Hugging Face's logo

torchao-testing
/
opt-125m-ModuleFqnToConfig-v1-regex-0.14.0.dev

PyTorch
opt
torchao
Model card Files
xet
Community
jerryzh168 commited on
Commit
3846aef
·
verified ·
1 Parent(s): e6d0cfd

Upload OPTForCausalLM

Browse files
Files changed (1) hide show
  1. config.json +108 -108
config.json CHANGED
@@ -62,7 +62,111 @@
62
  "_type": "IntxWeightOnlyConfig",
63
  "_version": 2
64
  },
65
- "model.decoder.layers.*.self_attn.k_proj": {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
66
  "_data": {
67
  "activation_dtype": {
68
  "_data": "float8_e4m3fn",
@@ -104,7 +208,7 @@
104
  "_type": "Float8DynamicActivationFloat8WeightConfig",
105
  "_version": 2
106
  },
107
- "model.decoder.layers.*.self_attn.q_proj": {
108
  "_data": {
109
  "activation_dtype": {
110
  "_data": "float8_e4m3fn",
@@ -146,7 +250,7 @@
146
  "_type": "Float8DynamicActivationFloat8WeightConfig",
147
  "_version": 2
148
  },
149
- "model.decoder.layers.*.self_attn.qkv_proj": {
150
  "_data": {
151
  "activation_dtype": {
152
  "_data": "float8_e4m3fn",
@@ -188,7 +292,7 @@
188
  "_type": "Float8DynamicActivationFloat8WeightConfig",
189
  "_version": 2
190
  },
191
- "model.decoder.layers.*.self_attn.v_proj": {
192
  "_data": {
193
  "activation_dtype": {
194
  "_data": "float8_e4m3fn",
@@ -229,110 +333,6 @@
229
  },
230
  "_type": "Float8DynamicActivationFloat8WeightConfig",
231
  "_version": 2
232
- },
233
- "model.decoder.layers.3.self_attn.k_proj": {
234
- "_data": {
235
- "group_size": 128,
236
- "int4_choose_qparams_algorithm": {
237
- "_data": "TINYGEMM",
238
- "_type": "Int4ChooseQParamsAlgorithm"
239
- },
240
- "int4_packing_format": "tile_packed_to_4d",
241
- "layout": {
242
- "_data": {
243
- "inner_k_tiles": 8
244
- },
245
- "_type": "TensorCoreTiledLayout",
246
- "_version": 1
247
- },
248
- "preserve_zero": null,
249
- "set_inductor_config": true,
250
- "use_hqq": false,
251
- "zero_point_domain": {
252
- "_data": "NONE",
253
- "_type": "ZeroPointDomain"
254
- }
255
- },
256
- "_type": "Int4WeightOnlyConfig",
257
- "_version": 2
258
- },
259
- "model.decoder.layers.3.self_attn.q_proj": {
260
- "_data": {
261
- "group_size": 128,
262
- "int4_choose_qparams_algorithm": {
263
- "_data": "TINYGEMM",
264
- "_type": "Int4ChooseQParamsAlgorithm"
265
- },
266
- "int4_packing_format": "tile_packed_to_4d",
267
- "layout": {
268
- "_data": {
269
- "inner_k_tiles": 8
270
- },
271
- "_type": "TensorCoreTiledLayout",
272
- "_version": 1
273
- },
274
- "preserve_zero": null,
275
- "set_inductor_config": true,
276
- "use_hqq": false,
277
- "zero_point_domain": {
278
- "_data": "NONE",
279
- "_type": "ZeroPointDomain"
280
- }
281
- },
282
- "_type": "Int4WeightOnlyConfig",
283
- "_version": 2
284
- },
285
- "model.decoder.layers.3.self_attn.qkv_proj": {
286
- "_data": {
287
- "group_size": 128,
288
- "int4_choose_qparams_algorithm": {
289
- "_data": "TINYGEMM",
290
- "_type": "Int4ChooseQParamsAlgorithm"
291
- },
292
- "int4_packing_format": "tile_packed_to_4d",
293
- "layout": {
294
- "_data": {
295
- "inner_k_tiles": 8
296
- },
297
- "_type": "TensorCoreTiledLayout",
298
- "_version": 1
299
- },
300
- "preserve_zero": null,
301
- "set_inductor_config": true,
302
- "use_hqq": false,
303
- "zero_point_domain": {
304
- "_data": "NONE",
305
- "_type": "ZeroPointDomain"
306
- }
307
- },
308
- "_type": "Int4WeightOnlyConfig",
309
- "_version": 2
310
- },
311
- "model.decoder.layers.3.self_attn.v_proj": {
312
- "_data": {
313
- "group_size": 128,
314
- "int4_choose_qparams_algorithm": {
315
- "_data": "TINYGEMM",
316
- "_type": "Int4ChooseQParamsAlgorithm"
317
- },
318
- "int4_packing_format": "tile_packed_to_4d",
319
- "layout": {
320
- "_data": {
321
- "inner_k_tiles": 8
322
- },
323
- "_type": "TensorCoreTiledLayout",
324
- "_version": 1
325
- },
326
- "preserve_zero": null,
327
- "set_inductor_config": true,
328
- "use_hqq": false,
329
- "zero_point_domain": {
330
- "_data": "NONE",
331
- "_type": "ZeroPointDomain"
332
- }
333
- },
334
- "_type": "Int4WeightOnlyConfig",
335
- "_version": 2
336
  }
337
  }
338
  },
 
