Hugging Face's logo

RedHatAI
/
gemma-3-4b-it-quantized.w4a16

Image-Text-to-Text
Transformers
Safetensors
gemma3
vllm
vision
w4a16
conversational
text-generation-inference
compressed-tensors
Model card Files
xet
Community
1
Shubhra Pandit commited on
Commit
c537037
·
1 Parent(s): 6343593

Upload model files

Browse files
Files changed (14) hide show
  1. .gitattributes +2 -0
  2. chat_template.json +3 -0
  3. config.json +254 -0
  4. generation_config.json +13 -0
  5. model-00001-of-00002.safetensors +3 -0
  6. model-00002-of-00002.safetensors +3 -0
  7. model.safetensors +3 -0
  8. model.safetensors.index.json +0 -0
  9. preprocessor_config.json +29 -0
  10. processor_config.json +4 -0
  11. recipe.yaml +11 -0
  12. special_tokens_map.json +33 -0
  13. tokenizer.json +3 -0
  14. tokenizer_config.json +0 -0
.gitattributes CHANGED
@@ -33,3 +33,5 @@ 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
37
+
chat_template.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "chat_template": "{{ bos_token }}\n{%- if messages[0]['role'] == 'system' -%}\n {%- if messages[0]['content'] is string -%}\n {%- set first_user_prefix = messages[0]['content'] + '\n\n' -%}\n {%- else -%}\n {%- set first_user_prefix = messages[0]['content'][0]['text'] + '\n\n' -%}\n {%- endif -%}\n {%- set loop_messages = messages[1:] -%}\n{%- else -%}\n {%- set first_user_prefix = \"\" -%}\n {%- set loop_messages = messages -%}\n{%- endif -%}\n{%- for message in loop_messages -%}\n {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}\n {{ raise_exception(\"Conversation roles must alternate user/assistant/user/assistant/...\") }}\n {%- endif -%}\n {%- if (message['role'] == 'assistant') -%}\n {%- set role = \"model\" -%}\n {%- else -%}\n {%- set role = message['role'] -%}\n {%- endif -%}\n {{ '<start_of_turn>' + role + '\n' + (first_user_prefix if loop.first else \"\") }}\n {%- if message['content'] is string -%}\n {{ message['content'] | trim }}\n {%- elif message['content'] is iterable -%}\n {%- for item in message['content'] -%}\n {%- if item['type'] == 'image' -%}\n {{ '<start_of_image>' }}\n {%- elif item['type'] == 'text' -%}\n {{ item['text'] | trim }}\n {%- endif -%}\n {%- endfor -%}\n {%- else -%}\n {{ raise_exception(\"Invalid content type\") }}\n {%- endif -%}\n {{ '<end_of_turn>\n' }}\n{%- endfor -%}\n{%- if add_generation_prompt -%}\n {{'<start_of_turn>model\n'}}\n{%- endif -%}\n"
3
+ }
config.json ADDED
@@ -0,0 +1,254 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "eoi_token_index": 256000,
7
+ "eos_token_id": [
8
+ 1,
9
+ 106
10
+ ],
11
+ "image_token_index": 262144,
12
+ "initializer_range": 0.02,
13
+ "mm_tokens_per_image": 256,
14
+ "model_type": "gemma3",
15
+ "quantization_config": {
16
+ "config_groups": {
17
+ "group_0": {
18
+ "input_activations": null,
19
+ "output_activations": null,
20
+ "targets": [
21
+ "Linear"
22
+ ],
23
+ "weights": {
24
+ "actorder": "weight",
25
+ "block_structure": null,
26
+ "dynamic": false,
27
+ "group_size": 128,
28
+ "num_bits": 4,
29
+ "observer": "mse",
30
+ "observer_kwargs": {},
31
+ "strategy": "group",
32
+ "symmetric": true,
33
+ "type": "int"
34
+ }
35
+ }
36
+ },
37
+ "format": "pack-quantized",
38
+ "global_compression_ratio": null,
39
+ "ignore": [
40
+ "vision_tower.vision_model.encoder.layers.0.self_attn.k_proj",
41
+ "vision_tower.vision_model.encoder.layers.0.self_attn.v_proj",
42
+ "vision_tower.vision_model.encoder.layers.0.self_attn.q_proj",
43
+ "vision_tower.vision_model.encoder.layers.0.self_attn.out_proj",
44
+ "vision_tower.vision_model.encoder.layers.0.mlp.fc1",
45
+ "vision_tower.vision_model.encoder.layers.0.mlp.fc2",
46
