |
+
"transformer.visual.transformer.resblocks.44.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 765 |
+
"transformer.visual.transformer.resblocks.44.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 766 |
+
"transformer.visual.transformer.resblocks.44.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 767 |
+
"transformer.visual.transformer.resblocks.44.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 768 |
+
"transformer.visual.transformer.resblocks.44.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 769 |
+
"transformer.visual.transformer.resblocks.44.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 770 |
+
"transformer.visual.transformer.resblocks.44.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 771 |
+
"transformer.visual.transformer.resblocks.44.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 772 |
+
"transformer.visual.transformer.resblocks.44.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 773 |
+
"transformer.visual.transformer.resblocks.45.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 774 |
+
"transformer.visual.transformer.resblocks.45.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 775 |
+
"transformer.visual.transformer.resblocks.45.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 776 |
+
"transformer.visual.transformer.resblocks.45.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 777 |
+
"transformer.visual.transformer.resblocks.45.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 778 |
+
"transformer.visual.transformer.resblocks.45.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 779 |
+
"transformer.visual.transformer.resblocks.45.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 780 |
+
"transformer.visual.transformer.resblocks.45.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 781 |
+
"transformer.visual.transformer.resblocks.45.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 782 |
+
"transformer.visual.transformer.resblocks.45.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 783 |
+
"transformer.visual.transformer.resblocks.45.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 784 |
+
"transformer.visual.transformer.resblocks.45.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 785 |
+
"transformer.visual.transformer.resblocks.46.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 786 |
+
"transformer.visual.transformer.resblocks.46.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 787 |
+
"transformer.visual.transformer.resblocks.46.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 788 |
+
"transformer.visual.transformer.resblocks.46.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 789 |
+
"transformer.visual.transformer.resblocks.46.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 790 |
+
"transformer.visual.transformer.resblocks.46.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 791 |
+
"transformer.visual.transformer.resblocks.46.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 792 |
+
"transformer.visual.transformer.resblocks.46.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 793 |
+
"transformer.visual.transformer.resblocks.46.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 794 |
+
"transformer.visual.transformer.resblocks.46.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 795 |
+
"transformer.visual.transformer.resblocks.46.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 796 |
+
"transformer.visual.transformer.resblocks.46.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 797 |
+
"transformer.visual.transformer.resblocks.47.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 798 |
+
"transformer.visual.transformer.resblocks.47.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 799 |
+
"transformer.visual.transformer.resblocks.47.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 800 |
+
"transformer.visual.transformer.resblocks.47.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 801 |
+
"transformer.visual.transformer.resblocks.47.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 802 |
+
"transformer.visual.transformer.resblocks.47.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 803 |
+
"transformer.visual.transformer.resblocks.47.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 804 |
+
"transformer.visual.transformer.resblocks.47.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 805 |
+
"transformer.visual.transformer.resblocks.47.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 806 |
+
"transformer.visual.transformer.resblocks.47.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 807 |
+
"transformer.visual.transformer.resblocks.47.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 808 |
+
"transformer.visual.transformer.resblocks.47.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 809 |
+
"transformer.visual.transformer.resblocks.5.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 810 |
+
"transformer.visual.transformer.resblocks.5.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 811 |
+
"transformer.visual.transformer.resblocks.5.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 812 |
+
"transformer.visual.transformer.resblocks.5.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 813 |
+
"transformer.visual.transformer.resblocks.5.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 814 |
+
"transformer.visual.transformer.resblocks.5.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 815 |
+
"transformer.visual.transformer.resblocks.5.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 816 |
+
"transformer.visual.transformer.resblocks.5.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 817 |
+
"transformer.visual.transformer.resblocks.5.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 818 |
+
"transformer.visual.transformer.resblocks.5.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 819 |
+
"transformer.visual.transformer.resblocks.5.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 820 |
+
"transformer.visual.transformer.resblocks.5.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 821 |
+
"transformer.visual.transformer.resblocks.6.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 822 |
+
"transformer.visual.transformer.resblocks.6.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 823 |
+
"transformer.visual.transformer.resblocks.6.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 824 |
+
"transformer.visual.transformer.resblocks.6.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 825 |
+
"transformer.visual.transformer.resblocks.6.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 826 |
+
"transformer.visual.transformer.resblocks.6.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 827 |
+
"transformer.visual.transformer.resblocks.6.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 828 |
+
"transformer.visual.transformer.resblocks.6.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 829 |
+
"transformer.visual.transformer.resblocks.6.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 830 |
+
"transformer.visual.transformer.resblocks.6.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 831 |
+
"transformer.visual.transformer.resblocks.6.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 832 |
+
"transformer.visual.transformer.resblocks.6.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 833 |
+
"transformer.visual.transformer.resblocks.7.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 834 |
+
"transformer.visual.transformer.resblocks.7.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 835 |
+
"transformer.visual.transformer.resblocks.7.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 836 |
+
"transformer.visual.transformer.resblocks.7.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 837 |
+
"transformer.visual.transformer.resblocks.7.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 838 |
+
"transformer.visual.transformer.resblocks.7.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 839 |
+
"transformer.visual.transformer.resblocks.7.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 840 |
+
"transformer.visual.transformer.resblocks.7.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 841 |
+
"transformer.visual.transformer.resblocks.7.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 842 |
+
"transformer.visual.transformer.resblocks.7.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 843 |
+
"transformer.visual.transformer.resblocks.7.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 844 |
+
"transformer.visual.transformer.resblocks.7.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 845 |
+
"transformer.visual.transformer.resblocks.8.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 846 |
+
"transformer.visual.transformer.resblocks.8.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 847 |
+
"transformer.visual.transformer.resblocks.8.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 848 |
+
"transformer.visual.transformer.resblocks.8.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 849 |
+
"transformer.visual.transformer.resblocks.8.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 850 |
+
"transformer.visual.transformer.resblocks.8.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 851 |
+
"transformer.visual.transformer.resblocks.8.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 852 |
+
"transformer.visual.transformer.resblocks.8.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 853 |
+
"transformer.visual.transformer.resblocks.8.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 854 |
+
"transformer.visual.transformer.resblocks.8.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 855 |
+
"transformer.visual.transformer.resblocks.8.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 856 |
+
"transformer.visual.transformer.resblocks.8.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 857 |
+
"transformer.visual.transformer.resblocks.9.attn.in_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 858 |
+
"transformer.visual.transformer.resblocks.9.attn.in_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 859 |
+
"transformer.visual.transformer.resblocks.9.attn.out_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 860 |
+
"transformer.visual.transformer.resblocks.9.attn.out_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 861 |
+
"transformer.visual.transformer.resblocks.9.ln_1.bias": "pytorch_model-00002-of-00002.bin",
|
| 862 |
+
"transformer.visual.transformer.resblocks.9.ln_1.weight": "pytorch_model-00002-of-00002.bin",
|
| 863 |
+
"transformer.visual.transformer.resblocks.9.ln_2.bias": "pytorch_model-00002-of-00002.bin",
|
| 864 |
+
"transformer.visual.transformer.resblocks.9.ln_2.weight": "pytorch_model-00002-of-00002.bin",
|
| 865 |
+
"transformer.visual.transformer.resblocks.9.mlp.c_fc.bias": "pytorch_model-00002-of-00002.bin",
|
| 866 |
+
"transformer.visual.transformer.resblocks.9.mlp.c_fc.weight": "pytorch_model-00002-of-00002.bin",
|
| 867 |
+
"transformer.visual.transformer.resblocks.9.mlp.c_proj.bias": "pytorch_model-00002-of-00002.bin",
|
| 868 |
+
"transformer.visual.transformer.resblocks.9.mlp.c_proj.weight": "pytorch_model-00002-of-00002.bin",
|
| 869 |
+
"transformer.wte.weight": "pytorch_model-00001-of-00002.bin"
|
| 870 |
+
}
|
| 871 |
+
}
|
tokenization_qwen.py
ADDED
|
@@ -0,0 +1,593 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Copyright (c) Alibaba Cloud.
|
| 2 |
+
#
|
| 3 |
+
# This source code is licensed under the license found in the
|
| 4 |
+
# LICENSE file in the root directory of this source tree.
|
| 5 |
+
|
| 6 |
+
"""Tokenization classes for QWen."""
