Hugging Face's logo

nm-testing
/
pixtral-12b-w4a16-actorder-group

Safetensors
llava
compressed-tensors
Model card Files Community
1
Shubhra Pandit commited on
Commit
a76f261
·
1 Parent(s): fcf1a17

Track tokenizer.json with Git LFS

Browse files
Files changed (13) hide show
  1. .gitattributes +1 -0
  2. chat_template.json +3 -0
  3. config.json +247 -0
  4. generation_config.json +6 -0
  5. model-00001-of-00002.safetensors +3 -0
  6. model-00002-of-00002.safetensors +3 -0
  7. model.safetensors.index.json +0 -0
  8. preprocessor_config.json +27 -0
  9. processor_config.json +7 -0
  10. recipe.yaml +11 -0
  11. special_tokens_map.json +24 -0
  12. tokenizer.json +3 -0
  13. tokenizer_config.json +0 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
chat_template.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "chat_template": "{%- if messages[0][\"role\"] == \"system\" %}\n {%- set system_message = messages[0][\"content\"] %}\n {%- set loop_messages = messages[1:] %}\n{%- else %}\n {%- set loop_messages = messages %}\n{%- endif %}\n\n{{- bos_token }}\n{%- for message in loop_messages %}\n {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}\n {{- raise_exception('After the optional system message, conversation roles must alternate user/assistant/user/assistant/...') }}\n {%- endif %}\n {%- if message[\"role\"] == \"user\" %}\n {%- if loop.last and system_message is defined %}\n {{- \"[INST]\" + system_message + \"\n\n\" }}\n {%- else %}\n {{- \"[INST]\" }}\n {%- endif %}\n {%- if message[\"content\"] is not string %}\n {%- for chunk in message[\"content\"] %}\n {%- if chunk[\"type\"] == \"text\" %}\n {{- chunk[\"content\"] }}\n {%- elif chunk[\"type\"] == \"image\" %}\n {{- \"[IMG]\" }}\n {%- else %}\n {{- raise_exception(\"Unrecognized content type!\") }}\n {%- endif %}\n {%- endfor %}\n {%- else %}\n {{- message[\"content\"] }}\n {%- endif %}\n {{- \"[/INST]\" }}\n {%- elif message[\"role\"] == \"assistant\" %}\n {{- message[\"content\"] + eos_token}}\n {%- else %}\n {{- raise_exception(\"Only user and assistant roles are supported, with the exception of an initial optional system message!\") }}\n {%- endif %}\n{%- endfor %}"
3
+ }
config.json ADDED
@@ -0,0 +1,247 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "mgoin/pixtral-12b",
3
+ "architectures": [
4
+ "LlavaForConditionalGeneration"
5
+ ],
6
+ "ignore_index": -100,
7
+ "image_seq_length": 1,
8
+ "image_token_index": 10,
9
+ "model_type": "llava",
10
+ "projector_hidden_act": "gelu",
11
+ "quantization_config": {
12
+ "config_groups": {
13
+ "group_0": {
14
+ "input_activations": null,
15
+ "output_activations": null,
16
+ "targets": [
17
+ "Linear"
18
+ ],
19
+ "weights": {
20
+ "actorder": "group",
21
+ "block_structure": null,
22
+ "dynamic": false,
23
+ "group_size": 128,
24
+ "num_bits": 4,
25
+ "observer": "minmax",
26
+ "observer_kwargs": {},
27
+ "strategy": "group",
28
+ "symmetric": true,
29
+ "type": "int"
30
+ }
31
+ }
32
+ },
33
+ "format": "pack-quantized",
34
+ "global_compression_ratio": 1.4377437657736856,
35
+ "ignore": [
36
+ "vision_tower.transformer.layers.0.feed_forward.gate_proj",
37
+ "vision_tower.transformer.layers.0.feed_forward.up_proj",
38
+ "vision_tower.transformer.layers.0.feed_forward.down_proj",
39
+ "vision_tower.transformer.layers.0.attention.k_proj",
40
+ "vision_tower.transformer.layers.0.attention.v_proj",