+ "vision_tower.vision_model.encoder.layers.1.self_attn.k_proj",
47
+ "vision_tower.vision_model.encoder.layers.1.self_attn.v_proj",
48
+ "vision_tower.vision_model.encoder.layers.1.self_attn.q_proj",
49
+ "vision_tower.vision_model.encoder.layers.1.self_attn.out_proj",
50
+ "vision_tower.vision_model.encoder.layers.1.mlp.fc1",
51
+ "vision_tower.vision_model.encoder.layers.1.mlp.fc2",
52
+ "vision_tower.vision_model.encoder.layers.2.self_attn.k_proj",
53
+ "vision_tower.vision_model.encoder.layers.2.self_attn.v_proj",
54
+ "vision_tower.vision_model.encoder.layers.2.self_attn.q_proj",
55
+ "vision_tower.vision_model.encoder.layers.2.self_attn.out_proj",
56
+ "vision_tower.vision_model.encoder.layers.2.mlp.fc1",
57
+ "vision_tower.vision_model.encoder.layers.2.mlp.fc2",
58
+ "vision_tower.vision_model.encoder.layers.3.self_attn.k_proj",
59
+ "vision_tower.vision_model.encoder.layers.3.self_attn.v_proj",
60
+ "vision_tower.vision_model.encoder.layers.3.self_attn.q_proj",
61
+ "vision_tower.vision_model.encoder.layers.3.self_attn.out_proj",
62
+ "vision_tower.vision_model.encoder.layers.3.mlp.fc1",
63
+ "vision_tower.vision_model.encoder.layers.3.mlp.fc2",
64
+ "vision_tower.vision_model.encoder.layers.4.self_attn.k_proj",
65
+ "vision_tower.vision_model.encoder.layers.4.self_attn.v_proj",
66
+ "vision_tower.vision_model.encoder.layers.4.self_attn.q_proj",
67
+ "vision_tower.vision_model.encoder.layers.4.self_attn.out_proj",
68
+ "vision_tower.vision_model.encoder.layers.4.mlp.fc1",
69
+ "vision_tower.vision_model.encoder.layers.4.mlp.fc2",
70
+ "vision_tower.vision_model.encoder.layers.5.self_attn.k_proj",
71
+ "vision_tower.vision_model.encoder.layers.5.self_attn.v_proj",
72
+ "vision_tower.vision_model.encoder.layers.5.self_attn.q_proj",
73
+ "vision_tower.vision_model.encoder.layers.5.self_attn.out_proj",
74
+ "vision_tower.vision_model.encoder.layers.5.mlp.fc1",
75
+ "vision_tower.vision_model.encoder.layers.5.mlp.fc2",
76
+ "vision_tower.vision_model.encoder.layers.6.self_attn.k_proj",
77
+ "vision_tower.vision_model.encoder.layers.6.self_attn.v_proj",
78
+ "vision_tower.vision_model.encoder.layers.6.self_attn.q_proj",
79
+ "vision_tower.vision_model.encoder.layers.6.self_attn.out_proj",
80
+ "vision_tower.vision_model.encoder.layers.6.mlp.fc1",
81
+ "vision_tower.vision_model.encoder.layers.6.mlp.fc2",
82
+ "vision_tower.vision_model.encoder.layers.7.self_attn.k_proj",
83
+ "vision_tower.vision_model.encoder.layers.7.self_attn.v_proj",
84
+ "vision_tower.vision_model.encoder.layers.7.self_attn.q_proj",
85
+ "vision_tower.vision_model.encoder.layers.7.self_attn.out_proj",
86
+ "vision_tower.vision_model.encoder.layers.7.mlp.fc1",
87
+ "vision_tower.vision_model.encoder.layers.7.mlp.fc2",
88
+ "vision_tower.vision_model.encoder.layers.8.self_attn.k_proj",
89
+ "vision_tower.vision_model.encoder.layers.8.self_attn.v_proj",
90
+ "vision_tower.vision_model.encoder.layers.8.self_attn.q_proj",
91
+ "vision_tower.vision_model.encoder.layers.8.self_attn.out_proj",
92
+ "vision_tower.vision_model.encoder.layers.8.mlp.fc1",
93
+ "vision_tower.vision_model.encoder.layers.8.mlp.fc2",
94
+ "vision_tower.vision_model.encoder.layers.9.self_attn.k_proj",
95
+ "vision_tower.vision_model.encoder.layers.9.self_attn.v_proj",
96
+ "vision_tower.vision_model.encoder.layers.9.self_attn.q_proj",
97
+ "vision_tower.vision_model.encoder.layers.9.self_attn.out_proj",
98
+ "vision_tower.vision_model.encoder.layers.9.mlp.fc1",
99
+ "vision_tower.vision_model.encoder.layers.9.mlp.fc2",
100
+ "vision_tower.vision_model.encoder.layers.10.self_attn.k_proj",
101
+ "vision_tower.vision_model.encoder.layers.10.self_attn.v_proj",
102