|
| 7 |
+
|
| 8 |
+
import base64
|
| 9 |
+
import logging
|
| 10 |
+
import os
|
| 11 |
+
import requests
|
| 12 |
+
import unicodedata
|
| 13 |
+
from typing import Collection, Dict, List, Set, Tuple, Union, Any, Callable, Optional
|
| 14 |
+
import pdb
|
| 15 |
+
import tiktoken
|
| 16 |
+
import numpy as np
|
| 17 |
+
from PIL import Image
|
| 18 |
+
from PIL import ImageFont
|
| 19 |
+
from PIL import ImageDraw
|
| 20 |
+
from transformers import PreTrainedTokenizer, AddedToken
|
| 21 |
+
from transformers.utils import try_to_load_from_cache
|
| 22 |
+
|
| 23 |
+
import matplotlib.colors as mcolors
|
| 24 |
+
from matplotlib.font_manager import FontProperties
|
| 25 |
+
|
| 26 |
+
logger = logging.getLogger(__name__)
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
VOCAB_FILES_NAMES = {"vocab_file": "qwen.tiktoken", "ttf": "SimSun.ttf"}
|
| 30 |
+
# pdb.set_trace()
|
| 31 |
+
FONT_PATH = try_to_load_from_cache("Qwen/Qwen-VL-Chat", "SimSun.ttf")
|
| 32 |
+
if FONT_PATH is None:
|
| 33 |
+
# if not os.path.exists("SimSun.ttf"):
|
| 34 |
+
# ttf = requests.get("https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/SimSun.ttf")
|
| 35 |
+
# open("SimSun.ttf", "wb").write(ttf.content)
|
| 36 |
+
FONT_PATH = "SimSun.ttf"
|
| 37 |
+
|
| 38 |
+
PAT_STR = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""
|
| 39 |
+
ENDOFTEXT = "<|endoftext|>"
|
| 40 |
+
IMSTART = "<|im_start|>"
|
| 41 |
+
IMEND = "<|im_end|>"
|
| 42 |
+
# as the default behavior is changed to allow special tokens in
|
| 43 |
+
# regular texts, the surface forms of special tokens need to be
|
| 44 |
+
# as different as possible to minimize the impact
|
| 45 |
+
EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
|
| 46 |
+
SPECIAL_TOKENS = (
|
| 47 |
+
ENDOFTEXT,
|
| 48 |
+
IMSTART,
|
| 49 |
+
IMEND,
|
| 50 |
+
) + EXTRAS
|
| 51 |
+
IMG_TOKEN_SPAN = 512
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
|
| 55 |
+
with open(tiktoken_bpe_file, "rb") as f:
|
| 56 |
+
contents = f.read()
|
| 57 |
+
return {
|
| 58 |
+
base64.b64decode(token): int(rank)
|
| 59 |
+
for token, rank in (line.split() for line in contents.splitlines() if line)
|
| 60 |
+
}
|
| 61 |
+
|
| 62 |
+
def _list_find(
|
| 63 |
+
input_list: List[Any],
|
| 64 |
+
candidates: Tuple[Any],
|
| 65 |
+
start: int = 0,
|
| 66 |
+
):
|
| 67 |
+
for i in range(start, len(input_list)):
|
| 68 |
+
if input_list[i] in candidates:
|
| 69 |
+
return i
|
| 70 |
+
return -1
|
| 71 |
+
|
| 72 |
+
def _replace_closed_tag(
|
| 73 |
+
input_tokens: List[Any],
|
| 74 |
+
start_tags: Union[Any, Tuple[Any]],
|
| 75 |
+
end_tags: Union[Any, Tuple[Any]],
|
| 76 |
+
inclusive_replace_func: Callable,
|
| 77 |
+
exclusive_replace_func: Callable = lambda x: x,
|
| 78 |
+
):
|
| 79 |
+
if isinstance(start_tags, (str, int)):
|
| 80 |
+
start_tags = (start_tags,)
|
| 81 |
+
if isinstance(end_tags, (str, int)):
|
| 82 |
+
end_tags = (end_tags,)
|
| 83 |
+
assert len(start_tags) == len(end_tags)
|
| 84 |
+
|
| 85 |
+
output_tokens = []
|
| 86 |
+
end = 0
|
| 87 |
+
while True:
|
| 88 |
+
start = _list_find(input_tokens, start_tags, end)
|
| 89 |
+
if start == -1:
|
| 90 |
+
break
|
| 91 |
+
output_tokens.extend(exclusive_replace_func(input_tokens[end : start]))
|
| 92 |
+
tag_idx = start_tags.index(input_tokens[start])
|
| 93 |
+
end = _list_find(input_tokens, (end_tags[tag_idx],), start)
|
| 94 |
+
if end == -1:
|
| 95 |
+
raise ValueError("Unclosed image token")
|
| 96 |
+
output_tokens.extend(inclusive_replace_func(input_tokens[start : end + 1]))
|
| 97 |
+
end += 1
|
| 98 |
+
output_tokens.extend(exclusive_replace_func(input_tokens[end : ]))
|
| 99 |
+
return output_tokens
|
| 100 |
+
|
| 101 |
+
class QWenTokenizer(PreTrainedTokenizer):
|
| 102 |
+
"""QWen tokenizer."""
|
| 103 |
+
|
| 104 |
+
vocab_files_names = VOCAB_FILES_NAMES
|
| 105 |
+
|
| 106 |
+
def __init__(
|
| 107 |
+
self,
|
| 108 |
+
vocab_file,
|
| 109 |
+
errors="replace",
|
| 110 |
+
image_start_tag='<img>',
|
| 111 |
+
image_end_tag='</img>',
|
| 112 |
+
image_pad_tag='<imgpad>',
|
| 113 |
+
ref_start_tag='<ref>',
|
| 114 |
+
ref_end_tag='</ref>',
|
| 115 |
+
box_start_tag='<box>',
|
| 116 |
+
box_end_tag='</box>',
|
| 117 |
+
quad_start_tag='<quad>',
|
| 118 |
+
quad_end_tag='</quad>',
|
| 119 |
+
**kwargs,
|
| 120 |
+
):
|
| 121 |
+
super().__init__(**kwargs)
|
| 122 |
+
self.image_start_tag = image_start_tag
|
| 123 |
+
self.image_end_tag = image_end_tag
|
| 124 |
+
self.image_pad_tag = image_pad_tag
|
| 125 |
+
self.ref_start_tag = ref_start_tag
|
| 126 |
+
self.ref_end_tag = ref_end_tag
|
| 127 |
+
self.box_start_tag = box_start_tag
|
| 128 |
+
self.box_end_tag = box_end_tag
|
| 129 |
+
self.quad_start_tag = quad_start_tag
|
| 130 |
+
self.quad_end_tag = quad_end_tag
|
| 131 |
+
self.IMAGE_ST = (
|
| 132 |
+
ref_start_tag, ref_end_tag,
|
| 133 |
+
box_start_tag, box_end_tag,
|
| 134 |
+
quad_start_tag, quad_end_tag,
|
| 135 |
+
image_start_tag, image_end_tag,
|
| 136 |
+
image_pad_tag
|
| 137 |
+
)
|
| 138 |
+
|
| 139 |
+
self.errors = errors # how to handle errors in decoding
|
| 140 |
+
|
| 141 |
+
self.mergeable_ranks = _load_tiktoken_bpe(vocab_file) # type: dict[bytes, int]
|
| 142 |
+
self.special_tokens = {
|
| 143 |
+
token: index
|
| 144 |
+
for index, token in enumerate(
|
| 145 |
+
SPECIAL_TOKENS + self.IMAGE_ST, start=len(self.mergeable_ranks)
|
| 146 |
+
)
|
| 147 |
+
}
|
| 148 |
+
self.img_start_id = self.special_tokens[self.image_start_tag]
|
| 149 |
+
self.img_end_id = self.special_tokens[self.image_end_tag]
|
| 150 |
+
self.img_pad_id = self.special_tokens[self.image_pad_tag]
|
| 151 |
+
self.ref_start_id = self.special_tokens[self.ref_start_tag]
|
| 152 |
+
self.ref_end_id = self.special_tokens[self.ref_end_tag]
|
| 153 |
+
self.box_start_id = self.special_tokens[self.box_start_tag]
|
| 154 |
+
self.box_end_id = self.special_tokens[self.box_end_tag]
|
| 155 |
+
self.quad_start_id = self.special_tokens[self.quad_start_tag]
|
| 156 |
+
self.quad_end_id = self.special_tokens[self.quad_end_tag]
|
| 157 |
+
|
| 158 |
+
enc = tiktoken.Encoding(
|
| 159 |
+
"Qwen",
|
| 160 |
+
pat_str=PAT_STR,
|
| 161 |
+
mergeable_ranks=self.mergeable_ranks,
|
| 162 |
+
special_tokens=self.special_tokens,
|
| 163 |
+
)
|
| 164 |
+
assert (
|
| 165 |
+
len(self.mergeable_ranks) + len(self.special_tokens) == enc.n_vocab
|
| 166 |
+
), f"{len(self.mergeable_ranks) + len(self.special_tokens)} != {enc.n_vocab} in encoding"
|
| 167 |
+
|
| 168 |
+
self.decoder = {
|
| 169 |
+
v: k for k, v in self.mergeable_ranks.items()
|
| 170 |
+
} # type: dict[int, bytes|str]
|
| 171 |
+
self.decoder.update({v: k for k, v in self.special_tokens.items()})
|
| 172 |
+
|
| 173 |
+