41
+ "vision_tower.transformer.layers.0.attention.q_proj",
42
+ "vision_tower.transformer.layers.0.attention.o_proj",
43
+ "vision_tower.transformer.layers.1.feed_forward.gate_proj",
44
+ "vision_tower.transformer.layers.1.feed_forward.up_proj",
45
+ "vision_tower.transformer.layers.1.feed_forward.down_proj",
46
+ "vision_tower.transformer.layers.1.attention.k_proj",
47
+ "vision_tower.transformer.layers.1.attention.v_proj",
48
+ "vision_tower.transformer.layers.1.attention.q_proj",
49
+ "vision_tower.transformer.layers.1.attention.o_proj",
50
+ "vision_tower.transformer.layers.2.feed_forward.gate_proj",
51
+ "vision_tower.transformer.layers.2.feed_forward.up_proj",
52
+ "vision_tower.transformer.layers.2.feed_forward.down_proj",
53
+ "vision_tower.transformer.layers.2.attention.k_proj",
54
+ "vision_tower.transformer.layers.2.attention.v_proj",
55
+ "vision_tower.transformer.layers.2.attention.q_proj",
56
+ "vision_tower.transformer.layers.2.attention.o_proj",
57
+ "vision_tower.transformer.layers.3.feed_forward.gate_proj",
58
+ "vision_tower.transformer.layers.3.feed_forward.up_proj",
59
+ "vision_tower.transformer.layers.3.feed_forward.down_proj",
60
+ "vision_tower.transformer.layers.3.attention.k_proj",
61
+ "vision_tower.transformer.layers.3.attention.v_proj",
62
+ "vision_tower.transformer.layers.3.attention.q_proj",
63
+ "vision_tower.transformer.layers.3.attention.o_proj",
64
+ "vision_tower.transformer.layers.4.feed_forward.gate_proj",
65
+ "vision_tower.transformer.layers.4.feed_forward.up_proj",
66
+ "vision_tower.transformer.layers.4.feed_forward.down_proj",
67
+ "vision_tower.transformer.layers.4.attention.k_proj",
68
+ "vision_tower.transformer.layers.4.attention.v_proj",
69
+ "vision_tower.transformer.layers.4.attention.q_proj",
70
+ "vision_tower.transformer.layers.4.attention.o_proj",
71
+ "vision_tower.transformer.layers.5.feed_forward.gate_proj",
72
+ "vision_tower.transformer.layers.5.feed_forward.up_proj",
73
+ "vision_tower.transformer.layers.5.feed_forward.down_proj",
74
+ "vision_tower.transformer.layers.5.attention.k_proj",
75
+ "vision_tower.transformer.layers.5.attention.v_proj",
76
+ "vision_tower.transformer.layers.5.attention.q_proj",
77
+ "vision_tower.transformer.layers.5.attention.o_proj",
78
+ "vision_tower.transformer.layers.6.feed_forward.gate_proj",
79
+ "vision_tower.transformer.layers.6.feed_forward.up_proj",
80
+ "vision_tower.transformer.layers.6.feed_forward.down_proj",
81
+ "vision_tower.transformer.layers.6.attention.k_proj",
82
+ "vision_tower.transformer.layers.6.attention.v_proj",
83
+ "vision_tower.transformer.layers.6.attention.q_proj",
84
+ "vision_tower.transformer.layers.6.attention.o_proj",
85
+ "vision_tower.transformer.layers.7.feed_forward.gate_proj",
86
+ "vision_tower.transformer.layers.7.feed_forward.up_proj",
87
+ "vision_tower.transformer.layers.7.feed_forward.down_proj",
88
+ "vision_tower.transformer.layers.7.attention.k_proj",
89
+ "vision_tower.transformer.layers.7.attention.v_proj",
90
+ "vision_tower.transformer.layers.7.attention.q_proj",
91
+ "vision_tower.transformer.layers.7.attention.o_proj",
92
+ "vision_tower.transformer.layers.8.feed_forward.gate_proj",
93
+ "vision_tower.transformer.layers.8.feed_forward.up_proj",
94
+ "vision_tower.transformer.layers.8.feed_forward.down_proj",
95
+ "vision_tower.transformer.layers.8.attention.k_proj",
96