+ "vision_tower.vision_model.encoder.layers.10.self_attn.q_proj",
103
+ "vision_tower.vision_model.encoder.layers.10.self_attn.out_proj",
104
+ "vision_tower.vision_model.encoder.layers.10.mlp.fc1",
105
+ "vision_tower.vision_model.encoder.layers.10.mlp.fc2",
106
+ "vision_tower.vision_model.encoder.layers.11.self_attn.k_proj",
107
+ "vision_tower.vision_model.encoder.layers.11.self_attn.v_proj",
108
+ "vision_tower.vision_model.encoder.layers.11.self_attn.q_proj",
109
+ "vision_tower.vision_model.encoder.layers.11.self_attn.out_proj",
110
+ "vision_tower.vision_model.encoder.layers.11.mlp.fc1",
111
+ "vision_tower.vision_model.encoder.layers.11.mlp.fc2",
112
+ "vision_tower.vision_model.encoder.layers.12.self_attn.k_proj",
113
+ "vision_tower.vision_model.encoder.layers.12.self_attn.v_proj",
114
+ "vision_tower.vision_model.encoder.layers.12.self_attn.q_proj",
115
+ "vision_tower.vision_model.encoder.layers.12.self_attn.out_proj",
116
+ "vision_tower.vision_model.encoder.layers.12.mlp.fc1",
117
+ "vision_tower.vision_model.encoder.layers.12.mlp.fc2",
118
+ "vision_tower.vision_model.encoder.layers.13.self_attn.k_proj",
119
+ "vision_tower.vision_model.encoder.layers.13.self_attn.v_proj",
120
+ "vision_tower.vision_model.encoder.layers.13.self_attn.q_proj",
121
+ "vision_tower.vision_model.encoder.layers.13.self_attn.out_proj",
122
+ "vision_tower.vision_model.encoder.layers.13.mlp.fc1",
123
+ "vision_tower.vision_model.encoder.layers.13.mlp.fc2",
124
+ "vision_tower.vision_model.encoder.layers.14.self_attn.k_proj",
125
+ "vision_tower.vision_model.encoder.layers.14.self_attn.v_proj",
126
+ "vision_tower.vision_model.encoder.layers.14.self_attn.q_proj",
127
+ "vision_tower.vision_model.encoder.layers.14.self_attn.out_proj",
128
+ "vision_tower.vision_model.encoder.layers.14.mlp.fc1",
129
+ "vision_tower.vision_model.encoder.layers.14.mlp.fc2",
130
+ "vision_tower.vision_model.encoder.layers.15.self_attn.k_proj",
131
+ "vision_tower.vision_model.encoder.layers.15.self_attn.v_proj",
132
+ "vision_tower.vision_model.encoder.layers.15.self_attn.q_proj",
133
+ "vision_tower.vision_model.encoder.layers.15.self_attn.out_proj",
134
+ "vision_tower.vision_model.encoder.layers.15.mlp.fc1",
135
+ "vision_tower.vision_model.encoder.layers.15.mlp.fc2",
136
+ "vision_tower.vision_model.encoder.layers.16.self_attn.k_proj",
137
+ "vision_tower.vision_model.encoder.layers.16.self_attn.v_proj",
138
+ "vision_tower.vision_model.encoder.layers.16.self_attn.q_proj",
139
+ "vision_tower.vision_model.encoder.layers.16.self_attn.out_proj",
140
+ "vision_tower.vision_model.encoder.layers.16.mlp.fc1",
141
+ "vision_tower.vision_model.encoder.layers.16.mlp.fc2",
142
+ "vision_tower.vision_model.encoder.layers.17.self_attn.k_proj",
143
+ "vision_tower.vision_model.encoder.layers.17.self_attn.v_proj",
144
+ "vision_tower.vision_model.encoder.layers.17.self_attn.q_proj",
145
+ "vision_tower.vision_model.encoder.layers.17.self_attn.out_proj",
146
+ "vision_tower.vision_model.encoder.layers.17.mlp.fc1",
147
+ "vision_tower.vision_model.encoder.layers.17.mlp.fc2",
148
+ "vision_tower.vision_model.encoder.layers.18.self_attn.k_proj",
149
+ "vision_tower.vision_model.encoder.layers.18.self_attn.v_proj",
150
+ "vision_tower.vision_model.encoder.layers.18.self_attn.q_proj",
151
+ "vision_tower.vision_model.encoder.layers.18.self_attn.out_proj",
152
+ "vision_tower.vision_model.encoder.layers.18.mlp.fc1",
153
+ "vision_tower.vision_model.encoder.layers.18.mlp.fc2",
154
+ "vision_tower.vision_model.encoder.layers.19.self_attn.k_proj",
155
+ "vision_tower.vision_model.encoder.layers.19.self_attn.v_proj",
156