self.tokenizer = enc # type: tiktoken.Encoding
|
| 174 |
+
|
| 175 |
+
self.eod_id = self.tokenizer.eot_token
|
| 176 |
+
self.im_start_id = self.special_tokens[IMSTART]
|
| 177 |
+
self.im_end_id = self.special_tokens[IMEND]
|
| 178 |
+
|
| 179 |
+
def __getstate__(self):
|
| 180 |
+
# for pickle lovers
|
| 181 |
+
state = self.__dict__.copy()
|
| 182 |
+
del state['tokenizer']
|
| 183 |
+
return state
|
| 184 |
+
|
| 185 |
+
def __setstate__(self, state):
|
| 186 |
+
# tokenizer is not python native; don't pass it; rebuild it
|
| 187 |
+
self.__dict__.update(state)
|
| 188 |
+
enc = tiktoken.Encoding(
|
| 189 |
+
"Qwen",
|
| 190 |
+
pat_str=PAT_STR,
|
| 191 |
+
mergeable_ranks=self.mergeable_ranks,
|
| 192 |
+
special_tokens=self.special_tokens,
|
| 193 |
+
)
|
| 194 |
+
self.tokenizer = enc
|
| 195 |
+
|
| 196 |
+
|
| 197 |
+
def __len__(self) -> int:
|
| 198 |
+
return self.tokenizer.n_vocab
|
| 199 |
+
|
| 200 |
+
def get_vocab(self) -> Dict[bytes, int]:
|
| 201 |
+
return self.mergeable_ranks
|
| 202 |
+
|
| 203 |
+
def convert_tokens_to_ids(
|
| 204 |
+
self, tokens: Union[bytes, str, List[Union[bytes, str]]]
|
| 205 |
+
) -> List[int]:
|
| 206 |
+
ids = []
|
| 207 |
+
if isinstance(tokens, (str, bytes)):
|
| 208 |
+
if tokens in self.special_tokens:
|
| 209 |
+
return self.special_tokens[tokens]
|
| 210 |
+
else:
|
| 211 |
+
return self.mergeable_ranks.get(tokens)
|
| 212 |
+
for token in tokens:
|
| 213 |
+
if token in self.special_tokens:
|
| 214 |
+
ids.append(self.special_tokens[token])
|
| 215 |
+
else:
|
| 216 |
+
ids.append(self.mergeable_ranks.get(token))
|
| 217 |
+
return ids
|
| 218 |
+
|
| 219 |
+
def _add_tokens(self, new_tokens: Union[List[str], List[AddedToken]], special_tokens: bool = False) -> int:
|
| 220 |
+
if not special_tokens and new_tokens:
|
| 221 |
+
raise ValueError('Adding regular tokens is not supported')
|
| 222 |
+
for token in new_tokens:
|
| 223 |
+
surface_form = token.content if isinstance(token, AddedToken) else token
|
| 224 |
+
if surface_form not in SPECIAL_TOKENS + self.IMAGE_ST:
|
| 225 |
+
raise ValueError('Adding unknown special tokens is not supported')
|
| 226 |
+
return 0
|
| 227 |
+
|
| 228 |
+
def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
|
| 229 |
+
"""
|
| 230 |
+
Save only the vocabulary of the tokenizer (vocabulary).
|
| 231 |
+
|
| 232 |
+
Returns:
|
| 233 |
+
`Tuple(str)`: Paths to the files saved.
|
| 234 |
+
"""
|
| 235 |
+
file_path = os.path.join(save_directory, "qwen.tiktoken")
|
| 236 |
+
with open(file_path, "w", encoding="utf8") as w:
|
| 237 |
+
for k, v in self.mergeable_ranks.items():
|
| 238 |
+
line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
|
| 239 |
+
w.write(line)
|
| 240 |
+
return (file_path,)
|
| 241 |
+
|
| 242 |
+
def tokenize(
|
| 243 |
+
self,
|
| 244 |
+
text: str,
|
| 245 |
+
allowed_special: Union[Set, str] = "all",
|
| 246 |
+
disallowed_special: Union[Collection, str] = (),
|
| 247 |
+
**kwargs,
|
| 248 |
+
) -> List[Union[bytes, str]]:
|
| 249 |
+
"""
|
| 250 |
+
Converts a string in a sequence of tokens.
|
| 251 |
+
|
| 252 |
+
Args:
|
| 253 |
+
text (`str`):
|
| 254 |
+
The sequence to be encoded.
|
| 255 |
+
allowed_special (`Literal["all"]` or `set`):
|
| 256 |
+
The surface forms of the tokens to be encoded as special tokens in regular texts.
|
| 257 |
+
Default to "all".
|
| 258 |
+
disallowed_special (`Literal["all"]` or `Collection`):
|
| 259 |
+
The surface forms of the tokens that should not be in regular texts and trigger errors.
|
| 260 |
+
Default to an empty tuple.
|
| 261 |
+
|
| 262 |
+
kwargs (additional keyword arguments, *optional*):
|
| 263 |
+
Will be passed to the underlying model specific encode method.
|
| 264 |
+
|
| 265 |
+
Returns:
|
| 266 |
+
`List[bytes|str]`: The list of tokens.
|
| 267 |
+
"""
|
| 268 |
+
tokens = []
|
| 269 |
+
text = unicodedata.normalize("NFC", text)
|
| 270 |
+
|
| 271 |
+
# this implementation takes a detour: text -> token id -> token surface forms
|
| 272 |
+
for t in self.tokenizer.encode(
|
| 273 |
+
text, allowed_special=allowed_special, disallowed_special=disallowed_special
|
| 274 |
+
):
|
| 275 |
+
tokens.append(self.decoder[t])
|
| 276 |
+
|
| 277 |
+
def _encode_imgurl(img_tokens):
|
| 278 |
+
assert img_tokens[0] == self.image_start_tag and img_tokens[-1] == self.image_end_tag
|
| 279 |
+
img_tokens = img_tokens[1:-1]
|
| 280 |
+
img_url = b''.join(img_tokens)
|
| 281 |
+
out_img_tokens = list(map(self.decoder.get, img_url))
|
| 282 |
+
if len(out_img_tokens) > IMG_TOKEN_SPAN:
|
| 283 |
+
raise ValueError("The content in {}..{} is too long".format(
|
| 284 |
+
self.image_start_tag, self.image_end_tag))
|
| 285 |
+
out_img_tokens.extend([self.image_pad_tag] * (IMG_TOKEN_SPAN - len(out_img_tokens)))
|
| 286 |
+
out_img_tokens = [self.image_start_tag] + out_img_tokens + [self.image_end_tag]
|
| 287 |
+
return out_img_tokens
|
| 288 |
+
|
| 289 |
+
return _replace_closed_tag(tokens, self.image_start_tag, self.image_end_tag, _encode_imgurl)
|
| 290 |
+
|
| 291 |
+
def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
|
| 292 |
+
"""
|
| 293 |
+
Converts a sequence of tokens in a single string.
|
| 294 |
+
"""
|
| 295 |
+
text = ""
|
| 296 |
+
temp = b""
|
| 297 |
+
for t in tokens:
|
| 298 |
+
if isinstance(t, str):
|
| 299 |
+
if temp:
|
| 300 |
+
text += temp.decode("utf-8", errors=self.errors)
|
| 301 |
+
temp = b""
|
| 302 |
+
text += t
|
| 303 |
+
elif isinstance(t, bytes):
|
| 304 |
+
temp += t
|
| 305 |
+
else:
|
| 306 |
+
raise TypeError("token should only be of type types or str")
|
| 307 |
+
if temp:
|
| 308 |
+
text += temp.decode("utf-8", errors=self.errors)
|
| 309 |
+
return text
|
| 310 |
+
|
| 311 |
+
@property
|
| 312 |
+
def vocab_size(self):
|
| 313 |
+
return self.tokenizer.n_vocab
|
| 314 |
+
|
| 315 |
+
def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
|
| 316 |
+
"""Converts an id to a token, special tokens included"""
|
| 317 |
+
if index in self.decoder:
|
| 318 |
+
return self.decoder[index]
|
| 319 |
+
raise ValueError("unknown ids")
|
| 320 |
+
|
| 321 |
+
def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
|
| 322 |
+
"""Converts a token to an id using the vocab, special tokens included"""
|
| 323 |
+
if token in self.special_tokens:
|
| 324 |
+
return self.special_tokens[token]
|
| 325 |
+
if token in self.mergeable_ranks:
|
| 326 |
+
return self.mergeable_ranks[token]
|
| 327 |
+
raise ValueError("unknown token")
|
| 328 |
+
|
| 329 |
+
def _tokenize(self, text: str, **kwargs):
|
| 330 |
+
"""
|
| 331 |
+
Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
|
| 332 |
+
vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
|
| 333 |
+
|
| 334 |
+
Do NOT take care of added tokens.
