ixim/Image21-INT8
162
1{
2 "component": "text_encoder",
3 "method": "bitsandbytes LLM.int8",
4 "threshold": 6.0,
5 "skip_modules": [
6 "lm_head",
7 "visual"
8 ],
9 "linear8bit_modules": [
10 "model.visual.blocks.0.attn.qkv",
11 "model.visual.blocks.0.attn.proj",
12 "model.visual.blocks.0.mlp.linear_fc1",
13 "model.visual.blocks.0.mlp.linear_fc2",
14 "model.visual.blocks.1.attn.qkv",
15 "model.visual.blocks.1.attn.proj",
16 "model.visual.blocks.1.mlp.linear_fc1",
17 "model.visual.blocks.1.mlp.linear_fc2",
18 "model.visual.blocks.2.attn.qkv",
19 "model.visual.blocks.2.attn.proj",
20 "model.visual.blocks.2.mlp.linear_fc1",
21 "model.visual.blocks.2.mlp.linear_fc2",
22 "model.visual.blocks.3.attn.qkv",
23 "model.visual.blocks.3.attn.proj",
24 "model.visual.blocks.3.mlp.linear_fc1",
25 "model.visual.blocks.3.mlp.linear_fc2",
26 "model.visual.blocks.4.attn.qkv",
27 "model.visual.blocks.4.attn.proj",
28 "model.visual.blocks.4.mlp.linear_fc1",
29 "model.visual.blocks.4.mlp.linear_fc2",
30 "model.visual.blocks.5.attn.qkv",
31 "model.visual.blocks.5.attn.proj",
32 "model.visual.blocks.5.mlp.linear_fc1",
33 "model.visual.blocks.5.mlp.linear_fc2",
34 "model.visual.blocks.6.attn.qkv",
35 "model.visual.blocks.6.attn.proj",
36 "model.visual.blocks.6.mlp.linear_fc1",
37 "model.visual.blocks.6.mlp.linear_fc2",
38 "model.visual.blocks.7.attn.qkv",
39 "model.visual.blocks.7.attn.proj",
40 "model.visual.blocks.7.mlp.linear_fc1",
41 "model.visual.blocks.7.mlp.linear_fc2",
42 "model.visual.blocks.8.attn.qkv",
43 "model.visual.blocks.8.attn.proj",
44 "model.visual.blocks.8.mlp.linear_fc1",
45 "model.visual.blocks.8.mlp.linear_fc2",
46 "model.visual.blocks.9.attn.qkv",
47 "model.visual.blocks.9.attn.proj",
48 "model.visual.blocks.9.mlp.linear_fc1",
49 "model.visual.blocks.9.mlp.linear_fc2",
50 "model.visual.blocks.10.attn.qkv",
51 "model.visual.blocks.10.attn.proj",
52 "model.visual.blocks.10.mlp.linear_fc1",
53 "model.visual.blocks.10.mlp.linear_fc2",
54 "model.visual.blocks.11.attn.qkv",
55 "model.visual.blocks.11.attn.proj",
56 "model.visual.blocks.11.mlp.linear_fc1",
57 "model.visual.blocks.11.mlp.linear_fc2",
58 "model.visual.blocks.12.attn.qkv",
59 "model.visual.blocks.12.attn.proj",
60 "model.visual.blocks.12.mlp.linear_fc1",
61 "model.visual.blocks.12.mlp.linear_fc2",
62 "model.visual.blocks.13.attn.qkv",
63 "model.visual.blocks.13.attn.proj",
64 "model.visual.blocks.13.mlp.linear_fc1",
65 "model.visual.blocks.13.mlp.linear_fc2",
66 "model.visual.blocks.14.attn.qkv",
67 "model.visual.blocks.14.attn.proj",
68 "model.visual.blocks.14.mlp.linear_fc1",
69 "model.visual.blocks.14.mlp.linear_fc2",
70 "model.visual.blocks.15.attn.qkv",
71 "model.visual.blocks.15.attn.proj",
72 "model.visual.blocks.15.mlp.linear_fc1",
73 "model.visual.blocks.15.mlp.linear_fc2",
74 "model.visual.blocks.16.attn.qkv",
75 "model.visual.blocks.16.attn.proj",
76 "model.visual.blocks.16.mlp.linear_fc1",
77 "model.visual.blocks.16.mlp.linear_fc2",
78 "model.visual.blocks.17.attn.qkv",
79 "model.visual.blocks.17.attn.proj",
80 "model.visual.blocks.17.mlp.linear_fc1",
81 "model.visual.blocks.17.mlp.linear_fc2",
82 "model.visual.blocks.18.attn.qkv",
83 "model.visual.blocks.18.attn.proj",
84 "model.visual.blocks.18.mlp.linear_fc1",
85 "model.visual.blocks.18.mlp.linear_fc2",
86 "model.visual.blocks.19.attn.qkv",
87 "model.visual.blocks.19.attn.proj",
88 "model.visual.blocks.19.mlp.linear_fc1",
89 "model.visual.blocks.19.mlp.linear_fc2",
90 "model.visual.blocks.20.attn.qkv",
91 "model.visual.blocks.20.attn.proj",
92 "model.visual.blocks.20.mlp.linear_fc1",
93 "model.visual.blocks.20.mlp.linear_fc2",
94 "model.visual.blocks.21.attn.qkv",
95 "model.visual.blocks.21.attn.proj",
96 "model.visual.blocks.21.mlp.linear_fc1",