62
  "_type": "IntxWeightOnlyConfig",
63
  "_version": 2
64
  },
65
+ "model.decoder.layers.3.self_attn.k_proj": {
66
+ "_data": {
67
+ "group_size": 128,
68
+ "int4_choose_qparams_algorithm": {
69
+ "_data": "TINYGEMM",
70
+ "_type": "Int4ChooseQParamsAlgorithm"
71
+ },
72
+ "int4_packing_format": "tile_packed_to_4d",
73
+ "layout": {
74
+ "_data": {
75
+ "inner_k_tiles": 8
76
+ },
77
+ "_type": "TensorCoreTiledLayout",
78
+ "_version": 1
79
+ },
80
+ "preserve_zero": null,
81
+ "set_inductor_config": true,
82
+ "use_hqq": false,
83
+ "zero_point_domain": {
84
+ "_data": "NONE",
85
+ "_type": "ZeroPointDomain"
86
+ }
87
+ },
88
+ "_type": "Int4WeightOnlyConfig",
89
+ "_version": 2
90
+ },
91
+ "model.decoder.layers.3.self_attn.q_proj": {
92
+ "_data": {
93
+ "group_size": 128,
94
+ "int4_choose_qparams_algorithm": {
95
+ "_data": "TINYGEMM",
96
+ "_type": "Int4ChooseQParamsAlgorithm"
97
+ },
98
+ "int4_packing_format": "tile_packed_to_4d",
99
+ "layout": {
100
+ "_data": {
101
+ "inner_k_tiles": 8
102
+ },
103
+ "_type": "TensorCoreTiledLayout",
104
+ "_version": 1
105
+ },
106
+ "preserve_zero": null,
107
+ "set_inductor_config": true,
108
+ "use_hqq": false,
109
+ "zero_point_domain": {
110
+ "_data": "NONE",
111
+ "_type": "ZeroPointDomain"
112
+ }
113
+ },
114
+ "_type": "Int4WeightOnlyConfig",
115
+ "_version": 2
116
+ },
117
+ "model.decoder.layers.3.self_attn.qkv_proj": {
118
+ "_data": {
119
+ "group_size": 128,
120
+ "int4_choose_qparams_algorithm": {
121
+ "_data": "TINYGEMM",
122
+ "_type": "Int4ChooseQParamsAlgorithm"
123
+ },
124
+ "int4_packing_format": "tile_packed_to_4d",
125
+ "layout": {
126
+ "_data": {
127
+ "inner_k_tiles": 8
128
+ },
129
+ "_type": "TensorCoreTiledLayout",
130
+ "_version": 1
131
+ },
132
+ "preserve_zero": null,
133
+ "set_inductor_config": true,
134
+ "use_hqq": false,
135
+ "zero_point_domain": {
136
+ "_data": "NONE",
137
+ "_type": "ZeroPointDomain"
138
+ }
139
+ },
140
+ "_type": "Int4WeightOnlyConfig",
141
+ "_version": 2
142
+ },
143
+ "model.decoder.layers.3.self_attn.v_proj": {
144
+ "_data": {
145
+ "group_size": 128,
146
+ "int4_choose_qparams_algorithm": {
147
+ "_data": "TINYGEMM",
148
+ "_type": "Int4ChooseQParamsAlgorithm"
149
+ },
150
+ "int4_packing_format": "tile_packed_to_4d",
151
+ "layout": {
152
+ "_data": {
153
+ "inner_k_tiles": 8
154
+ },
155
+ "_type": "TensorCoreTiledLayout",
156
+ "_version": 1
157
+ },
158
+ "preserve_zero": null,
159
+ "set_inductor_config": true,
160
+ "use_hqq": false,
161
+ "zero_point_domain": {
162
+ "_data": "NONE",
163
+ "_type": "ZeroPointDomain"
164
+ }
165
+ },
166
+ "_type": "Int4WeightOnlyConfig",
167
+ "_version": 2
168
+ },
169
+ "re:model\\.decoder\\.layers\\..+\\.self_attn\\.k_proj": {
170
  "_data": {
171
  "activation_dtype": {
172
  "_data": "float8_e4m3fn",
 
208
  "_type": "Float8DynamicActivationFloat8WeightConfig",
209
  "_version": 2
210
  },
211
+ "re:model\\.decoder\\.layers\\..+\\.self_attn\\.q_proj": {
212
  "_data": {
213
  "activation_dtype": {
214
  "_data": "float8_e4m3fn",
 
250
  "_type": "Float8DynamicActivationFloat8WeightConfig",
251
  "_version": 2
252
  },
253
+ "re:model\\.decoder\\.layers\\..+\\.self_attn\\.qkv_proj": {
254
  "_data": {
255
  "activation_dtype": {
256
  "_data": "float8_e4m3fn",
 
292
  "_type": "Float8DynamicActivationFloat8WeightConfig",
293
  "_version": 2
294
  },
295
+ "re:model\\.decoder\\.layers\\..+\\.self_attn\\.v_proj": {
296
  "_data": {
297
  "activation_dtype": {
298
  "_data": "float8_e4m3fn",
 
333
  },
334
  "_type": "Float8DynamicActivationFloat8WeightConfig",
335
  "_version": 2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
336
  }
337
  }
338
  },