+ "vision_tower.vision_model.encoder.layers.19.self_attn.q_proj",
157
+ "vision_tower.vision_model.encoder.layers.19.self_attn.out_proj",
158
+ "vision_tower.vision_model.encoder.layers.19.mlp.fc1",
159
+ "vision_tower.vision_model.encoder.layers.19.mlp.fc2",
160
+ "vision_tower.vision_model.encoder.layers.20.self_attn.k_proj",
161
+ "vision_tower.vision_model.encoder.layers.20.self_attn.v_proj",
162
+ "vision_tower.vision_model.encoder.layers.20.self_attn.q_proj",
163
+ "vision_tower.vision_model.encoder.layers.20.self_attn.out_proj",
164
+ "vision_tower.vision_model.encoder.layers.20.mlp.fc1",
165
+ "vision_tower.vision_model.encoder.layers.20.mlp.fc2",
166
+ "vision_tower.vision_model.encoder.layers.21.self_attn.k_proj",
167
+ "vision_tower.vision_model.encoder.layers.21.self_attn.v_proj",
168
+ "vision_tower.vision_model.encoder.layers.21.self_attn.q_proj",
169
+ "vision_tower.vision_model.encoder.layers.21.self_attn.out_proj",
170
+ "vision_tower.vision_model.encoder.layers.21.mlp.fc1",
171
+ "vision_tower.vision_model.encoder.layers.21.mlp.fc2",
172
+ "vision_tower.vision_model.encoder.layers.22.self_attn.k_proj",
173
+ "vision_tower.vision_model.encoder.layers.22.self_attn.v_proj",
174
+ "vision_tower.vision_model.encoder.layers.22.self_attn.q_proj",
175
+ "vision_tower.vision_model.encoder.layers.22.self_attn.out_proj",
176
+ "vision_tower.vision_model.encoder.layers.22.mlp.fc1",
177
+ "vision_tower.vision_model.encoder.layers.22.mlp.fc2",
178
+ "vision_tower.vision_model.encoder.layers.23.self_attn.k_proj",
179
+ "vision_tower.vision_model.encoder.layers.23.self_attn.v_proj",
180
+ "vision_tower.vision_model.encoder.layers.23.self_attn.q_proj",
181
+ "vision_tower.vision_model.encoder.layers.23.self_attn.out_proj",
182
+ "vision_tower.vision_model.encoder.layers.23.mlp.fc1",
183
+ "vision_tower.vision_model.encoder.layers.23.mlp.fc2",
184
+ "vision_tower.vision_model.encoder.layers.24.self_attn.k_proj",
185
+ "vision_tower.vision_model.encoder.layers.24.self_attn.v_proj",
186
+ "vision_tower.vision_model.encoder.layers.24.self_attn.q_proj",
187
+ "vision_tower.vision_model.encoder.layers.24.self_attn.out_proj",
188
+ "vision_tower.vision_model.encoder.layers.24.mlp.fc1",
189
+ "vision_tower.vision_model.encoder.layers.24.mlp.fc2",
190
+ "vision_tower.vision_model.encoder.layers.25.self_attn.k_proj",
191
+ "vision_tower.vision_model.encoder.layers.25.self_attn.v_proj",
192
+ "vision_tower.vision_model.encoder.layers.25.self_attn.q_proj",
193
+ "vision_tower.vision_model.encoder.layers.25.self_attn.out_proj",
194
+ "vision_tower.vision_model.encoder.layers.25.mlp.fc1",
195
+ "vision_tower.vision_model.encoder.layers.25.mlp.fc2",
196
+ "vision_tower.vision_model.encoder.layers.26.self_attn.k_proj",
197
+ "vision_tower.vision_model.encoder.layers.26.self_attn.v_proj",
198
+ "vision_tower.vision_model.encoder.layers.26.self_attn.q_proj",
199
+ "vision_tower.vision_model.encoder.layers.26.self_attn.out_proj",
200
+ "vision_tower.vision_model.encoder.layers.26.mlp.fc1",
201
+ "vision_tower.vision_model.encoder.layers.26.mlp.fc2",
202
+ "language_model.lm_head"
203
+ ],
204
+ "kv_cache_scheme": null,
205
+ "quant_method": "compressed-tensors",
206
+ "quantization_status": "compressed"
207
+ },
208
+ "text_config": {
209
+ "_attn_implementation_autoset": true,
210
+ "attention_bias": false,
211
+ "attention_dropout": 0.0,
212
+ "attn_logit_softcapping": null,
213
+ "cache_implementation": "hybrid",
214
+ "final_logit_softcapping": null,
215
+ "head_dim": 256,
216
+ "hidden_activation": "gelu_pytorch_tanh",
217
+ "hidden_size": 2560,
218
+ "initializer_range": 0.02,
219
+ "intermediate_size": 10240,
220