97 "model.visual.blocks.21.mlp.linear_fc2",
98 "model.visual.blocks.22.attn.qkv",
99 "model.visual.blocks.22.attn.proj",
100 "model.visual.blocks.22.mlp.linear_fc1",
101 "model.visual.blocks.22.mlp.linear_fc2",
102 "model.visual.blocks.23.attn.qkv",
103 "model.visual.blocks.23.attn.proj",
104 "model.visual.blocks.23.mlp.linear_fc1",
105 "model.visual.blocks.23.mlp.linear_fc2",
106 "model.visual.blocks.24.attn.qkv",
107 "model.visual.blocks.24.attn.proj",
108 "model.visual.blocks.24.mlp.linear_fc1",
109 "model.visual.blocks.24.mlp.linear_fc2",
110 "model.visual.blocks.25.attn.qkv",
111 "model.visual.blocks.25.attn.proj",
112 "model.visual.blocks.25.mlp.linear_fc1",
113 "model.visual.blocks.25.mlp.linear_fc2",
114 "model.visual.blocks.26.attn.qkv",
115 "model.visual.blocks.26.attn.proj",
116 "model.visual.blocks.26.mlp.linear_fc1",
117 "model.visual.blocks.26.mlp.linear_fc2",
118 "model.visual.merger.linear_fc1",
119 "model.visual.merger.linear_fc2",
120 "model.visual.deepstack_merger_list.0.linear_fc1",
121 "model.visual.deepstack_merger_list.0.linear_fc2",
122 "model.visual.deepstack_merger_list.1.linear_fc1",
123 "model.visual.deepstack_merger_list.1.linear_fc2",
124 "model.visual.deepstack_merger_list.2.linear_fc1",
125 "model.visual.deepstack_merger_list.2.linear_fc2",
126 "model.language_model.layers.0.self_attn.q_proj",
127 "model.language_model.layers.0.self_attn.k_proj",
128 "model.language_model.layers.0.self_attn.v_proj",
129 "model.language_model.layers.0.self_attn.o_proj",
130 "model.language_model.layers.0.mlp.gate_proj",
131 "model.language_model.layers.0.mlp.up_proj",
132 "model.language_model.layers.0.mlp.down_proj",
133 "model.language_model.layers.1.self_attn.q_proj",
134 "model.language_model.layers.1.self_attn.k_proj",
135 "model.language_model.layers.1.self_attn.v_proj",
136 "model.language_model.layers.1.self_attn.o_proj",
137 "model.language_model.layers.1.mlp.gate_proj",
138 "model.language_model.layers.1.mlp.up_proj",
139 "model.language_model.layers.1.mlp.down_proj",
140 "model.language_model.layers.2.self_attn.q_proj",
141 "model.language_model.layers.2.self_attn.k_proj",
142 "model.language_model.layers.2.self_attn.v_proj",
143 "model.language_model.layers.2.self_attn.o_proj",
144 "model.language_model.layers.2.mlp.gate_proj",
145 "model.language_model.layers.2.mlp.up_proj",
146 "model.language_model.layers.2.mlp.down_proj",
147 "model.language_model.layers.3.self_attn.q_proj",
148 "model.language_model.layers.3.self_attn.k_proj",
149 "model.language_model.layers.3.self_attn.v_proj",
150 "model.language_model.layers.3.self_attn.o_proj",
151 "model.language_model.layers.3.mlp.gate_proj",
152 "model.language_model.layers.3.mlp.up_proj",
153 "model.language_model.layers.3.mlp.down_proj",
154 "model.language_model.layers.4.self_attn.q_proj",
155 "model.language_model.layers.4.self_attn.k_proj",
156 "model.language_model.layers.4.self_attn.v_proj",
157 "model.language_model.layers.4.self_attn.o_proj",
158 "model.language_model.layers.4.mlp.gate_proj",
159 "model.language_model.layers.4.mlp.up_proj",
160 "model.language_model.layers.4.mlp.down_proj",
161 "model.language_model.layers.5.self_attn.q_proj",
162 "model.language_model.layers.5.self_attn.k_proj",
163 "model.language_model.layers.5.self_attn.v_proj",
164 "model.language_model.layers.5.self_attn.o_proj",
165 "model.language_model.layers.5.mlp.gate_proj",
166 "model.language_model.layers.5.mlp.up_proj",
167 "model.language_model.layers.5.mlp.down_proj",
168 "model.language_model.layers.6.self_attn.q_proj",
169 "model.language_model.layers.6.self_attn.k_proj",
170 "model.language_model.layers.6.self_attn.v_proj",
171 "model.language_model.layers.6.self_attn.o_proj",