|
| 335 |
+
"""
|
| 336 |
+
raise NotImplementedError
|
| 337 |
+
|
| 338 |
+
def _decode(
|
| 339 |
+
self,
|
| 340 |
+
token_ids: Union[int, List[int]],
|
| 341 |
+
skip_special_tokens: bool = False,
|
| 342 |
+
errors: str = None,
|
| 343 |
+
**kwargs,
|
| 344 |
+
) -> str:
|
| 345 |
+
if isinstance(token_ids, int):
|
| 346 |
+
token_ids = [token_ids]
|
| 347 |
+
|
| 348 |
+
def _decode_imgurl(img_token_ids):
|
| 349 |
+
assert img_token_ids[0] == self.img_start_id and img_token_ids[-1] == self.img_end_id
|
| 350 |
+
img_token_ids = img_token_ids[1:-1]
|
| 351 |
+
img_token_ids = img_token_ids[ : img_token_ids.index(self.img_pad_id)]
|
| 352 |
+
img_url = bytes(img_token_ids).decode('utf-8')
|
| 353 |
+
return [self.img_start_id] + self.tokenizer.encode(img_url) + [self.img_end_id]
|
| 354 |
+
|
| 355 |
+
token_ids = _replace_closed_tag(token_ids, self.img_start_id, self.img_end_id, _decode_imgurl)
|
| 356 |
+
|
| 357 |
+
if skip_special_tokens:
|
| 358 |
+
token_ids = [i for i in token_ids if i < self.eod_id]
|
| 359 |
+
return self.tokenizer.decode(token_ids, errors=errors or self.errors)
|
| 360 |
+
|
| 361 |
+
def to_list_format(self, text: str):
|
| 362 |
+
# pdb.set_trace()
|
| 363 |
+
text = unicodedata.normalize("NFC", text)
|
| 364 |
+
token_ids = self.tokenizer.encode(
|
| 365 |
+
text, allowed_special=set(self.IMAGE_ST + (ENDOFTEXT,)))
|
| 366 |
+
|
| 367 |
+
def _encode_vl_info(tokens):
|
| 368 |
+
if len(tokens) == 0:
|
| 369 |
+
return []
|
| 370 |
+
if tokens[0] == self.img_start_id and tokens[-1] == self.img_end_id:
|
| 371 |
+
key = 'image'
|
| 372 |
+
elif tokens[0] == self.ref_start_id and tokens[-1] == self.ref_end_id:
|
| 373 |
+
key = 'ref'
|
| 374 |
+
elif tokens[0] == self.box_start_id and tokens[-1] == self.box_end_id:
|
| 375 |
+
key = 'box'
|
| 376 |
+
elif tokens[0] == self.quad_start_id and tokens[-1] == self.quad_end_id:
|
| 377 |
+
key = 'quad'
|
| 378 |
+
else:
|
| 379 |
+
_tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
|
| 380 |
+
return [{'text': b''.join(map(_tobytes, map(self.decoder.get, tokens))).decode('utf-8')}]
|
| 381 |
+
_tobytes = lambda x: x.encode('utf-8') if isinstance(x, str) else x
|
| 382 |
+
val = b''.join(map(_tobytes, map(self.decoder.get, tokens[1:-1]))).decode('utf-8')
|
| 383 |
+
return [{key: val}]
|
| 384 |
+
|
| 385 |
+
return _replace_closed_tag(
|
| 386 |
+
token_ids,
|
| 387 |
+
(self.img_start_id, self.ref_start_id, self.box_start_id, self.quad_start_id),
|
| 388 |
+
(self.img_end_id, self.ref_end_id, self.box_end_id, self.quad_end_id),
|
| 389 |
+
_encode_vl_info,
|
| 390 |
+
_encode_vl_info,
|
| 391 |
+
)
|
| 392 |
+
|
| 393 |
+
def from_list_format(self, list_format: List[Dict]):
|
| 394 |
+
# pdb.set_trace()
|
| 395 |
+
text = ''
|
| 396 |
+
num_images = 0
|
| 397 |
+
for ele in list_format:
|
| 398 |
+
if 'image' in ele:
|
| 399 |
+
num_images += 1
|
| 400 |
+
text += f'Picture {num_images}:'
|
| 401 |
+
text += self.image_start_tag + ele['image'] + self.image_end_tag
|
| 402 |
+
text += '\n'
|
| 403 |
+
elif 'text' in ele:
|
| 404 |
+
text += ele['text']
|
| 405 |
+
elif 'box' in ele:
|
| 406 |
+
if 'ref' in ele:
|
| 407 |
+
text += self.ref_start_tag + ele['ref'] + self.ref_end_tag
|
| 408 |
+
for box in ele['box']:
|
| 409 |
+
text += self.box_start_tag + '(%d,%d),(%d,%d)' % (box[0], box[1], box[2], box[3]) + self.box_end_tag
|
| 410 |
+
else:
|
| 411 |
+
raise ValueError("Unsupport element: " + str(ele))
|
| 412 |
+
return text
|
| 413 |
+
|
| 414 |
+
def _fetch_latest_picture(self, response, history):
|
| 415 |
+
if history is None:
|
| 416 |
+
history = []
|
| 417 |
+
_history = history + [(response, None)]
|
| 418 |
+
for q, r in _history[::-1]:
|
| 419 |
+
for ele in self.to_list_format(q)[::-1]:
|
| 420 |
+
if 'image' in ele:
|
| 421 |
+
return ele['image']
|
| 422 |
+
return None
|
| 423 |
+
|
| 424 |
+
def _fetch_all_box_with_ref(self, text):
|
| 425 |
+
list_format = self.to_list_format(text)
|
| 426 |
+
output = []
|
| 427 |
+
for i, ele in enumerate(list_format):
|
| 428 |
+
if 'box' in ele:
|
| 429 |
+
bbox = tuple(map(int, ele['box'].replace('(', '').replace(')', '').split(',')))
|
| 430 |
+
assert len(bbox) == 4
|
| 431 |
+
output.append({'box': bbox})
|
| 432 |
+
if i > 0 and 'ref' in list_format[i-1]:
|
| 433 |
+
output[-1]['ref'] = list_format[i-1]['ref'].strip()
|
| 434 |
+
return output
|
| 435 |
+
|
| 436 |
+
def draw_bbox_on_latest_picture(
|
| 437 |
+
self,
|
| 438 |
+
response,
|
| 439 |
+
history=None,
|
| 440 |
+
) -> Optional[Image.Image]:
|
| 441 |
+
image = self._fetch_latest_picture(response, history)
|
| 442 |
+
if image is None:
|
| 443 |
+
return None
|
| 444 |
+
if image.startswith("http://") or image.startswith("https://"):
|
| 445 |
+
image = Image.open(requests.get(image, stream=True).raw).convert("RGB")
|
| 446 |
+
h, w = image.height, image.width
|
| 447 |
+
else:
|
| 448 |
+
image = np.asarray(Image.open(image).convert("RGB"))
|
| 449 |
+
h, w = image.shape[0], image.shape[1]
|
| 450 |
+
visualizer = Visualizer(image)
|
| 451 |
+
|
| 452 |
+
boxes = self._fetch_all_box_with_ref(response)
|
| 453 |
+
if not boxes:
|
| 454 |
+
return None
|
| 455 |
+
color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()]) # init color