+ "vision_tower.transformer.layers.8.attention.v_proj",
97
+ "vision_tower.transformer.layers.8.attention.q_proj",
98
+ "vision_tower.transformer.layers.8.attention.o_proj",
99
+ "vision_tower.transformer.layers.9.feed_forward.gate_proj",
100
+ "vision_tower.transformer.layers.9.feed_forward.up_proj",
101
+ "vision_tower.transformer.layers.9.feed_forward.down_proj",
102
+ "vision_tower.transformer.layers.9.attention.k_proj",
103
+ "vision_tower.transformer.layers.9.attention.v_proj",
104
+ "vision_tower.transformer.layers.9.attention.q_proj",
105
+ "vision_tower.transformer.layers.9.attention.o_proj",
106
+ "vision_tower.transformer.layers.10.feed_forward.gate_proj",
107
+ "vision_tower.transformer.layers.10.feed_forward.up_proj",
108
+ "vision_tower.transformer.layers.10.feed_forward.down_proj",
109
+ "vision_tower.transformer.layers.10.attention.k_proj",
110
+ "vision_tower.transformer.layers.10.attention.v_proj",
111
+ "vision_tower.transformer.layers.10.attention.q_proj",
112
+ "vision_tower.transformer.layers.10.attention.o_proj",
113
+ "vision_tower.transformer.layers.11.feed_forward.gate_proj",
114
+ "vision_tower.transformer.layers.11.feed_forward.up_proj",
115
+ "vision_tower.transformer.layers.11.feed_forward.down_proj",
116
+ "vision_tower.transformer.layers.11.attention.k_proj",
117
+ "vision_tower.transformer.layers.11.attention.v_proj",
118
+ "vision_tower.transformer.layers.11.attention.q_proj",
119
+ "vision_tower.transformer.layers.11.attention.o_proj",
120
+ "vision_tower.transformer.layers.12.feed_forward.gate_proj",
121
+ "vision_tower.transformer.layers.12.feed_forward.up_proj",
122
+ "vision_tower.transformer.layers.12.feed_forward.down_proj",
123
+ "vision_tower.transformer.layers.12.attention.k_proj",
124
+ "vision_tower.transformer.layers.12.attention.v_proj",
125
+ "vision_tower.transformer.layers.12.attention.q_proj",
126
+ "vision_tower.transformer.layers.12.attention.o_proj",
127
+ "vision_tower.transformer.layers.13.feed_forward.gate_proj",
128
+ "vision_tower.transformer.layers.13.feed_forward.up_proj",
129
+ "vision_tower.transformer.layers.13.feed_forward.down_proj",
130
+ "vision_tower.transformer.layers.13.attention.k_proj",
131
+ "vision_tower.transformer.layers.13.attention.v_proj",
132
+ "vision_tower.transformer.layers.13.attention.q_proj",
133
+ "vision_tower.transformer.layers.13.attention.o_proj",
134
+ "vision_tower.transformer.layers.14.feed_forward.gate_proj",
135
+ "vision_tower.transformer.layers.14.feed_forward.up_proj",
136
+ "vision_tower.transformer.layers.14.feed_forward.down_proj",
137
+ "vision_tower.transformer.layers.14.attention.k_proj",
138
+ "vision_tower.transformer.layers.14.attention.v_proj",
139
+ "vision_tower.transformer.layers.14.attention.q_proj",
140
+ "vision_tower.transformer.layers.14.attention.o_proj",
141
+ "vision_tower.transformer.layers.15.feed_forward.gate_proj",
142
+ "vision_tower.transformer.layers.15.feed_forward.up_proj",
143
+ "vision_tower.transformer.layers.15.feed_forward.down_proj",
144
+ "vision_tower.transformer.layers.15.attention.k_proj",
145
+ "vision_tower.transformer.layers.15.attention.v_proj",
146
+ "vision_tower.transformer.layers.15.attention.q_proj",
147
+ "vision_tower.transformer.layers.15.attention.o_proj",
148
+ "vision_tower.transformer.layers.16.feed_forward.gate_proj",
149
+ "vision_tower.transformer.layers.16.feed_forward.up_proj",