172 "model.language_model.layers.6.mlp.gate_proj",
173 "model.language_model.layers.6.mlp.up_proj",
174 "model.language_model.layers.6.mlp.down_proj",
175 "model.language_model.layers.7.self_attn.q_proj",
176 "model.language_model.layers.7.self_attn.k_proj",
177 "model.language_model.layers.7.self_attn.v_proj",
178 "model.language_model.layers.7.self_attn.o_proj",
179 "model.language_model.layers.7.mlp.gate_proj",
180 "model.language_model.layers.7.mlp.up_proj",
181 "model.language_model.layers.7.mlp.down_proj",
182 "model.language_model.layers.8.self_attn.q_proj",
183 "model.language_model.layers.8.self_attn.k_proj",
184 "model.language_model.layers.8.self_attn.v_proj",
185 "model.language_model.layers.8.self_attn.o_proj",
186 "model.language_model.layers.8.mlp.gate_proj",
187 "model.language_model.layers.8.mlp.up_proj",
188 "model.language_model.layers.8.mlp.down_proj",
189 "model.language_model.layers.9.self_attn.q_proj",
190 "model.language_model.layers.9.self_attn.k_proj",
191 "model.language_model.layers.9.self_attn.v_proj",
192 "model.language_model.layers.9.self_attn.o_proj",
193 "model.language_model.layers.9.mlp.gate_proj",
194 "model.language_model.layers.9.mlp.up_proj",
195 "model.language_model.layers.9.mlp.down_proj",
196 "model.language_model.layers.10.self_attn.q_proj",
197 "model.language_model.layers.10.self_attn.k_proj",
198 "model.language_model.layers.10.self_attn.v_proj",
199 "model.language_model.layers.10.self_attn.o_proj",
200 "model.language_model.layers.10.mlp.gate_proj",
201 "model.language_model.layers.10.mlp.up_proj",
202 "model.language_model.layers.10.mlp.down_proj",
203 "model.language_model.layers.11.self_attn.q_proj",
204 "model.language_model.layers.11.self_attn.k_proj",
205 "model.language_model.layers.11.self_attn.v_proj",
206 "model.language_model.layers.11.self_attn.o_proj",
207 "model.language_model.layers.11.mlp.gate_proj",
208 "model.language_model.layers.11.mlp.up_proj",
209 "model.language_model.layers.11.mlp.down_proj",
210 "model.language_model.layers.12.self_attn.q_proj",
211 "model.language_model.layers.12.self_attn.k_proj",
212 "model.language_model.layers.12.self_attn.v_proj",
213 "model.language_model.layers.12.self_attn.o_proj",
214 "model.language_model.layers.12.mlp.gate_proj",
215 "model.language_model.layers.12.mlp.up_proj",
216 "model.language_model.layers.12.mlp.down_proj",
217 "model.language_model.layers.13.self_attn.q_proj",
218 "model.language_model.layers.13.self_attn.k_proj",
219 "model.language_model.layers.13.self_attn.v_proj",
220 "model.language_model.layers.13.self_attn.o_proj",
221 "model.language_model.layers.13.mlp.gate_proj",
222 "model.language_model.layers.13.mlp.up_proj",
223 "model.language_model.layers.13.mlp.down_proj",
224 "model.language_model.layers.14.self_attn.q_proj",
225 "model.language_model.layers.14.self_attn.k_proj",
226 "model.language_model.layers.14.self_attn.v_proj",
227 "model.language_model.layers.14.self_attn.o_proj",
228 "model.language_model.layers.14.mlp.gate_proj",
229 "model.language_model.layers.14.mlp.up_proj",
230 "model.language_model.layers.14.mlp.down_proj",
231 "model.language_model.layers.15.self_attn.q_proj",
232 "model.language_model.layers.15.self_attn.k_proj",
233 "model.language_model.layers.15.self_attn.v_proj",
234 "model.language_model.layers.15.self_attn.o_proj",
235 "model.language_model.layers.15.mlp.gate_proj",
236 "model.language_model.layers.15.mlp.up_proj",
237 "model.language_model.layers.15.mlp.down_proj",
238 "model.language_model.layers.16.self_attn.q_proj",
239 "model.language_model.layers.16.self_attn.k_proj",
240 "model.language_model.layers.16.self_attn.v_proj",
241 "model.language_model.layers.16.self_attn.o_proj",
242 "model.language_model.layers.16.mlp.gate_proj",
243 "model.language_model.layers.16.mlp.up_proj",
244 "model.language_model.layers.16.mlp.down_proj",
245 "model.language_model.layers.17.self_attn.q_proj",
246 "model.language_model.layers.17.self_attn.k_proj",
247 "model.language_model.layers.17.self_attn.v_proj",
248 "model.language_model.layers.17.self_attn.o_proj",
249 "model.language_model.layers.17.mlp.gate_proj",
250 "model.language_model.layers.17.mlp.up_proj",
251 "model.language_model.layers.17.mlp.down_proj",
252 "model.language_model.layers.18.self_attn.q_proj",
253 "model.language_model.layers.18.self_attn.k_proj",
254 "model.language_model.layers.18.self_attn.v_proj",
255 "model.language_model.layers.18.self_attn.o_proj",
256 "model.language_model.layers.18.mlp.gate_proj",
257 "model.language_model.layers.18.mlp.up_proj",
258 "model.language_model.layers.18.mlp.down_proj",
259 "model.language_model.layers.19.self_attn.q_proj",
260 "model.language_model.layers.19.self_attn.k_proj",
261 "model.language_model.layers.19.self_attn.v_proj",
262 "model.language_model.layers.19.self_attn.o_proj",
263 "model.language_model.layers.19.mlp.gate_proj",
264 "model.language_model.layers.19.mlp.up_proj",
265 "model.language_model.layers.19.mlp.down_proj",
266 "model.language_model.layers.20.self_attn.q_proj",
267 "model.language_model.layers.20.self_attn.k_proj",
268 "model.language_model.layers.20.self_attn.v_proj",
269 "model.language_model.layers.20.self_attn.o_proj",
270 "model.language_model.layers.20.mlp.gate_proj",
271 "model.language_model.layers.20.mlp.up_proj",
272 "model.language_model.layers.20.mlp.down_proj",
273 "model.language_model.layers.21.self_attn.q_proj",
274 "model.language_model.layers.21.self_attn.k_proj",
275 "model.language_model.layers.21.self_attn.v_proj",
276 "model.language_model.layers.21.self_attn.o_proj",
277 "model.language_model.layers.21.mlp.gate_proj",
278 "model.language_model.layers.21.mlp.up_proj",
279 "model.language_model.layers.21.mlp.down_proj",
280 "model.language_model.layers.22.self_attn.q_proj",
281 "model.language_model.layers.22.self_attn.k_proj",
282 "model.language_model.layers.22.self_attn.v_proj",
283 "model.language_model.layers.22.self_attn.o_proj",
284 "model.language_model.layers.22.mlp.gate_proj",
285 "model.language_model.layers.22.mlp.up_proj",
286 "model.language_model.layers.22.mlp.down_proj",
287 "model.language_model.layers.23.self_attn.q_proj",
288 "model.language_model.layers.23.self_attn.k_proj",
289 "model.language_model.layers.23.self_attn.v_proj",
290 "model.language_model.layers.23.self_attn.o_proj",
291 "model.language_model.layers.23.mlp.gate_proj",
292 "model.language_model.layers.23.mlp.up_proj",
293 "model.language_model.layers.23.mlp.down_proj",
294 "model.language_model.layers.24.self_attn.q_proj",
295 "model.language_model.layers.24.self_attn.k_proj",
296 "model.language_model.layers.24.self_attn.v_proj",
297 "model.language_model.layers.24.self_attn.o_proj",
298 "model.language_model.layers.24.mlp.gate_proj",
299 "model.language_model.layers.24.mlp.up_proj",
300 "model.language_model.layers.24.mlp.down_proj",
301 "model.language_model.layers.25.self_attn.q_proj",
302 "model.language_model.layers.25.self_attn.k_proj",
303 "model.language_model.layers.25.self_attn.v_proj",
304 "model.language_model.layers.25.self_attn.o_proj",
305 "model.language_model.layers.25.mlp.gate_proj",
306 "model.language_model.layers.25.mlp.up_proj",
307 "model.language_model.layers.25.mlp.down_proj",
308 "model.language_model.layers.26.self_attn.q_proj",
309 "model.language_model.layers.26.self_attn.k_proj",
310 "model.language_model.layers.26.self_attn.v_proj",
