Hugging Face's logo

aisingapore
/
Gemma-SEA-LION-v4-27B-IT-FP8-Dynamic

Text Generation
Transformers
Safetensors
gemma3
image-text-to-text
conversational
text-generation-inference
compressed-tensors
Model card Files
xet
 Community
waiyiaisg commited on
Commit
5c6b5a0
·
verified ·
1 Parent(s): f4593fe

Update config.json

Browse files
Files changed (1) hide show
  1. config.json +289 -2
config.json CHANGED
@@ -12,29 +12,316 @@
12
  "initializer_range": 0.02,
13
  "mm_tokens_per_image": 256,
14
  "model_type": "gemma3",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
15
  "text_config": {
 
 
 
 
 
 
16
  "head_dim": 128,
 
17
  "hidden_size": 5376,
 
18
  "intermediate_size": 21504,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  "model_type": "gemma3_text",
20
  "num_attention_heads": 32,
21
  "num_hidden_layers": 62,
22
  "num_key_value_heads": 16,
23
  "query_pre_attn_scalar": 168,
 
 
24
  "rope_scaling": {
25
  "factor": 8.0,
26
  "rope_type": "linear"
27
  },
28
- "sliding_window": 1024
 
 
 
 
29
  },
30
  "torch_dtype": "bfloat16",
31
- "transformers_version": "4.50.0.dev0",
32
  "vision_config": {
 
 
33
  "hidden_size": 1152,
34
  "image_size": 896,
35
  "intermediate_size": 4304,
 
36
  "model_type": "siglip_vision_model",
37
  "num_attention_heads": 16,
 
38
  "num_hidden_layers": 27,
39
  "patch_size": 14,
40
  "vision_use_head": false
 