150
+ "vision_tower.transformer.layers.16.feed_forward.down_proj",
151
+ "vision_tower.transformer.layers.16.attention.k_proj",
152
+ "vision_tower.transformer.layers.16.attention.v_proj",
153
+ "vision_tower.transformer.layers.16.attention.q_proj",
154
+ "vision_tower.transformer.layers.16.attention.o_proj",
155
+ "vision_tower.transformer.layers.17.feed_forward.gate_proj",
156
+ "vision_tower.transformer.layers.17.feed_forward.up_proj",
157
+ "vision_tower.transformer.layers.17.feed_forward.down_proj",
158
+ "vision_tower.transformer.layers.17.attention.k_proj",
159
+ "vision_tower.transformer.layers.17.attention.v_proj",
160
+ "vision_tower.transformer.layers.17.attention.q_proj",
161
+ "vision_tower.transformer.layers.17.attention.o_proj",
162
+ "vision_tower.transformer.layers.18.feed_forward.gate_proj",
163
+ "vision_tower.transformer.layers.18.feed_forward.up_proj",
164
+ "vision_tower.transformer.layers.18.feed_forward.down_proj",
165
+ "vision_tower.transformer.layers.18.attention.k_proj",
166
+ "vision_tower.transformer.layers.18.attention.v_proj",
167
+ "vision_tower.transformer.layers.18.attention.q_proj",
168
+ "vision_tower.transformer.layers.18.attention.o_proj",
169
+ "vision_tower.transformer.layers.19.feed_forward.gate_proj",
170
+ "vision_tower.transformer.layers.19.feed_forward.up_proj",
171
+ "vision_tower.transformer.layers.19.feed_forward.down_proj",
172
+ "vision_tower.transformer.layers.19.attention.k_proj",
173
+ "vision_tower.transformer.layers.19.attention.v_proj",
174
+ "vision_tower.transformer.layers.19.attention.q_proj",
175
+ "vision_tower.transformer.layers.19.attention.o_proj",
176
+ "vision_tower.transformer.layers.20.feed_forward.gate_proj",
177
+ "vision_tower.transformer.layers.20.feed_forward.up_proj",
178
+ "vision_tower.transformer.layers.20.feed_forward.down_proj",
179
+ "vision_tower.transformer.layers.20.attention.k_proj",
180
+ "vision_tower.transformer.layers.20.attention.v_proj",
181
+ "vision_tower.transformer.layers.20.attention.q_proj",
182
+ "vision_tower.transformer.layers.20.attention.o_proj",
183
+ "vision_tower.transformer.layers.21.feed_forward.gate_proj",
184
+ "vision_tower.transformer.layers.21.feed_forward.up_proj",
185
+ "vision_tower.transformer.layers.21.feed_forward.down_proj",
186
+ "vision_tower.transformer.layers.21.attention.k_proj",
187
+ "vision_tower.transformer.layers.21.attention.v_proj",
188
+ "vision_tower.transformer.layers.21.attention.q_proj",
189
+ "vision_tower.transformer.layers.21.attention.o_proj",
190
+ "vision_tower.transformer.layers.22.feed_forward.gate_proj",
191
+ "vision_tower.transformer.layers.22.feed_forward.up_proj",
192
+ "vision_tower.transformer.layers.22.feed_forward.down_proj",
193
+ "vision_tower.transformer.layers.22.attention.k_proj",
194
+ "vision_tower.transformer.layers.22.attention.v_proj",
195
+ "vision_tower.transformer.layers.22.attention.q_proj",
196
+ "vision_tower.transformer.layers.22.attention.o_proj",
197
+ "vision_tower.transformer.layers.23.feed_forward.gate_proj",
198
+ "vision_tower.transformer.layers.23.feed_forward.up_proj",
199
+ "vision_tower.transformer.layers.23.feed_forward.down_proj",
200
+ "vision_tower.transformer.layers.23.attention.k_proj",
201
+ "vision_tower.transformer.layers.23.attention.v_proj",
202
+ "vision_tower.transformer.layers.23.attention.q_proj",
203