311 "model.language_model.layers.26.self_attn.o_proj",
312 "model.language_model.layers.26.mlp.gate_proj",
313 "model.language_model.layers.26.mlp.up_proj",
314 "model.language_model.layers.26.mlp.down_proj",
315 "model.language_model.layers.27.self_attn.q_proj",
316 "model.language_model.layers.27.self_attn.k_proj",
317 "model.language_model.layers.27.self_attn.v_proj",
318 "model.language_model.layers.27.self_attn.o_proj",
319 "model.language_model.layers.27.mlp.gate_proj",
320 "model.language_model.layers.27.mlp.up_proj",
321 "model.language_model.layers.27.mlp.down_proj",
322 "model.language_model.layers.28.self_attn.q_proj",
323 "model.language_model.layers.28.self_attn.k_proj",
324 "model.language_model.layers.28.self_attn.v_proj",
325 "model.language_model.layers.28.self_attn.o_proj",
326 "model.language_model.layers.28.mlp.gate_proj",
327 "model.language_model.layers.28.mlp.up_proj",
328 "model.language_model.layers.28.mlp.down_proj",
329 "model.language_model.layers.29.self_attn.q_proj",
330 "model.language_model.layers.29.self_attn.k_proj",
331 "model.language_model.layers.29.self_attn.v_proj",
332 "model.language_model.layers.29.self_attn.o_proj",
333 "model.language_model.layers.29.mlp.gate_proj",
334 "model.language_model.layers.29.mlp.up_proj",
335 "model.language_model.layers.29.mlp.down_proj",
336 "model.language_model.layers.30.self_attn.q_proj",
337 "model.language_model.layers.30.self_attn.k_proj",
338 "model.language_model.layers.30.self_attn.v_proj",
339 "model.language_model.layers.30.self_attn.o_proj",
340 "model.language_model.layers.30.mlp.gate_proj",
341 "model.language_model.layers.30.mlp.up_proj",
342 "model.language_model.layers.30.mlp.down_proj",
343 "model.language_model.layers.31.self_attn.q_proj",
344 "model.language_model.layers.31.self_attn.k_proj",
345 "model.language_model.layers.31.self_attn.v_proj",
346 "model.language_model.layers.31.self_attn.o_proj",
347 "model.language_model.layers.31.mlp.gate_proj",
348 "model.language_model.layers.31.mlp.up_proj",
349 "model.language_model.layers.31.mlp.down_proj",
350 "model.language_model.layers.32.self_attn.q_proj",
351 "model.language_model.layers.32.self_attn.k_proj",
352 "model.language_model.layers.32.self_attn.v_proj",
353 "model.language_model.layers.32.self_attn.o_proj",
354 "model.language_model.layers.32.mlp.gate_proj",
355 "model.language_model.layers.32.mlp.up_proj",
356 "model.language_model.layers.32.mlp.down_proj",
357 "model.language_model.layers.33.self_attn.q_proj",
358 "model.language_model.layers.33.self_attn.k_proj",
359 "model.language_model.layers.33.self_attn.v_proj",
360 "model.language_model.layers.33.self_attn.o_proj",
361 "model.language_model.layers.33.mlp.gate_proj",
362 "model.language_model.layers.33.mlp.up_proj",
363 "model.language_model.layers.33.mlp.down_proj",
364 "model.language_model.layers.34.self_attn.q_proj",
365 "model.language_model.layers.34.self_attn.k_proj",
366 "model.language_model.layers.34.self_attn.v_proj",
367 "model.language_model.layers.34.self_attn.o_proj",
368 "model.language_model.layers.34.mlp.gate_proj",
369 "model.language_model.layers.34.mlp.up_proj",
370 "model.language_model.layers.34.mlp.down_proj",
371 "model.language_model.layers.35.self_attn.q_proj",
372 "model.language_model.layers.35.self_attn.k_proj",
373 "model.language_model.layers.35.self_attn.v_proj",
374 "model.language_model.layers.35.self_attn.o_proj",
375 "model.language_model.layers.35.mlp.gate_proj",
376 "model.language_model.layers.35.mlp.up_proj",
377 "model.language_model.layers.35.mlp.down_proj"
378 ],
379 "parameter_dtypes": {
380 "torch.bfloat16": 1249853680,
381 "torch.int8": 7517270016
382 },
383 "int8_parameters": 7517270016
384}