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": {
19
+ "actorder": null,
20
+ "block_structure": null,
21
+ "dynamic": true,
22
+ "group_size": null,
23
+ "num_bits": 8,
24
+ "observer": null,
25
+ "observer_kwargs": {},
26
+ "strategy": "token",
27
+ "symmetric": true,
28
+ "type": "float"
29
+ },
30
+ "output_activations": null,
31
+ "targets": [
32
+ "Linear"
33
+ ],
34
+ "weights": {
35
+ "actorder": null,
36
+ "block_structure": null,
37
+ "dynamic": false,
38
+ "group_size": null,
39
+ "num_bits": 8,
40
+ "observer": "minmax",
41
+ "observer_kwargs": {},
42
+ "strategy": "channel",
43
+ "symmetric": true,
44
+ "type": "float"
45
+ }
46
+ }
47
+ },
48
+ "format": "float-quantized",
49
+ "global_compression_ratio": null,
50
+ "ignore": [
51
+ "model.vision_tower.vision_model.encoder.layers.0.self_attn.k_proj",
52
+ "model.vision_tower.vision_model.encoder.layers.0.self_attn.v_proj",
53
+ "model.vision_tower.vision_model.encoder.layers.0.self_attn.q_proj",
54
+ "model.vision_tower.vision_model.encoder.layers.0.self_attn.out_proj",
55
+ "model.vision_tower.vision_model.encoder.layers.0.mlp.fc1",
56
+ "model.vision_tower.vision_model.encoder.layers.0.mlp.fc2",
57
+ "model.vision_tower.vision_model.encoder.layers.1.self_attn.k_proj",
58
+ "model.vision_tower.vision_model.encoder.layers.1.self_attn.v_proj",
59
+ "model.vision_tower.vision_model.encoder.layers.1.self_attn.q_proj",
60
+ "model.vision_tower.vision_model.encoder.layers.1.self_attn.out_proj",
61
+ "model.vision_tower.vision_model.encoder.layers.1.mlp.fc1",
62
+ "model.vision_tower.vision_model.encoder.layers.1.mlp.fc2",
63
+ "model.vision_tower.vision_model.encoder.layers.2.self_attn.k_proj",
64
+ "model.vision_tower.vision_model.encoder.layers.2.self_attn.v_proj",
65
+ "model.vision_tower.vision_model.encoder.layers.2.self_attn.q_proj",
66
+ "model.vision_tower.vision_model.encoder.layers.2.self_attn.out_proj",
67
+ "model.vision_tower.vision_model.encoder.layers.2.mlp.fc1",
68
+ "model.vision_tower.vision_model.encoder.layers.2.mlp.fc2",
69
+ "model.vision_tower.vision_model.encoder.layers.3.self_attn.k_proj",
70
+ "model.vision_tower.vision_model.encoder.layers.3.self_attn.v_proj",
71
+ "model.vision_tower.vision_model.encoder.layers.3.self_attn.q_proj",
72
+ "model.vision_tower.vision_model.encoder.layers.3.self_attn.out_proj",
73
+ "model.vision_tower.vision_model.encoder.layers.3.mlp.fc1",
74
+ "model.vision_tower.vision_model.encoder.layers.3.mlp.fc2",
75
+ "model.vision_tower.vision_model.encoder.layers.4.self_attn.k_proj",
76
+ "model.vision_tower.vision_model.encoder.layers.4.self_attn.v_proj",
77
+ "model.vision_tower.vision_model.encoder.layers.4.self_attn.q_proj",
78
+ "model.vision_tower.vision_model.encoder.layers.4.self_attn.out_proj",
79
+ "model.vision_tower.vision_model.encoder.layers.4.mlp.fc1",
80
+ "model.vision_tower.vision_model.encoder.layers.4.mlp.fc2",
81
+ "model.vision_tower.vision_model.encoder.layers.5.self_attn.k_proj",
82
+ "model.vision_tower.vision_model.encoder.layers.5.self_attn.v_proj",
83
+ "model.vision_tower.vision_model.encoder.layers.5.self_attn.q_proj",
84
+ "model.vision_tower.vision_model.encoder.layers.5.self_attn.out_proj",
85
+ "model.vision_tower.vision_model.encoder.layers.5.mlp.fc1",
86
+ "model.vision_tower.vision_model.encoder.layers.5.mlp.fc2",
87
+ "model.vision_tower.vision_model.encoder.layers.6.self_attn.k_proj",
88
+ "model.vision_tower.vision_model.encoder.layers.6.self_attn.v_proj",
89
+ "model.vision_tower.vision_model.encoder.layers.6.self_attn.q_proj",
90
+ "model.vision_tower.vision_model.encoder.layers.6.self_attn.out_proj",
91
+ "model.vision_tower.vision_model.encoder.layers.6.mlp.fc1",
92
+ "model.vision_tower.vision_model.encoder.layers.6.mlp.fc2",
93
+ "model.vision_tower.vision_model.encoder.layers.7.self_attn.k_proj",
94
+ "model.vision_tower.vision_model.encoder.layers.7.self_attn.v_proj",
95
+ "model.vision_tower.vision_model.encoder.layers.7.self_attn.q_proj",
96
+ "model.vision_tower.vision_model.encoder.layers.7.self_attn.out_proj",
97
+ "model.vision_tower.vision_model.encoder.layers.7.mlp.fc1",
98
+ "model.vision_tower.vision_model.encoder.layers.7.mlp.fc2",
99
+ "model.vision_tower.vision_model.encoder.layers.8.self_attn.k_proj",
100
+ "model.vision_tower.vision_model.encoder.layers.8.self_attn.v_proj",
101
+ "model.vision_tower.vision_model.encoder.layers.8.self_attn.q_proj",
102
+ "model.vision_tower.vision_model.encoder.layers.8.self_attn.out_proj",
103
+ "model.vision_tower.vision_model.encoder.layers.8.mlp.fc1",
104
+ "model.vision_tower.vision_model.encoder.layers.8.mlp.fc2",
105
+ "model.vision_tower.vision_model.encoder.layers.9.self_attn.k_proj",
106
+ "model.vision_tower.vision_model.encoder.layers.9.self_attn.v_proj",
107
+ "model.vision_tower.vision_model.encoder.layers.9.self_attn.q_proj",
108
+ "model.vision_tower.vision_model.encoder.layers.9.self_attn.out_proj",
109
+ "model.vision_tower.vision_model.encoder.layers.9.mlp.fc1",
110
+ "model.vision_tower.vision_model.encoder.layers.9.mlp.fc2",
111
+ "model.vision_tower.vision_model.encoder.layers.10.self_attn.k_proj",
112
+ "model.vision_tower.vision_model.encoder.layers.10.self_attn.v_proj",
113
+ "model.vision_tower.vision_model.encoder.layers.10.self_attn.q_proj",
114
+ "model.vision_tower.vision_model.encoder.layers.10.self_attn.out_proj",
115
+ "model.vision_tower.vision_model.encoder.layers.10.mlp.fc1",
116
+ "model.vision_tower.vision_model.encoder.layers.10.mlp.fc2",
117
+ "model.vision_tower.vision_model.encoder.layers.11.self_attn.k_proj",
118
+ "model.vision_tower.vision_model.encoder.layers.11.self_attn.v_proj",
119
+ "model.vision_tower.vision_model.encoder.layers.11.self_attn.q_proj",
120
+ "model.vision_tower.vision_model.encoder.layers.11.self_attn.out_proj",
121
+ "model.vision_tower.vision_model.encoder.layers.11.mlp.fc1",
122
+ "model.vision_tower.vision_model.encoder.layers.11.mlp.fc2",
123
+ "model.vision_tower.vision_model.encoder.layers.12.self_attn.k_proj",
124
+ "model.vision_tower.vision_model.encoder.layers.12.self_attn.v_proj",
125
+ "model.vision_tower.vision_model.encoder.layers.12.self_attn.q_proj",
126
+ "model.vision_tower.vision_model.encoder.layers.12.self_attn.out_proj",
127
+ "model.vision_tower.vision_model.encoder.layers.12.mlp.fc1",
128
+ "model.vision_tower.vision_model.encoder.layers.12.mlp.fc2",
129
+ "model.vision_tower.vision_model.encoder.layers.13.self_attn.k_proj",
130
+ "model.vision_tower.vision_model.encoder.layers.13.self_attn.v_proj",
131
+ "model.vision_tower.vision_model.encoder.layers.13.self_attn.q_proj",
132
+ "model.vision_tower.vision_model.encoder.layers.13.self_attn.out_proj",
133
+ "model.vision_tower.vision_model.encoder.layers.13.mlp.fc1",
134
+ "model.vision_tower.vision_model.encoder.layers.13.mlp.fc2",
135
+ "model.vision_tower.vision_model.encoder.layers.14.self_attn.k_proj",
136
+ "model.vision_tower.vision_model.encoder.layers.14.self_attn.v_proj",
137
+ "model.vision_tower.vision_model.encoder.layers.14.self_attn.q_proj",
138
+ "model.vision_tower.vision_model.encoder.layers.14.self_attn.out_proj",
139
+ "model.vision_tower.vision_model.encoder.layers.14.mlp.fc1",
140
+ "model.vision_tower.vision_model.encoder.layers.14.mlp.fc2",
141
+ "model.vision_tower.vision_model.encoder.layers.15.self_attn.k_proj",
142
+ "model.vision_tower.vision_model.encoder.layers.15.self_attn.v_proj",
143
+ "model.vision_tower.vision_model.encoder.layers.15.self_attn.q_proj",
144
+ "model.vision_tower.vision_model.encoder.layers.15.self_attn.out_proj",
145
+ "model.vision_tower.vision_model.encoder.layers.15.mlp.fc1",
146
+ "model.vision_tower.vision_model.encoder.layers.15.mlp.fc2",
147
+ "model.vision_tower.vision_model.encoder.layers.16.self_attn.k_proj",
148
+ "model.vision_tower.vision_model.encoder.layers.16.self_attn.v_proj",
149
+ "model.vision_tower.vision_model.encoder.layers.16.self_attn.q_proj",
150
+ "model.vision_tower.vision_model.encoder.layers.16.self_attn.out_proj",
151
+ "model.vision_tower.vision_model.encoder.layers.16.mlp.fc1",
152
+ "model.vision_tower.vision_model.encoder.layers.16.mlp.fc2",
153
+ "model.vision_tower.vision_model.encoder.layers.17.self_attn.k_proj",
154
+ "model.vision_tower.vision_model.encoder.layers.17.self_attn.v_proj",
155
+ "model.vision_tower.vision_model.encoder.layers.17.self_attn.q_proj",
156
+ "model.vision_tower.vision_model.encoder.layers.17.self_attn.out_proj",
157
+ "model.vision_tower.vision_model.encoder.layers.17.mlp.fc1",
158
+ "model.vision_tower.vision_model.encoder.layers.17.mlp.fc2",
159
+ "model.vision_tower.vision_model.encoder.layers.18.self_attn.k_proj",
160