|
| 456 |
+
for box in boxes:
|
| 457 |
+
if 'ref' in box: # random new color for new refexps
|
| 458 |
+
color = random.choice([_ for _ in mcolors.TABLEAU_COLORS.keys()])
|
| 459 |
+
x1, y1, x2, y2 = box['box']
|
| 460 |
+
x1, y1, x2, y2 = (int(x1 / 1000 * w), int(y1 / 1000 * h), int(x2 / 1000 * w), int(y2 / 1000 * h))
|
| 461 |
+
visualizer.draw_box((x1, y1, x2, y2), alpha=1, edge_color=color)
|
| 462 |
+
if 'ref' in box:
|
| 463 |
+
visualizer.draw_text(box['ref'], (x1, y1), color=color, horizontal_alignment="left")
|
| 464 |
+
return visualizer.output
|
| 465 |
+
|
| 466 |
+
|
| 467 |
+
import colorsys
|
| 468 |
+
import logging
|
| 469 |
+
import math
|
| 470 |
+
import numpy as np
|
| 471 |
+
import matplotlib as mpl
|
| 472 |
+
import matplotlib.colors as mplc
|
| 473 |
+
import matplotlib.figure as mplfigure
|
| 474 |
+
import torch
|
| 475 |
+
from matplotlib.backends.backend_agg import FigureCanvasAgg
|
| 476 |
+
from PIL import Image
|
| 477 |
+
import random
|
| 478 |
+
|
| 479 |
+
logger = logging.getLogger(__name__)
|
| 480 |
+
|
| 481 |
+
|
| 482 |
+
class VisImage:
|
| 483 |
+
def __init__(self, img, scale=1.0):
|
| 484 |
+
self.img = img
|
| 485 |
+
self.scale = scale
|
| 486 |
+
self.width, self.height = img.shape[1], img.shape[0]
|
| 487 |
+
self._setup_figure(img)
|
| 488 |
+
|
| 489 |
+
def _setup_figure(self, img):
|
| 490 |
+
fig = mplfigure.Figure(frameon=False)
|
| 491 |
+
self.dpi = fig.get_dpi()
|
| 492 |
+
# add a small 1e-2 to avoid precision lost due to matplotlib's truncation
|
| 493 |
+
# (https://github.com/matplotlib/matplotlib/issues/15363)
|
| 494 |
+
fig.set_size_inches(
|
| 495 |
+
(self.width * self.scale + 1e-2) / self.dpi,
|
| 496 |
+
(self.height * self.scale + 1e-2) / self.dpi,
|
| 497 |
+
)
|
| 498 |
+
self.canvas = FigureCanvasAgg(fig)
|
| 499 |
+
# self.canvas = mpl.backends.backend_cairo.FigureCanvasCairo(fig)
|
| 500 |
+
ax = fig.add_axes([0.0, 0.0, 1.0, 1.0])
|
| 501 |
+
ax.axis("off")
|
| 502 |
+
self.fig = fig
|
| 503 |
+
self.ax = ax
|
| 504 |
+
self.reset_image(img)
|
| 505 |
+
|
| 506 |
+
def reset_image(self, img):
|
| 507 |
+
img = img.astype("uint8")
|
| 508 |
+
self.ax.imshow(img, extent=(0, self.width, self.height, 0), interpolation="nearest")
|
| 509 |
+
|
| 510 |
+
def save(self, filepath):
|
| 511 |
+
self.fig.savefig(filepath)
|
| 512 |
+
|
| 513 |
+
def get_image(self):
|
| 514 |
+
canvas = self.canvas
|
| 515 |
+
s, (width, height) = canvas.print_to_buffer()
|
| 516 |
+
|
| 517 |
+
buffer = np.frombuffer(s, dtype="uint8")
|
| 518 |
+
|
| 519 |
+
img_rgba = buffer.reshape(height, width, 4)
|
| 520 |
+
rgb, alpha = np.split(img_rgba, [3], axis=2)
|
| 521 |
+
return rgb.astype("uint8")
|
| 522 |
+
|
| 523 |
+
|
| 524 |
+
class Visualizer:
|
| 525 |
+
def __init__(self, img_rgb, metadata=None, scale=1.0):
|
| 526 |
+
self.img = np.asarray(img_rgb).clip(0, 255).astype(np.uint8)
|
| 527 |
+
self.font_path = FONT_PATH
|
| 528 |
+
self.output = VisImage(self.img, scale=scale)
|
| 529 |
+
self.cpu_device = torch.device("cpu")
|
| 530 |
+
|
| 531 |
+
# too small texts are useless, therefore clamp to 14
|
| 532 |
+
self._default_font_size = max(
|
| 533 |
+
np.sqrt(self.output.height * self.output.width) // 30, 15 // scale
|
| 534 |
+
)
|
| 535 |
+
|
| 536 |
+
def draw_text(
|
| 537 |
+
self,
|
| 538 |
+
text,
|
| 539 |
+
position,
|
| 540 |
+
*,
|
| 541 |
+
font_size=None,
|
| 542 |
+
color="g",
|
| 543 |
+
horizontal_alignment="center",
|
| 544 |
+
rotation=0,
|
| 545 |
+
):
|
| 546 |
+
if not font_size:
|
| 547 |
+
font_size = self._default_font_size
|
| 548 |
+
|
| 549 |
+
# since the text background is dark, we don't want the text to be dark
|
| 550 |
+
color = np.maximum(list(mplc.to_rgb(color)), 0.2)
|
| 551 |
+
color[np.argmax(color)] = max(0.8, np.max(color))
|
| 552 |
+
|
| 553 |
+
x, y = position
|
| 554 |
+
self.output.ax.text(
|
| 555 |
+
x,
|
| 556 |
+
y,
|
| 557 |
+
text,
|
| 558 |
+
size=font_size * self.output.scale,
|
| 559 |
+
fontproperties=FontProperties(fname=self.font_path),
|
| 560 |
+
bbox={"facecolor": "black", "alpha": 0.8, "pad": 0.7, "edgecolor": "none"},
|
| 561 |
+
verticalalignment="top",
|
| 562 |
+
horizontalalignment=horizontal_alignment,
|
| 563 |
+
color=color,
|
| 564 |
+
zorder=10,
|
| 565 |
+
rotation=rotation,
|
| 566 |
+
)
|
| 567 |
+
return self.output
|
| 568 |
+
|
| 569 |
+
def draw_box(self, box_coord, alpha=0.5, edge_color="g", line_style="-"):
|
| 570 |
+
|
| 571 |
+
x0, y0, x1, y1 = box_coord
|
| 572 |
+
width = x1 - x0
|
| 573 |
+
height = y1 - y0
|
| 574 |
+
|
| 575 |
+
linewidth = max(self._default_font_size / 4, 1)
|
| 576 |
+
|
| 577 |
+
self.output.ax.add_patch(
|
| 578 |
+
mpl.patches.Rectangle(
|
| 579 |
+
(x0, y0),
|
| 580 |
+
width,
|
| 581 |
+
height,
|
| 582 |
+
fill=False,
|
| 583 |
+
edgecolor=edge_color,
|
| 584 |
+
linewidth=linewidth * self.output.scale,
|
| 585 |
+
alpha=alpha,
|
| 586 |
+
linestyle=line_style,
|
| 587 |
+
)
|
| 588 |
+
)
|
| 589 |
+
return self.output
|
| 590 |
+
|
| 591 |
+
def get_output(self):
|
| 592 |
+
|
| 593 |
+
return self.output
|
visual.py
ADDED
|
@@ -0,0 +1,482 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Copyright (c) Alibaba Cloud.