+ "vision_tower.transformer.layers.23.attention.o_proj",
204
+ "multi_modal_projector.linear_1",
205
+ "multi_modal_projector.linear_2",
206
+ "language_model.lm_head"
207
+ ],
208
+ "kv_cache_scheme": null,
209
+ "quant_method": "compressed-tensors",
210
+ "quantization_status": "compressed",
211
+ "sparsity_config": {
212
+ "format": "dense",
213
+ "global_sparsity": 0.13750025450228134,
214
+ "ignore": [],
215
+ "registry_requires_subclass": false,
216
+ "sparsity_structure": "unstructured",
217
+ "targets": []
218
+ }
219
+ },
220
+ "text_config": {
221
+ "hidden_size": 5120,
222
+ "intermediate_size": 14336,
223
+ "is_composition": true,
224
+ "max_position_embeddings": 1024000,
225
+ "model_type": "mistral",
226
+ "num_hidden_layers": 40,
227
+ "num_key_value_heads": 8,
228
+ "rms_norm_eps": 1e-05,
229
+ "rope_theta": 1000000000.0,
230
+ "sliding_window": null,
231
+ "vocab_size": 131072
232
+ },
233
+ "torch_dtype": "bfloat16",
234
+ "transformers_version": "4.47.0",
235
+ "vision_config": {
236
+ "head_dim": 64,
237
+ "hidden_act": "silu",
238
+ "image_size": 1024,
239
+ "is_composition": true,
240
+ "model_type": "pixtral",
241
+ "patch_size": 16,
242
+ "rope_theta": 10000.0,
243
+ "tie_word_embeddings": false
244
+ },
245
+ "vision_feature_layer": -1,
246
+ "vision_feature_select_strategy": "full"
247
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "transformers_version": "4.47.0"
6
+ }
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed8d76e289eba8b02d899f812d25260753c126e8c321695fd337dbd653ae2969
3
+ size 4989702904
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ae4db8eaa84c7601a6201fb7ac9399d966e59e8932b331fd2d06eeeb61b2363
3
+ size 4195619904
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
preprocessor_config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": true,
3
+ "do_normalize": true,
4
+ "do_rescale": true,
5
+ "do_resize": true,
6
+ "image_mean": [
7
+ 0.48145466,
8
+ 0.4578275,
9
+ 0.40821073
10
+ ],
11
+ "image_processor_type": "PixtralImageProcessorFast",
12
+ "image_std": [
13
+ 0.26862954,
14
+ 0.26130258,
15
+ 0.27577711
16
+ ],
17
+ "patch_size": {
18
+ "height": 16,
19
+ "width": 16
20
+ },
21
+ "processor_class": "PixtralProcessor",
22
+ "resample": 3,
23
+ "rescale_factor": 0.00392156862745098,
24
+ "size": {
25
+ "longest_edge": 1024
26
+ }
27
+ }
processor_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_break_token": "[IMG_BREAK]",
3
+ "image_end_token": "[IMG_END]",
4
+ "image_token": "[IMG]",
5
+ "patch_size": 16,
6
+ "processor_class": "PixtralProcessor"
7
+ }
recipe.yaml ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ DEFAULT_stage:
2
+ DEFAULT_modifiers:
3
+ GPTQModifier:
4
+ dampening_frac: 0.05
5
+ config_groups:
6
+ config_group:
7
+ targets: [Linear]
8
+ weights: {num_bits: 4, type: int, symmetric: true, group_size: 128, strategy: group,
9
+ dynamic: false, actorder: group, observer: minmax}
10
+ targets: Linear
11
+ ignore: ['re:.*lm_head', 're:multi_modal_projector.*', 're:vision_tower.*']
special_tokens_map.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": "</s>",
17
+ "unk_token": {
18
+ "content": "<unk>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ }
24
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fcdf3e6b96371b9a5c8673bdf2f1bc838b994b73a6fa995dca62196346aaddb5
3
+ size 17077311
tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff