shubhrapandit commited on
Commit
87784ea
·
1 Parent(s): 0d21073

Update files to match base model keeping quant_config intact

Browse files
Files changed (1) hide show
  1. config.json +204 -0
config.json CHANGED
@@ -28,6 +28,210 @@
28
  "transformers_version": "4.41.2",
29
  "use_cache": true,
30
  "use_sliding_window": false,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
  "vision_config": {
32
  "depth": 32,
33
  "hidden_act": "silu",
 
28
  "transformers_version": "4.41.2",
29
  "use_cache": true,
30
  "use_sliding_window": false,
31
+ "quantization_config": {
32
+ "config_groups": {
33
+ "group_0": {
34
+ "input_activations": {
35
+ "actorder": null,
36
+ "block_structure": null,
37
+ "dynamic": true,
38
+ "group_size": null,
39
+ "num_bits": 8,
40
+ "observer": null,
41
+ "observer_kwargs": {},
42
+ "strategy": "token",
43
+ "symmetric": true,
44
+ "type": "int"
45
+ },
46
+ "output_activations": null,
47
+ "targets": [
48
+ "Linear"
49
+ ],
50
+ "weights": {
51
+ "actorder": null,
52
+ "block_structure": null,
53
+ "dynamic": false,
54
+ "group_size": null,
55
+ "num_bits": 8,
56
+ "observer": "minmax",
57
+ "observer_kwargs": {},
58
+ "strategy": "channel",
59
+ "symmetric": true,
60
+ "type": "int"
61
+ }
62
+ }
63
+ },
64
+ "format": "int-quantized",
65
+ "global_compression_ratio": 1.323305644571225,
66
+ "ignore": [
67
+ "visual.blocks.0.attn.qkv",
68
+ "visual.blocks.0.attn.proj",
69
+ "visual.blocks.0.mlp.gate_proj",
70
+ "visual.blocks.0.mlp.up_proj",
71
+ "visual.blocks.0.mlp.down_proj",
72
+ "visual.blocks.1.attn.qkv",
73
+ "visual.blocks.1.attn.proj",
74
+ "visual.blocks.1.mlp.gate_proj",
75
+ "visual.blocks.1.mlp.up_proj",
76
+ "visual.blocks.1.mlp.down_proj",
77
+ "visual.blocks.2.attn.qkv",
78
+ "visual.blocks.2.attn.proj",
79
+ "visual.blocks.2.mlp.gate_proj",
80
+ "visual.blocks.2.mlp.up_proj",
81
+ "visual.blocks.2.mlp.down_proj",
82
+ "visual.blocks.3.attn.qkv",
83
+ "visual.blocks.3.attn.proj",
84
+ "visual.blocks.3.mlp.gate_proj",
85
+ "visual.blocks.3.mlp.up_proj",
86
+ "visual.blocks.3.mlp.down_proj",
87
+ "visual.blocks.4.attn.qkv",
88
+ "visual.blocks.4.attn.proj",
89
+ "visual.blocks.4.mlp.gate_proj",
90
+ "visual.blocks.4.mlp.up_proj",
91
+ "visual.blocks.4.mlp.down_proj",
92
+ "visual.blocks.5.attn.qkv",
93
+ "visual.blocks.5.attn.proj",
94
+ "visual.blocks.5.mlp.gate_proj",
95
+ "visual.blocks.5.mlp.up_proj",
96
+ "visual.blocks.5.mlp.down_proj",
97
+ "visual.blocks.6.attn.qkv",
98
+ "visual.blocks.6.attn.proj",
99
+ "visual.blocks.6.mlp.gate_proj",
100
+ "visual.blocks.6.mlp.up_proj",
101
+ "visual.blocks.6.mlp.down_proj",
102
+ "visual.blocks.7.attn.qkv",
103
+ "visual.blocks.7.attn.proj",
104
+ "visual.blocks.7.mlp.gate_proj",
105
+ "visual.blocks.7.mlp.up_proj",
106
+ "visual.blocks.7.mlp.down_proj",
107
+ "visual.blocks.8.attn.qkv",
108
+ "visual.blocks.8.attn.proj",
109
+ "visual.blocks.8.mlp.gate_proj",
110
+ "visual.blocks.8.mlp.up_proj",
111
+ "visual.blocks.8.mlp.down_proj",
112
+ "visual.blocks.9.attn.qkv",
113
+ "visual.blocks.9.attn.proj",
114
+ "visual.blocks.9.mlp.gate_proj",
115
+ "visual.blocks.9.mlp.up_proj",
116
+ "visual.blocks.9.mlp.down_proj",
117
+ "visual.blocks.10.attn.qkv",
118
+ "visual.blocks.10.attn.proj",
119
+ "visual.blocks.10.mlp.gate_proj",
120
+ "visual.blocks.10.mlp.up_proj",
121
+ "visual.blocks.10.mlp.down_proj",
122
+ "visual.blocks.11.attn.qkv",
123
+ "visual.blocks.11.attn.proj",
124
+ "visual.blocks.11.mlp.gate_proj",
125
+ "visual.blocks.11.mlp.up_proj",
126
+ "visual.blocks.11.mlp.down_proj",
127
+ "visual.blocks.12.attn.qkv",
128
+ "visual.blocks.12.attn.proj",
129
+ "visual.blocks.12.mlp.gate_proj",
130
+ "visual.blocks.12.mlp.up_proj",
131
+ "visual.blocks.12.mlp.down_proj",
132
+ "visual.blocks.13.attn.qkv",
133
+ "visual.blocks.13.attn.proj",
134
+ "visual.blocks.13.mlp.gate_proj",
135
+ "visual.blocks.13.mlp.up_proj",
136
+ "visual.blocks.13.mlp.down_proj",
137
+ "visual.blocks.14.attn.qkv",
138
+ "visual.blocks.14.attn.proj",
139
+ "visual.blocks.14.mlp.gate_proj",
140
+ "visual.blocks.14.mlp.up_proj",
141
+ "visual.blocks.14.mlp.down_proj",
142
+ "visual.blocks.15.attn.qkv",
143
+ "visual.blocks.15.attn.proj",
144
+ "visual.blocks.15.mlp.gate_proj",
145
+ "visual.blocks.15.mlp.up_proj",
146
+ "visual.blocks.15.mlp.down_proj",
147
+ "visual.blocks.16.attn.qkv",
148
+ "visual.blocks.16.attn.proj",
149
+ "visual.blocks.16.mlp.gate_proj",
150
+ "visual.blocks.16.mlp.up_proj",
151
+ "visual.blocks.16.mlp.down_proj",
152
+ "visual.blocks.17.attn.qkv",
153
+ "visual.blocks.17.attn.proj",
154
+ "visual.blocks.17.mlp.gate_proj",
155
+ "visual.blocks.17.mlp.up_proj",
156
+ "visual.blocks.17.mlp.down_proj",
157
+ "visual.blocks.18.attn.qkv",
158
+ "visual.blocks.18.attn.proj",
159
+ "visual.blocks.18.mlp.gate_proj",
160
+ "visual.blocks.18.mlp.up_proj",
161
+ "visual.blocks.18.mlp.down_proj",
162
+ "visual.blocks.19.attn.qkv",
163
+ "visual.blocks.19.attn.proj",
164
+ "visual.blocks.19.mlp.gate_proj",
165
+ "visual.blocks.19.mlp.up_proj",
166
+ "visual.blocks.19.mlp.down_proj",
167
+ "visual.blocks.20.attn.qkv",
168
+ "visual.blocks.20.attn.proj",
169
+ "visual.blocks.20.mlp.gate_proj",
170
+ "visual.blocks.20.mlp.up_proj",
171
+ "visual.blocks.20.mlp.down_proj",
172
+ "visual.blocks.21.attn.qkv",
173
+ "visual.blocks.21.attn.proj",
174
+ "visual.blocks.21.mlp.gate_proj",
175
+ "visual.blocks.21.mlp.up_proj",
176
+ "visual.blocks.21.mlp.down_proj",
177
+ "visual.blocks.22.attn.qkv",
178
+ "visual.blocks.22.attn.proj",
179
+ "visual.blocks.22.mlp.gate_proj",
180
+ "visual.blocks.22.mlp.up_proj",
181
+ "visual.blocks.22.mlp.down_proj",
182
+ "visual.blocks.23.attn.qkv",
183
+ "visual.blocks.23.attn.proj",
184
+ "visual.blocks.23.mlp.gate_proj",
185
+ "visual.blocks.23.mlp.up_proj",
186
+ "visual.blocks.23.mlp.down_proj",
187
+ "visual.blocks.24.attn.qkv",
188
+ "visual.blocks.24.attn.proj",
189
+ "visual.blocks.24.mlp.gate_proj",
190
+ "visual.blocks.24.mlp.up_proj",
191
+ "visual.blocks.24.mlp.down_proj",
192
+ "visual.blocks.25.attn.qkv",
193
+ "visual.blocks.25.attn.proj",
194
+ "visual.blocks.25.mlp.gate_proj",
195
+ "visual.blocks.25.mlp.up_proj",
196
+ "visual.blocks.25.mlp.down_proj",
197
+ "visual.blocks.26.attn.qkv",
198
+ "visual.blocks.26.attn.proj",
199
+ "visual.blocks.26.mlp.gate_proj",
200
+ "visual.blocks.26.mlp.up_proj",
201
+ "visual.blocks.26.mlp.down_proj",
202
+ "visual.blocks.27.attn.qkv",
203
+ "visual.blocks.27.attn.proj",
204
+ "visual.blocks.27.mlp.gate_proj",
205
+ "visual.blocks.27.mlp.up_proj",
206
+ "visual.blocks.27.mlp.down_proj",
207
+ "visual.blocks.28.attn.qkv",
208
+ "visual.blocks.28.attn.proj",
209
+ "visual.blocks.28.mlp.gate_proj",
210
+ "visual.blocks.28.mlp.up_proj",
211
+ "visual.blocks.28.mlp.down_proj",
212
+ "visual.blocks.29.attn.qkv",
213
+ "visual.blocks.29.attn.proj",
214
+ "visual.blocks.29.mlp.gate_proj",
215
+ "visual.blocks.29.mlp.up_proj",
216
+ "visual.blocks.29.mlp.down_proj",
217
+ "visual.blocks.30.attn.qkv",
218
+ "visual.blocks.30.attn.proj",
219
+ "visual.blocks.30.mlp.gate_proj",
220
+ "visual.blocks.30.mlp.up_proj",
221
+ "visual.blocks.30.mlp.down_proj",
222
+ "visual.blocks.31.attn.qkv",
223
+ "visual.blocks.31.attn.proj",
224
+ "visual.blocks.31.mlp.gate_proj",
225
+ "visual.blocks.31.mlp.up_proj",
226
+ "visual.blocks.31.mlp.down_proj",
227
+ "visual.merger.mlp.0",
228
+ "visual.merger.mlp.2",
229
+ "lm_head"
230
+ ],
231
+ "kv_cache_scheme": null,
232
+ "quant_method": "compressed-tensors",
233
+ "quantization_status": "compressed"
234
+ },
235
  "vision_config": {
236
  "depth": 32,
237
  "hidden_act": "silu",