+ "max_position_embeddings": 131072,
221
+ "model_type": "gemma3_text",
222
+ "num_attention_heads": 8,
223
+ "num_hidden_layers": 34,
224
+ "num_key_value_heads": 4,
225
+ "query_pre_attn_scalar": 256,
226
+ "rms_norm_eps": 1e-06,
227
+ "rope_local_base_freq": 10000.0,
228
+ "rope_scaling": {
229
+ "factor": 8.0,
230
+ "rope_type": "linear"
231
+ },
232
+ "rope_theta": 1000000.0,
233
+ "sliding_window": 1024,
234
+ "sliding_window_pattern": 6,
235
+ "use_cache": true,
236
+ "vocab_size": 262208
237
+ },
238
+ "torch_dtype": "bfloat16",
239
+ "transformers_version": "4.51.3",
240
+ "vision_config": {
241
+ "attention_dropout": 0.0,
242
+ "hidden_act": "gelu_pytorch_tanh",
243
+ "hidden_size": 1152,
244
+ "image_size": 896,
245
+ "intermediate_size": 4304,
246
+ "layer_norm_eps": 1e-06,
247
+ "model_type": "siglip_vision_model",
248
+ "num_attention_heads": 16,
249
+ "num_channels": 3,
250
+ "num_hidden_layers": 27,
251
+ "patch_size": 14,
252
+ "vision_use_head": false
253
+ }
254
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "cache_implementation": "hybrid",
4
+ "do_sample": true,
5
+ "eos_token_id": [
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "top_k": 64,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.51.3"
13
+ }
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:90d98c8c878c0cfd9e7fe6504e81f5c9f10b777c588dd02d86e36bad06e1073e
3
+ size 3837513624
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb73be0b65b63a6587425d2c2e035dd61057651ecf69e74bc73a1902df340010
3
+ size 1342505104
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e972fa5445e79d96d53663af92a43a3efda3bb470109e18a4225c3289e188363
3
+ size 4183628640
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
preprocessor_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": null,
3
+ "do_normalize": true,
4
+ "do_pan_and_scan": null,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Gemma3ImageProcessor",
13
+ "image_seq_length": 256,
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "pan_and_scan_max_num_crops": null,
20
+ "pan_and_scan_min_crop_size": null,
21
+ "pan_and_scan_min_ratio_to_activate": null,
22
+ "processor_class": "Gemma3Processor",
23
+ "resample": 2,
24
+ "rescale_factor": 0.00392156862745098,
25
+ "size": {
26
+ "height": 896,
27
+ "width": 896
28
+ }
29
+ }
processor_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "image_seq_length": 256,
3
+ "processor_class": "Gemma3Processor"
4
+ }
recipe.yaml ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ quant_stage:
2
+ quant_modifiers:
3
+ GPTQModifier:
4
+ ignore: ['re:.*lm_head.*', 're:.*embed_tokens.*', 're:vision_tower.*', 're:multi_modal_projector.*']
5
+ sequential_targets: [Gemma3DecoderLayer]
6
+ dampening_frac: 0.05
7
+ config_groups:
8
+ group0:
9
+ targets: [Linear]
10
+ weights: {num_bits: 4, type: int, strategy: group, group_size: 128, symmetric: true,
11
+ actorder: weight, observer: mse}
special_tokens_map.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "boi_token": "<start_of_image>",
3
+ "bos_token": {
4
+ "content": "<bos>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "eoi_token": "<end_of_image>",
11
+ "eos_token": {
12
+ "content": "<eos>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false
17
+ },
18
+ "image_token": "<image_soft_token>",
19
+ "pad_token": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "unk_token": {
27
+ "content": "<unk>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ }
33
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4667f2089529e8e7657cfb6d1c19910ae71ff5f28aa7ab2ff2763330affad795
3
+ size 33384568
tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff