ixim/Image21-INT8
162
1{
2 "component": "transformer",
3 "method": "bitsandbytes LLM.int8",
4 "threshold": 6.0,
5 "skip_modules": [
6 "proj_out",
7 "img_in",
8 "txt_in",
9 "time_text_embed",
10 "modulation",
11 "norm_out"
12 ],
13 "linear8bit_modules": [
14 "transformer_blocks.0.attn.to_q",
15 "transformer_blocks.0.attn.to_k",
16 "transformer_blocks.0.attn.to_v",
17 "transformer_blocks.0.attn.to_out.0",
18 "transformer_blocks.0.img_mlp.proj",
19 "transformer_blocks.0.img_mlp.out",
20 "transformer_blocks.0.img_mlp.gate_layer",
21 "transformer_blocks.1.attn.to_q",
22 "transformer_blocks.1.attn.to_k",
23 "transformer_blocks.1.attn.to_v",
24 "transformer_blocks.1.attn.to_out.0",
25 "transformer_blocks.1.img_mlp.proj",
26 "transformer_blocks.1.img_mlp.out",
27 "transformer_blocks.1.img_mlp.gate_layer",
28 "transformer_blocks.2.attn.to_q",
29 "transformer_blocks.2.attn.to_k",
30 "transformer_blocks.2.attn.to_v",
31 "transformer_blocks.2.attn.to_out.0",
32 "transformer_blocks.2.img_mlp.proj",
33 "transformer_blocks.2.img_mlp.out",
34 "transformer_blocks.2.img_mlp.gate_layer",
35 "transformer_blocks.3.attn.to_q",
36 "transformer_blocks.3.attn.to_k",
37 "transformer_blocks.3.attn.to_v",
38 "transformer_blocks.3.attn.to_out.0",
39 "transformer_blocks.3.img_mlp.proj",
40 "transformer_blocks.3.img_mlp.out",
41 "transformer_blocks.3.img_mlp.gate_layer",
42 "transformer_blocks.4.attn.to_q",
43 "transformer_blocks.4.attn.to_k",
44 "transformer_blocks.4.attn.to_v",
45 "transformer_blocks.4.attn.to_out.0",
46 "transformer_blocks.4.img_mlp.proj",
47 "transformer_blocks.4.img_mlp.out",
48 "transformer_blocks.4.img_mlp.gate_layer",
49 "transformer_blocks.5.attn.to_q",
50 "transformer_blocks.5.attn.to_k",
51 "transformer_blocks.5.attn.to_v",
52 "transformer_blocks.5.attn.to_out.0",
53 "transformer_blocks.5.img_mlp.proj",
54 "transformer_blocks.5.img_mlp.out",
55 "transformer_blocks.5.img_mlp.gate_layer",
56 "transformer_blocks.6.attn.to_q",
57 "transformer_blocks.6.attn.to_k",
58 "transformer_blocks.6.attn.to_v",
59 "transformer_blocks.6.attn.to_out.0",
60 "transformer_blocks.6.img_mlp.proj",
61 "transformer_blocks.6.img_mlp.out",
62 "transformer_blocks.6.img_mlp.gate_layer",
63 "transformer_blocks.7.attn.to_q",
64 "transformer_blocks.7.attn.to_k",
65 "transformer_blocks.7.attn.to_v",
66 "transformer_blocks.7.attn.to_out.0",
67 "transformer_blocks.7.img_mlp.proj",
68 "transformer_blocks.7.img_mlp.out",
69 "transformer_blocks.7.img_mlp.gate_layer",
70 "transformer_blocks.8.attn.to_q",
71 "transformer_blocks.8.attn.to_k",
72 "transformer_blocks.8.attn.to_v",
73 "transformer_blocks.8.attn.to_out.0",
74 "transformer_blocks.8.img_mlp.proj",
75 "transformer_blocks.8.img_mlp.out",
76 "transformer_blocks.8.img_mlp.gate_layer",
77 "transformer_blocks.9.attn.to_q",
78 "transformer_blocks.9.attn.to_k",
79 "transformer_blocks.9.attn.to_v",
80 "transformer_blocks.9.attn.to_out.0",
81 "transformer_blocks.9.img_mlp.proj",
82 "transformer_blocks.9.img_mlp.out",
83 "transformer_blocks.9.img_mlp.gate_layer",
84 "transformer_blocks.10.attn.to_q",
85 "transformer_blocks.10.attn.to_k",
86 "transformer_blocks.10.attn.to_v",
87 "transformer_blocks.10.attn.to_out.0",
88 "transformer_blocks.10.img_mlp.proj",
89 "transformer_blocks.10.img_mlp.out",
90 "transformer_blocks.10.img_mlp.gate_layer",
91 "transformer_blocks.11.attn.to_q",
92 "transformer_blocks.11.attn.to_k",
93 "transformer_blocks.11.attn.to_v",
94 "transformer_blocks.11.attn.to_out.0",
95 "transformer_blocks.11.img_mlp.proj",
96 "transformer_blocks.11.img_mlp.out",
97 "transformer_blocks.11.img_mlp.gate_layer",
98 "transformer_blocks.12.attn.to_q",
99 "transformer_blocks.12.attn.to_k",
100 "transformer_blocks.12.attn.to_v",
101 "transformer_blocks.12.attn.to_out.0",
102 "transformer_blocks.12.img_mlp.proj",
103 "transformer_blocks.12.img_mlp.out",
104 "transformer_blocks.12.img_mlp.gate_layer",
105 "transformer_blocks.13.attn.to_q",
106 "transformer_blocks.13.attn.to_k",
107 "transformer_blocks.13.attn.to_v",
108 "transformer_blocks.13.attn.to_out.0",
109 "transformer_blocks.13.img_mlp.proj",
110 "transformer_blocks.13.img_mlp.out",
111 "transformer_blocks.13.img_mlp.gate_layer",
112 "transformer_blocks.14.attn.to_q",
113 "transformer_blocks.14.attn.to_k",
114 "transformer_blocks.14.attn.to_v",
115 "transformer_blocks.14.attn.to_out.0",
116 "transformer_blocks.14.img_mlp.proj",
117 "transformer_blocks.14.img_mlp.out",
118 "transformer_blocks.14.img_mlp.gate_layer",
119 "transformer_blocks.15.attn.to_q",
120 "transformer_blocks.15.attn.to_k",
121 "transformer_blocks.15.attn.to_v",
122 "transformer_blocks.15.attn.to_out.0",
123 "transformer_blocks.15.img_mlp.proj",
124 "transformer_blocks.15.img_mlp.out",
125 "transformer_blocks.15.img_mlp.gate_layer",
126 "transformer_blocks.16.attn.to_q",
127 "transformer_blocks.16.attn.to_k",
128 "transformer_blocks.16.attn.to_v",
129 "transformer_blocks.16.attn.to_out.0",
130 "transformer_blocks.16.img_mlp.proj",
131 "transformer_blocks.16.img_mlp.out",
132 "transformer_blocks.16.img_mlp.gate_layer",
133 "transformer_blocks.17.attn.to_q",
134 "transformer_blocks.17.attn.to_k",
135 "transformer_blocks.17.attn.to_v",
136 "transformer_blocks.17.attn.to_out.0",
137 "transformer_blocks.17.img_mlp.proj",
138 "transformer_blocks.17.img_mlp.out",
139 "transformer_blocks.17.img_mlp.gate_layer",
140 "transformer_blocks.18.attn.to_q",
141 "transformer_blocks.18.attn.to_k",
142 "transformer_blocks.18.attn.to_v",
143 "transformer_blocks.18.attn.to_out.0",
144 "transformer_blocks.18.img_mlp.proj",
145 "transformer_blocks.18.img_mlp.out",
146 "transformer_blocks.18.img_mlp.gate_layer",
147 "transformer_blocks.19.attn.to_q",
148 "transformer_blocks.19.attn.to_k",
149 "transformer_blocks.19.attn.to_v",
150 "transformer_blocks.19.attn.to_out.0",
151 "transformer_blocks.19.img_mlp.proj",
152 "transformer_blocks.19.img_mlp.out",
153 "transformer_blocks.19.img_mlp.gate_layer",
154 "transformer_blocks.20.attn.to_q",
155 "transformer_blocks.20.attn.to_k",
156 "transformer_blocks.20.attn.to_v",
157 "transformer_blocks.20.attn.to_out.0",
158 "transformer_blocks.20.img_mlp.proj",
159 "transformer_blocks.20.img_mlp.out",
160 "transformer_blocks.20.img_mlp.gate_layer",
161 "transformer_blocks.21.attn.to_q",
162 "transformer_blocks.21.attn.to_k",
163 "transformer_blocks.21.attn.to_v",
164 "transformer_blocks.21.attn.to_out.0",