|
| 2 |
+
#
|
| 3 |
+
# This source code is licensed under the license found in the
|
| 4 |
+
# LICENSE file in the root directory of this source tree.
|
| 5 |
+
|
| 6 |
+
from collections import OrderedDict
|
| 7 |
+
import math
|
| 8 |
+
import requests
|
| 9 |
+
from io import BytesIO
|
| 10 |
+
from functools import partial
|
| 11 |
+
from PIL import Image
|
| 12 |
+
from typing import Callable, Optional, Sequence, Tuple, List
|
| 13 |
+
import numpy as np
|
| 14 |
+
import torch
|
| 15 |
+
from torch import nn
|
| 16 |
+
from torch.nn import functional as F
|
| 17 |
+
from torch.nn.init import trunc_normal_
|
| 18 |
+
from torchvision import transforms
|
| 19 |
+
from torchvision.transforms import InterpolationMode
|
| 20 |
+
import pdb
|
| 21 |
+
|
| 22 |
+
def sliding_window(matrix, window_size, stride):
|
| 23 |
+
b,c,height, width = matrix.shape
|
| 24 |
+
window_rows = (height - window_size[0]) // stride + 1
|
| 25 |
+
window_cols = (width - window_size[1]) // stride + 1
|
| 26 |
+
images_448 = F.interpolate(matrix, size=window_size, mode='bicubic')
|
| 27 |
+
windows = []
|
| 28 |
+
# pdb.set_trace()
|
| 29 |
+
for i in range(window_rows):
|
| 30 |
+
windows_col = []
|
| 31 |
+
for j in range(window_cols):
|
| 32 |
+
window = matrix[:,:, i*stride:i*stride+window_size[0], j*stride:j*stride+window_size[1]]
|
| 33 |
+
windows.append(window)
|
| 34 |
+
# windows.append(windows_col)
|
| 35 |
+
windows.append(images_448)
|
| 36 |
+
images = torch.cat(windows,dim=1)
|
| 37 |
+
images = images.reshape(b*5,c,window_size[0], window_size[0])
|
| 38 |
+
|
| 39 |
+
return images
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
def get_abs_pos(abs_pos, tgt_size):
|
| 43 |
+
# abs_pos: L, C
|
| 44 |
+
# tgt_size: M
|
| 45 |
+
# return: M, C
|
| 46 |
+
src_size = int(math.sqrt(abs_pos.size(0)))
|
| 47 |
+
tgt_size = int(math.sqrt(tgt_size))
|
| 48 |
+
dtype = abs_pos.dtype
|
| 49 |
+
|
| 50 |
+
if src_size != tgt_size:
|
| 51 |
+
return F.interpolate(
|
| 52 |
+
abs_pos.float().reshape(1, src_size, src_size, -1).permute(0, 3, 1, 2),
|
| 53 |
+
size=(tgt_size, tgt_size),
|
| 54 |
+
mode="bicubic",
|
| 55 |
+
align_corners=False,
|
| 56 |
+
).permute(0, 2, 3, 1).flatten(0, 2).to(dtype=dtype)
|
| 57 |
+
else:
|
| 58 |
+
return abs_pos
|
| 59 |
+
|
| 60 |
+
# https://github.com/facebookresearch/mae/blob/efb2a8062c206524e35e47d04501ed4f544c0ae8/util/pos_embed.py#L20
|
| 61 |
+
def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
|
| 62 |
+
"""
|
| 63 |
+
grid_size: int of the grid height and width
|
| 64 |
+
return:
|
| 65 |
+
pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
|
| 66 |
+
"""
|
| 67 |
+
grid_h = np.arange(grid_size, dtype=np.float32)
|
| 68 |
+
grid_w = np.arange(grid_size, dtype=np.float32)
|
| 69 |
+
grid = np.meshgrid(grid_w, grid_h) # here w goes first
|
| 70 |
+
grid = np.stack(grid, axis=0)
|
| 71 |
+
|
| 72 |
+
grid = grid.reshape([2, 1, grid_size, grid_size])
|
| 73 |
+
pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
|
| 74 |
+
if cls_token:
|
| 75 |
+
pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
|
| 76 |
+
return pos_embed
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
|
| 80 |
+
assert embed_dim % 2 == 0
|
| 81 |
+
|
| 82 |
+
# use half of dimensions to encode grid_h
|
| 83 |
+
emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0]) # (H*W, D/2)
|
| 84 |
+
emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1]) # (H*W, D/2)
|
| 85 |
+
|
| 86 |
+
emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
|
| 87 |
+
return emb
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
|
| 91 |
+
"""
|
| 92 |
+
embed_dim: output dimension for each position
|
| 93 |
+
pos: a list of positions to be encoded: size (M,)
|
| 94 |
+
out: (M, D)
|
| 95 |
+
"""
|
| 96 |
+
assert embed_dim % 2 == 0
|
| 97 |
+
omega = np.arange(embed_dim // 2, dtype=np.float32)
|
| 98 |
+
omega /= embed_dim / 2.
|
| 99 |
+
omega = 1. / 10000**omega # (D/2,)
|
| 100 |
+
|
| 101 |
+
pos = pos.reshape(-1) # (M,)
|
| 102 |
+
out = np.einsum('m,d->md', pos, omega) # (M, D/2), outer product
|
| 103 |
+
|
| 104 |
+
emb_sin = np.sin(out) # (M, D/2)
|
| 105 |
+
emb_cos = np.cos(out) # (M, D/2)
|
| 106 |
+
|
| 107 |
+
emb = np.concatenate([emb_sin, emb_cos], axis=1) # (M, D)
|
| 108 |
+
return emb
|
| 109 |
+
|
| 110 |
+
|
| 111 |
+
class Resampler(nn.Module):
|
| 112 |
+
"""
|
| 113 |
+
A 2D perceiver-resampler network with one cross attention layers by
|
| 114 |
+
(grid_size**2) learnable queries and 2d sincos pos_emb
|
| 115 |
+
Outputs:
|
| 116 |
+
A tensor with the shape of (grid_size**2, embed_dim)
|
| 117 |
+
"""
|
| 118 |
+
def __init__(
|
| 119 |
+
self,
|
| 120 |
+
grid_size,
|
| 121 |
+
embed_dim,
|
| 122 |
+
num_heads,
|
| 123 |
+
kv_dim=None,
|
| 124 |
+
norm_layer=nn.LayerNorm
|
| 125 |
+
):
|
| 126 |
+
super().__init__()
|
| 127 |
+
self.num_queries = grid_size ** 2
|
| 128 |
+
self.embed_dim = embed_dim
|
| 129 |
+
self.num_heads = num_heads
|
| 130 |
+
|
| 131 |
+
self.pos_embed = nn.Parameter(
|
| 132 |
+
torch.from_numpy(get_2d_sincos_pos_embed(embed_dim, grid_size)).float()
|
| 133 |
+
).requires_grad_(False)
|
| 134 |
+
|
| 135 |
+
self.query = nn.Parameter(torch.zeros(self.num_queries, embed_dim))
|
| 136 |
+
trunc_normal_(self.query, std=.02)
|
| 137 |
+
|
| 138 |
+
if kv_dim is not None and kv_dim != embed_dim:
|
| 139 |
+
self.kv_proj = nn.Linear(kv_dim, embed_dim, bias=False)
|
| 140 |
+
else:
|
| 141 |
+
self.kv_proj = nn.Identity()
|
| 142 |
+
|
| 143 |
+
self.attn = nn.MultiheadAttention(embed_dim, num_heads)
|
| 144 |
+
self.ln_q = norm_layer(embed_dim)
|
| 145 |
+
self.ln_kv = norm_layer(embed_dim)
|
| 146 |
+
|
| 147 |
+
self.apply(self._init_weights)
|
| 148 |
+
# pdb.set_trace()
|
| 149 |
+
#self.load_state_dict(torch.load('/cfs/cfs-lugcocyb/mingdali/code/qWen-VL/vl-chat/attn_params.pth'))
|
| 150 |
+
|
| 151 |
+
def _init_weights(self, m):
|
| 152 |
+
# self.load_state_dict(torch.load('/cfs/cfs-lugcocyb/mingdali/code/qWen-VL/vl-chat/attn_params.pth'))
|
| 153 |
+
#pdb.set_trace()
|
| 154 |
+
if isinstance(m, nn.Linear):
|
| 155 |
+
trunc_normal_(m.weight, std=.02)
|
| 156 |
+
if isinstance(m, nn.Linear) and m.bias is not None:
|
| 157 |
+
nn.init.constant_(m.bias, 0)
|
| 158 |
+
elif isinstance(m, nn.LayerNorm):
|
| 159 |
+
nn.init.constant_(m.bias, 0)
|
| 160 |
+
nn.init.constant_(m.weight, 1.0)
|
| 161 |
+
|
| 162 |
+
def forward(self, x, attn_mask=None):
|
| 163 |
+
#pdb.set_trace()
|
| 164 |
+
pos_embed = get_abs_pos(self.pos_embed, x.size(1))
|
| 165 |
+
|
| 166 |
+
x = self.kv_proj(x)
|
| 167 |
+
x = self.ln_kv(x).permute(1, 0, 2)
|
| 168 |
+
|
| 169 |
+
N = x.shape[1]
|
| 170 |
+
q = self.ln_q(self.query)
|
| 171 |
+
out = self.attn(
|
| 172 |
+
self._repeat(q, N) + self.pos_embed.unsqueeze(1),
|
| 173 |
+
x + pos_embed.unsqueeze(1),
|
| 174 |
+
x,
|
| 175 |
+
attn_mask=attn_mask)[0]
|
| 176 |
+
return out.permute(1, 0, 2)
|
| 177 |
+
|
| 178 |
+
def _repeat(self, query, N: int):
|
| 179 |
+
return query.unsqueeze(1).repeat(1, N, 1)
|
| 180 |
+
|
| 181 |
+
|
| 182 |
+
class VisualAttention(nn.Module):
|
| 183 |
+
"""self-attention layer class.
|
| 184 |
+
|
| 185 |
+
Self-attention layer takes input with size [s, b, h]
|
| 186 |
+
and returns output of the same size.
|
| 187 |
+
"""
|
| 188 |
+
|
| 189 |
+
def __init__(self, embed_dim, num_heads,
|
| 190 |
+
bias=True, kdim=None, vdim=None):
|
| 191 |
+
super(VisualAttention, self).__init__()
|
| 192 |
+
self.embed_dim = embed_dim
|
| 193 |
+
self.kdim = kdim if kdim is not None else embed_dim
|
| 194 |
+
self.vdim = vdim if vdim is not None else embed_dim
|
| 195 |
+
self._qkv_same_embed_dim = self.kdim == embed_dim and self.vdim == embed_dim
|
| 196 |
+
|
| 197 |
+
self.num_heads = num_heads
|
| 198 |
+
|
| 199 |
+
# Per attention head and per partition values.
|
| 200 |
+
assert embed_dim % num_heads == 0
|
| 201 |
+
self.hidden_size_per_attention_head = embed_dim // num_heads
|
| 202 |
+
self.num_attention_heads_per_partition = num_heads
|
| 203 |
+
self.hidden_size_per_partition = embed_dim
|
| 204 |
+
|
| 205 |
+
# Strided linear layer.
|
| 206 |
+
assert self._qkv_same_embed_dim, 'Only Support SelfAttention Currently'
|
| 207 |
+
self.in_proj = nn.Linear(embed_dim, 3 * embed_dim)
|
| 208 |
+
self.out_proj = nn.Linear(embed_dim, embed_dim)
|
| 209 |
+
self.norm_factor = math.sqrt(self.hidden_size_per_attention_head)
|
| 210 |
+
|
| 211 |
+
def forward(self, query, key, value, attn_mask = None):
|
| 212 |
+
# query/key/value: [sq, b, h]
|
| 213 |
+
sq, b, _ = query.size()
|
| 214 |
+
|
| 215 |
+
assert torch.allclose(query, key), 'Only Support Self-Attention Currently'
|
| 216 |
+
sk = sq
|
| 217 |
+
mixed_x_layer = self.in_proj(query)
|
| 218 |
+
|
| 219 |
+
# [sq, b, (np * 3 * hn)] --> [sq, b, np, 3 * hn]
|
| 220 |
+
new_tensor_shape = mixed_x_layer.size()[:-1] + \
|
| 221 |
+
(self.num_attention_heads_per_partition,
|
| 222 |
+
3 * self.hidden_size_per_attention_head)
|
| 223 |
+
mixed_x_layer = mixed_x_layer.view(*new_tensor_shape)
|
| 224 |
+
|
| 225 |
+
# [sq, b, np, 3 * hn] --> 3 [sq, b, np, hn]
|
| 226 |
+
query_layer, key_layer, value_layer = mixed_x_layer.split(
|
| 227 |
+
self.hidden_size_per_attention_head, dim=-1)
|
| 228 |
+
|
| 229 |
+
# [sq, b, np, hn] -> [sq, b * np, hn]
|
| 230 |
+
query_layer = query_layer.view(sq,
|
| 231 |
+
b * self.num_attention_heads_per_partition,
|
| 232 |
+
self.hidden_size_per_attention_head).transpose(0, 1)
|
| 233 |
+
# [sk, b, np, hn] -> [sk, b * np, hn]
|
| 234 |
+
key_layer = key_layer.view(sk,
|
| 235 |
+
b * self.num_attention_heads_per_partition,
|
| 236 |
+
self.hidden_size_per_attention_head).transpose(0, 1)
|
| 237 |
+
|
| 238 |
+
q_scaled = query_layer / self.norm_factor
|
| 239 |
+
if attn_mask is not None:
|
| 240 |
+
attention_probs = torch.baddbmm(attn_mask, q_scaled, key_layer.transpose(-2, -1))
|
| 241 |
+
else:
|
| 242 |
+
attention_probs = torch.bmm(q_scaled, key_layer.transpose(-2, -1))
|
| 243 |
+
attention_probs = attention_probs.softmax(dim=-1)
|
| 244 |
+
|
| 245 |
+
value_layer = value_layer.view(sk,
|
| 246 |
+
b * self.num_attention_heads_per_partition,
|
| 247 |
+
self.hidden_size_per_attention_head).transpose(0, 1)
|
| 248 |
+
|
| 249 |
+
# matmul: [b * np, sq, hn]
|
| 250 |
+
context_layer = torch.bmm(attention_probs, value_layer)
|
| 251 |
+
|
| 252 |
+
# change view [b, np, sq, hn]
|
| 253 |
+
context_layer = context_layer.view(b,
|
| 254 |
+
self.num_attention_heads_per_partition,
|
| 255 |
+
sq, self.hidden_size_per_attention_head)
|
| 256 |
+
|
| 257 |
+
# [b, np, sq, hn] --> [sq, b, np, hn]
|
| 258 |
+
context_layer = context_layer.permute(2, 0, 1, 3).contiguous()
|
| 259 |
+
|
| 260 |
+
# [sq, b, np, hn] --> [sq, b, hp]
|
| 261 |
+
new_context_layer_shape = context_layer.size()[:-2] + \
|
| 262 |
+
(self.hidden_size_per_partition,)
|
| 263 |
+
context_layer = context_layer.view(*new_context_layer_shape)
|
| 264 |
+
|
| 265 |
+
output = self.out_proj(context_layer)
|
| 266 |
+
|
| 267 |
+
return output
|
| 268 |
+
|
| 269 |
+
|
| 270 |
+
class VisualAttentionBlock(nn.Module):
|
| 271 |
+
def __init__(
|
| 272 |
+
self,
|
| 273 |
+
d_model: int,
|
| 274 |
+
n_head: int,
|
| 275 |
+
mlp_ratio: float = 4.0,
|
| 276 |
+
act_layer: Callable = nn.GELU,
|
| 277 |
+
norm_layer: Callable = nn.LayerNorm,
|
| 278 |
+
is_cross_attention: bool = False,
|
| 279 |
+
):
|
| 280 |
+
super().__init__()
|
| 281 |
+
|
| 282 |
+
self.ln_1 = norm_layer(d_model)
|
| 283 |
+
if is_cross_attention:
|
| 284 |
+
self.ln_1_kv = norm_layer(d_model)
|
| 285 |
+
|
| 286 |
+
self.ln_2 = norm_layer(d_model)
|
| 287 |
+
mlp_width = int(d_model * mlp_ratio)
|
| 288 |
+
self.attn = VisualAttention(d_model, n_head)
|
| 289 |
+
self.mlp = nn.Sequential(OrderedDict([
|
| 290 |
+
("c_fc", nn.Linear(d_model, mlp_width)),
|
| 291 |
+
("gelu", act_layer()),
|
| 292 |
+
("c_proj", nn.Linear(mlp_width, d_model))
|
| 293 |
+
]))
|
| 294 |
+
|
| 295 |
+
def attention(
|
| 296 |
+
self,
|
| 297 |
+
q_x: torch.Tensor,
|
| 298 |
+
k_x: Optional[torch.Tensor] = None,
|
| 299 |
+
v_x: Optional[torch.Tensor] = None,
|
| 300 |
+
attn_mask: Optional[torch.Tensor] = None,
|
| 301 |
+
):
|
| 302 |
+
k_x = k_x if k_x is not None else q_x
|
| 303 |
+
v_x = v_x if v_x is not None else q_x
|
| 304 |
+
|
| 305 |
+
attn_mask = attn_mask.to(q_x.dtype) if attn_mask is not None else None
|
| 306 |
+
return self.attn(q_x, k_x, v_x, attn_mask=attn_mask)
|
| 307 |
+
|
| 308 |
+
def forward(
|
| 309 |
+
self,
|
| 310 |
+
q_x: torch.Tensor,
|
| 311 |
+
k_x: Optional[torch.Tensor] = None,
|
| 312 |
+
v_x: Optional[torch.Tensor] = None,
|
| 313 |
+
attn_mask: Optional[torch.Tensor] = None,
|
| 314 |
+
):
|
| 315 |
+
k_x = self.ln_1_kv(k_x) if hasattr(self, "ln_1_kv") and k_x is not None else None
|
| 316 |
+
v_x = self.ln_1_kv(v_x) if hasattr(self, "ln_1_kv") and v_x is not None else None
|
| 317 |
+
|
| 318 |
+
x = q_x + self.attention(q_x=self.ln_1(q_x), k_x=k_x, v_x=v_x, attn_mask=attn_mask)
|
| 319 |
+
x = x + self.mlp(self.ln_2(x))
|
| 320 |
+
return x
|
| 321 |
+
|
| 322 |
+
|
| 323 |
+
class TransformerBlock(nn.Module):
|
| 324 |
+
def __init__(
|
| 325 |
+
self,
|
| 326 |
+
width: int,
|
| 327 |
+
layers: int,
|
| 328 |
+
heads: int,
|
| 329 |
+
mlp_ratio: float = 4.0,
|
| 330 |
