update json
#2
by haoyang-amd - opened
- config.json +0 -35
config.json
CHANGED
|
@@ -123,41 +123,6 @@
|
|
| 123 |
"scale_type": "float",
|
| 124 |
"symmetric": true
|
| 125 |
}
|
| 126 |
-
},
|
| 127 |
-
"model.layers.61.*": {
|
| 128 |
-
"bias": null,
|
| 129 |
-
"input_tensors": {
|
| 130 |
-
"ch_axis": -1,
|
| 131 |
-
"dtype": "fp4",
|
| 132 |
-
"group_size": 32,
|
| 133 |
-
"is_dynamic": true,
|
| 134 |
-
"is_scale_quant": false,
|
| 135 |
-
"mx_element_dtype": null,
|
| 136 |
-
"observer_cls": "PerBlockMXObserver",
|
| 137 |
-
"qscheme": "per_group",
|
| 138 |
-
"round_method": "half_even",
|
| 139 |
-
"scale_calculation_mode": "even",
|
| 140 |
-
"scale_format": "e8m0",
|
| 141 |
-
"scale_type": "float",
|
| 142 |
-
"symmetric": null
|
| 143 |
-
},
|
| 144 |
-
"output_tensors": null,
|
| 145 |
-
"target_device": null,
|
| 146 |
-
"weight": {
|
| 147 |
-
"ch_axis": -1,
|
| 148 |
-
"dtype": "fp4",
|
| 149 |
-
"group_size": 32,
|
| 150 |
-
"is_dynamic": false,
|
| 151 |
-
"is_scale_quant": false,
|
| 152 |
-
"mx_element_dtype": null,
|
| 153 |
-
"observer_cls": "PerBlockMXObserver",
|
| 154 |
-
"qscheme": "per_group",
|
| 155 |
-
"round_method": "half_even",
|
| 156 |
-
"scale_calculation_mode": "even",
|
| 157 |
-
"scale_format": "e8m0",
|
| 158 |
-
"scale_type": "float",
|
| 159 |
-
"symmetric": null
|
| 160 |
-
}
|
| 161 |
}
|
| 162 |
},
|
| 163 |
"layer_type_quant_config": {},
|
|
|
|
| 123 |
"scale_type": "float",
|
| 124 |
"symmetric": true
|
| 125 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 126 |
}
|
| 127 |
},
|
| 128 |
"layer_type_quant_config": {},
|