165 "transformer_blocks.21.img_mlp.proj",
166 "transformer_blocks.21.img_mlp.out",
167 "transformer_blocks.21.img_mlp.gate_layer",
168 "transformer_blocks.22.attn.to_q",
169 "transformer_blocks.22.attn.to_k",
170 "transformer_blocks.22.attn.to_v",
171 "transformer_blocks.22.attn.to_out.0",
172 "transformer_blocks.22.img_mlp.proj",
173 "transformer_blocks.22.img_mlp.out",
174 "transformer_blocks.22.img_mlp.gate_layer",
175 "transformer_blocks.23.attn.to_q",
176 "transformer_blocks.23.attn.to_k",
177 "transformer_blocks.23.attn.to_v",
178 "transformer_blocks.23.attn.to_out.0",
179 "transformer_blocks.23.img_mlp.proj",
180 "transformer_blocks.23.img_mlp.out",
181 "transformer_blocks.23.img_mlp.gate_layer",
182 "transformer_blocks.24.attn.to_q",
183 "transformer_blocks.24.attn.to_k",
184 "transformer_blocks.24.attn.to_v",
185 "transformer_blocks.24.attn.to_out.0",
186 "transformer_blocks.24.img_mlp.proj",
187 "transformer_blocks.24.img_mlp.out",
188 "transformer_blocks.24.img_mlp.gate_layer",
189 "transformer_blocks.25.attn.to_q",
190 "transformer_blocks.25.attn.to_k",
191 "transformer_blocks.25.attn.to_v",
192 "transformer_blocks.25.attn.to_out.0",
193 "transformer_blocks.25.img_mlp.proj",
194 "transformer_blocks.25.img_mlp.out",
195 "transformer_blocks.25.img_mlp.gate_layer",
196 "transformer_blocks.26.attn.to_q",
197 "transformer_blocks.26.attn.to_k",
198 "transformer_blocks.26.attn.to_v",
199 "transformer_blocks.26.attn.to_out.0",
200 "transformer_blocks.26.img_mlp.proj",
201 "transformer_blocks.26.img_mlp.out",
202 "transformer_blocks.26.img_mlp.gate_layer",
203 "transformer_blocks.27.attn.to_q",
204 "transformer_blocks.27.attn.to_k",
205 "transformer_blocks.27.attn.to_v",
206 "transformer_blocks.27.attn.to_out.0",
207 "transformer_blocks.27.img_mlp.proj",
208 "transformer_blocks.27.img_mlp.out",
209 "transformer_blocks.27.img_mlp.gate_layer",
210 "transformer_blocks.28.attn.to_q",
211 "transformer_blocks.28.attn.to_k",
212 "transformer_blocks.28.attn.to_v",
213 "transformer_blocks.28.attn.to_out.0",
214 "transformer_blocks.28.img_mlp.proj",
215 "transformer_blocks.28.img_mlp.out",
216 "transformer_blocks.28.img_mlp.gate_layer",
217 "transformer_blocks.29.attn.to_q",
218 "transformer_blocks.29.attn.to_k",
219 "transformer_blocks.29.attn.to_v",
220 "transformer_blocks.29.attn.to_out.0",
221 "transformer_blocks.29.img_mlp.proj",
222 "transformer_blocks.29.img_mlp.out",
223 "transformer_blocks.29.img_mlp.gate_layer",
224 "transformer_blocks.30.attn.to_q",
225 "transformer_blocks.30.attn.to_k",
226 "transformer_blocks.30.attn.to_v",
227 "transformer_blocks.30.attn.to_out.0",
228 "transformer_blocks.30.img_mlp.proj",
229 "transformer_blocks.30.img_mlp.out",
230 "transformer_blocks.30.img_mlp.gate_layer",
231 "transformer_blocks.31.attn.to_q",
232 "transformer_blocks.31.attn.to_k",
233 "transformer_blocks.31.attn.to_v",
234 "transformer_blocks.31.attn.to_out.0",
235 "transformer_blocks.31.img_mlp.proj",
236 "transformer_blocks.31.img_mlp.out",
237 "transformer_blocks.31.img_mlp.gate_layer"
238 ],
239 "parameter_dtypes": {
240 "torch.bfloat16": 135802880,
241 "torch.int8": 6979321856
242 },
243 "int8_parameters": 6979321856
244}