+
act_layer: Callable = nn.GELU,
|
| 331 |
+
norm_layer: Callable = nn.LayerNorm,
|
| 332 |
+
):
|
| 333 |
+
super().__init__()
|
| 334 |
+
self.width = width
|
| 335 |
+
self.layers = layers
|
| 336 |
+
|
| 337 |
+
self.resblocks = nn.ModuleList([
|
| 338 |
+
VisualAttentionBlock(
|
| 339 |
+
width, heads, mlp_ratio, act_layer=act_layer, norm_layer=norm_layer)
|
| 340 |
+
for _ in range(layers)
|
| 341 |
+
])
|
| 342 |
+
|
| 343 |
+
def get_cast_dtype(self) -> torch.dtype:
|
| 344 |
+
return self.resblocks[0].mlp.c_fc.weight.dtype
|
| 345 |
+
|
| 346 |
+
def get_cast_device(self) -> torch.device:
|
| 347 |
+
return self.resblocks[0].mlp.c_fc.weight.device
|
| 348 |
+
|
| 349 |
+
def forward(self, x: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
|
| 350 |
+
for r in self.resblocks:
|
| 351 |
+
x = r(x, attn_mask=attn_mask)
|
| 352 |
+
return x
|
| 353 |
+
|
| 354 |
+
|
| 355 |
+
class VisionTransformer(nn.Module):
|
| 356 |
+
|
| 357 |
+
def __init__(
|
| 358 |
+
self,
|
| 359 |
+
image_size: int,
|
| 360 |
+
patch_size: int,
|
| 361 |
+
width: int,
|
| 362 |
+
layers: int,
|
| 363 |
+
heads: int,
|
| 364 |
+
mlp_ratio: float,
|
| 365 |
+
n_queries: int = 256,
|
| 366 |
+
output_dim: int = 512,
|
| 367 |
+
**kwargs
|
| 368 |
+
):
|
| 369 |
+
super().__init__()
|
| 370 |
+
image_height, image_width = self.image_size = (image_size, image_size)
|
| 371 |
+
patch_height, patch_width = self.patch_size = (patch_size, patch_size)
|
| 372 |
+
self.grid_size = (image_height // patch_height, image_width // patch_width)
|
| 373 |
+
self.output_dim = output_dim
|
| 374 |
+
|
| 375 |
+
mean = (0.48145466, 0.4578275, 0.40821073)
|
| 376 |
+
std = (0.26862954, 0.26130258, 0.27577711)
|
| 377 |
+
self.image_transform = transforms.Compose([
|
| 378 |
+
transforms.Resize(
|
| 379 |
+
(image_size*2, image_size*2),
|
| 380 |
+
interpolation=InterpolationMode.BICUBIC
|
| 381 |
+
),
|
| 382 |
+
transforms.ToTensor(),
|
| 383 |
+
transforms.Normalize(mean=mean, std=std),
|
| 384 |
+
])
|
| 385 |
+
|
| 386 |
+
self.conv1 = nn.Conv2d(in_channels=3, out_channels=width, kernel_size=patch_size, stride=patch_size, bias=False)
|
| 387 |
+
|
| 388 |
+
# class embeddings and positional embeddings
|
| 389 |
+
scale = width ** -0.5
|
| 390 |
+
self.positional_embedding = nn.Parameter(scale * torch.randn(256, width))
|
| 391 |
+
|
| 392 |
+
norm_layer = partial(nn.LayerNorm, eps=1e-6)
|
| 393 |
+
act_layer = nn.GELU
|
| 394 |
+
|
| 395 |
+
self.ln_pre = norm_layer(width)
|
| 396 |
+
self.transformer = TransformerBlock(
|
| 397 |
+
width,
|
| 398 |
+
layers,
|
| 399 |
+
heads,
|
| 400 |
+
mlp_ratio,
|
| 401 |
+
act_layer=act_layer,
|
| 402 |
+
norm_layer=norm_layer,
|
| 403 |
+
)
|
| 404 |
+
# pdb.set_trace()
|
| 405 |
+
self.attn_pool = Resampler(
|
| 406 |
+
grid_size=int(math.sqrt(n_queries)),
|
| 407 |
+
embed_dim=output_dim,
|
| 408 |
+
num_heads=output_dim // 128,
|
| 409 |
+
kv_dim=width,
|
| 410 |
+
norm_layer=norm_layer,
|
| 411 |
+
)
|
| 412 |
+
self.attn_pool2 = Resampler(
|
| 413 |
+
grid_size=int(math.sqrt(n_queries)),
|
| 414 |
+
embed_dim=output_dim,
|
| 415 |
+
num_heads=output_dim // 128,
|
| 416 |
+
kv_dim=width,
|
| 417 |
+
norm_layer=norm_layer,
|
| 418 |
+
)
|
| 419 |
+
self.ln_post = norm_layer(output_dim)
|
| 420 |
+
self.proj = nn.Parameter((output_dim** -0.5) * torch.randn(output_dim, output_dim))
|
| 421 |
+
# self.attn_pool2.load_state_dict(torch.load('/cfs/cfs-lugcocyb/mingdali/code/qWen-VL/vl-chat/attn_params.pth'))
|
| 422 |
+
|
| 423 |
+
# def initialize_vision_modules(self,lpath):
|
| 424 |
+
# self.attn_pool2[0].load_state_dict(torch.load(lpath))
|
| 425 |
+
|
| 426 |
+
def forward(self, x: torch.Tensor):
|
| 427 |
+
#pdb.set_trace()
|
| 428 |
+
#torch.save(self.attn_pool.state_dict(), '/cfs/cfs-lugcocyb/mingdali/code/qWen-VL/vl-chat/attn_params.pth')
|
| 429 |
+
x = x.to(
|
| 430 |
+
dtype=self.transformer.get_cast_dtype(),
|
| 431 |
+
device=self.transformer.get_cast_device(),
|
| 432 |
+
)
|
| 433 |
+
# to patches
|
| 434 |
+
x = self.conv1(x) # shape = [*, width, grid, grid]
|
| 435 |
+
x = x.reshape(x.shape[0], x.shape[1], -1) # shape = [*, width, grid ** 2]
|
| 436 |
+
x = x.permute(0, 2, 1) # shape = [*, grid ** 2, width]
|
| 437 |
+
|
| 438 |
+
x = x + get_abs_pos(self.positional_embedding, x.size(1))
|
| 439 |
+
|
| 440 |
+
x = self.ln_pre(x)
|
| 441 |
+
|
| 442 |
+
x = x.permute(1, 0, 2) # NLD -> LND
|
| 443 |
+
x = self.transformer(x)
|
| 444 |
+
x = x.permute(1, 0, 2) # LND -> NLD
|
| 445 |
+
# pdb.set_trace()
|
| 446 |
+
src_size = int(math.sqrt(x.shape[1]))
|
| 447 |
+
x = x.reshape(x.shape[0]//5,5,-1, x.shape[-1])
|
| 448 |
+
x1 = x[:,4,:,:]
|
| 449 |
+
x = x[:,:4,:,:]
|
| 450 |
+
x = x.reshape(x.shape[0], -1, src_size, src_size, x.shape[-1])
|
| 451 |
+
x = x.transpose(1,2).reshape(x.shape[0], src_size,2,2, src_size, x.shape[-1])
|
| 452 |
+
x = x.transpose(1,2).reshape(x.shape[0], -1, x.shape[-1])
|
| 453 |
+
x = self.attn_pool2(x)
|
| 454 |
+
x1 = self.attn_pool(x1)
|
| 455 |
+
x = self.post_pro(x)
|
| 456 |
+
x1 = self.post_pro(x1)
|
| 457 |
+
# return x1
|
| 458 |
+
return torch.cat([x,x1],dim=1)
|
| 459 |
+
|
| 460 |
+
def post_pro(self, x):
|
| 461 |
+
x = self.ln_post(x)
|
| 462 |
+
x = x @ self.proj
|
| 463 |
+
return x
|
| 464 |
+
|
| 465 |
+
|
| 466 |
+
def encode(self, image_paths: List[str]):
|
| 467 |
+
images = []
|
| 468 |
+
# pdb.set_trace()
|
| 469 |
+
for image_path in image_paths:
|
| 470 |
+
try:
|
| 471 |
+
if image_path.startswith("http://") or image_path.startswith("https://"):
|
| 472 |
+
image = Image.open(requests.get(image_path, stream=True).raw)
|
| 473 |
+
else:
|
| 474 |
+
image = self.image_transform(Image.open(image_path).convert("RGB"))
|
| 475 |
+
except:
|
| 476 |
+
image = torch.zeros((3, 448*2, 448*2))
|
| 477 |
+
# pdb.set_trace()
|
| 478 |
+
images.append(image)
|
| 479 |
+
images = torch.stack(images, dim=0)
|
| 480 |
+
windows = sliding_window(images,window_size=(448,448),stride=448)
|
| 481 |
+
return self(windows)
|
| 482 |
+
|