Upload folder using huggingface_hub
Browse files- config.json +46 -6
config.json
CHANGED
|
@@ -21,7 +21,10 @@
|
|
| 21 |
"pad_token_id": -1,
|
| 22 |
"pretraining_tp": 1,
|
| 23 |
"quantization_config": {
|
| 24 |
-
"
|
|
|
|
|
|
|
|
|
|
| 25 |
"export": {
|
| 26 |
"kv_cache_group": [],
|
| 27 |
"pack_method": "reorder",
|
|
@@ -38,11 +41,48 @@
|
|
| 38 |
]
|
| 39 |
]
|
| 40 |
},
|
| 41 |
-
"
|
| 42 |
-
"
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
},
|
| 47 |
"rms_norm_eps": 1e-06,
|
| 48 |
"rope_scaling": null,
|
|
|
|
| 21 |
"pad_token_id": -1,
|
| 22 |
"pretraining_tp": 1,
|
| 23 |
"quantization_config": {
|
| 24 |
+
"algo_config": null,
|
| 25 |
+
"exclude": [
|
| 26 |
+
"lm_head"
|
| 27 |
+
],
|
| 28 |
"export": {
|
| 29 |
"kv_cache_group": [],
|
| 30 |
"pack_method": "reorder",
|
|
|
|
| 41 |
]
|
| 42 |
]
|
| 43 |
},
|
| 44 |
+
"global_quant_config": {
|
| 45 |
+
"bias": null,
|
| 46 |
+
"input_tensors": {
|
| 47 |
+
"ch_axis": null,
|
| 48 |
+
"dtype": "fp8_e4m3",
|
| 49 |
+
"group_size": null,
|
| 50 |
+
"is_dynamic": false,
|
| 51 |
+
"observer_cls": "PerTensorMinMaxObserver",
|
| 52 |
+
"qscheme": "per_tensor",
|
| 53 |
+
"round_method": null,
|
| 54 |
+
"scale_type": null,
|
| 55 |
+
"symmetric": null
|
| 56 |
+
},
|
| 57 |
+
"output_tensors": {
|
| 58 |
+
"ch_axis": null,
|
| 59 |
+
"dtype": "fp8_e4m3",
|
| 60 |
+
"group_size": null,
|
| 61 |
+
"is_dynamic": false,
|
| 62 |
+
"observer_cls": "PerTensorMinMaxObserver",
|
| 63 |
+
"qscheme": "per_tensor",
|
| 64 |
+
"round_method": null,
|
| 65 |
+
"scale_type": null,
|
| 66 |
+
"symmetric": null
|
| 67 |
+
},
|
| 68 |
+
"target_device": null,
|
| 69 |
+
"weight": {
|
| 70 |
+
"ch_axis": null,
|
| 71 |
+
"dtype": "fp8_e4m3",
|
| 72 |
+
"group_size": null,
|
| 73 |
+
"is_dynamic": false,
|
| 74 |
+
"observer_cls": "PerTensorMinMaxObserver",
|
| 75 |
+
"qscheme": "per_tensor",
|
| 76 |
+
"round_method": null,
|
| 77 |
+
"scale_type": null,
|
| 78 |
+
"symmetric": null
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"layer_quant_config": {},
|
| 82 |
+
"layer_type_quant_config": {},
|
| 83 |
+
"pack_method": "reorder",
|
| 84 |
+
"quant_method": "quark",
|
| 85 |
+
"quant_mode": 1
|
| 86 |
},
|
| 87 |
"rms_norm_eps": 1e-06,
|
| 88 |
"rope_scaling": null,
|