+ "model.vision_tower.vision_model.encoder.layers.18.self_attn.v_proj",
161
+ "model.vision_tower.vision_model.encoder.layers.18.self_attn.q_proj",
162
+ "model.vision_tower.vision_model.encoder.layers.18.self_attn.out_proj",
163
+ "model.vision_tower.vision_model.encoder.layers.18.mlp.fc1",
164
+ "model.vision_tower.vision_model.encoder.layers.18.mlp.fc2",
165
+ "model.vision_tower.vision_model.encoder.layers.19.self_attn.k_proj",
166
+ "model.vision_tower.vision_model.encoder.layers.19.self_attn.v_proj",
167
+ "model.vision_tower.vision_model.encoder.layers.19.self_attn.q_proj",
168
+ "model.vision_tower.vision_model.encoder.layers.19.self_attn.out_proj",
169
+ "model.vision_tower.vision_model.encoder.layers.19.mlp.fc1",
170
+ "model.vision_tower.vision_model.encoder.layers.19.mlp.fc2",
171
+ "model.vision_tower.vision_model.encoder.layers.20.self_attn.k_proj",
172
+ "model.vision_tower.vision_model.encoder.layers.20.self_attn.v_proj",
173
+ "model.vision_tower.vision_model.encoder.layers.20.self_attn.q_proj",
174
+ "model.vision_tower.vision_model.encoder.layers.20.self_attn.out_proj",
175
+ "model.vision_tower.vision_model.encoder.layers.20.mlp.fc1",
176
+ "model.vision_tower.vision_model.encoder.layers.20.mlp.fc2",
177
+ "model.vision_tower.vision_model.encoder.layers.21.self_attn.k_proj",
178
+ "model.vision_tower.vision_model.encoder.layers.21.self_attn.v_proj",
179
+ "model.vision_tower.vision_model.encoder.layers.21.self_attn.q_proj",
180
+ "model.vision_tower.vision_model.encoder.layers.21.self_attn.out_proj",
181
+ "model.vision_tower.vision_model.encoder.layers.21.mlp.fc1",
182
+ "model.vision_tower.vision_model.encoder.layers.21.mlp.fc2",
183
+ "model.vision_tower.vision_model.encoder.layers.22.self_attn.k_proj",
184
+ "model.vision_tower.vision_model.encoder.layers.22.self_attn.v_proj",
185
+ "model.vision_tower.vision_model.encoder.layers.22.self_attn.q_proj",
186
+ "model.vision_tower.vision_model.encoder.layers.22.self_attn.out_proj",
187
+ "model.vision_tower.vision_model.encoder.layers.22.mlp.fc1",
188
+ "model.vision_tower.vision_model.encoder.layers.22.mlp.fc2",
189
+ "model.vision_tower.vision_model.encoder.layers.23.self_attn.k_proj",
190
+ "model.vision_tower.vision_model.encoder.layers.23.self_attn.v_proj",
191
+ "model.vision_tower.vision_model.encoder.layers.23.self_attn.q_proj",
192
+ "model.vision_tower.vision_model.encoder.layers.23.self_attn.out_proj",
193
+ "model.vision_tower.vision_model.encoder.layers.23.mlp.fc1",
194
+ "model.vision_tower.vision_model.encoder.layers.23.mlp.fc2",
195
+ "model.vision_tower.vision_model.encoder.layers.24.self_attn.k_proj",
196
+ "model.vision_tower.vision_model.encoder.layers.24.self_attn.v_proj",
197
+ "model.vision_tower.vision_model.encoder.layers.24.self_attn.q_proj",
198
+ "model.vision_tower.vision_model.encoder.layers.24.self_attn.out_proj",
199
+ "model.vision_tower.vision_model.encoder.layers.24.mlp.fc1",
200
+ "model.vision_tower.vision_model.encoder.layers.24.mlp.fc2",
201
+ "model.vision_tower.vision_model.encoder.layers.25.self_attn.k_proj",
202
+ "model.vision_tower.vision_model.encoder.layers.25.self_attn.v_proj",
203
+ "model.vision_tower.vision_model.encoder.layers.25.self_attn.q_proj",
204
+ "model.vision_tower.vision_model.encoder.layers.25.self_attn.out_proj",
205
+ "model.vision_tower.vision_model.encoder.layers.25.mlp.fc1",
206
+ "model.vision_tower.vision_model.encoder.layers.25.mlp.fc2",
207
+ "model.vision_tower.vision_model.encoder.layers.26.self_attn.k_proj",
208
+ "model.vision_tower.vision_model.encoder.layers.26.self_attn.v_proj",
209
+ "model.vision_tower.vision_model.encoder.layers.26.self_attn.q_proj",
210
+ "model.vision_tower.vision_model.encoder.layers.26.self_attn.out_proj",
211
+ "model.vision_tower.vision_model.encoder.layers.26.mlp.fc1",
212
+ "model.vision_tower.vision_model.encoder.layers.26.mlp.fc2",
213
+ "lm_head"
214
+ ],
215
+ "kv_cache_scheme": null,
216
+ "quant_method": "compressed-tensors",
217
+ "quantization_status": "compressed"
218
+ },
219
  "text_config": {
220
+ "_sliding_window_pattern": 6,
221
+ "attention_bias": false,
222
+ "attention_dropout": 0.0,
223
+ "attn_logit_softcapping": null,
224
+ "cache_implementation": "hybrid",
225
+ "final_logit_softcapping": null,
226
  "head_dim": 128,
227
+ "hidden_activation": "gelu_pytorch_tanh",
228
  "hidden_size": 5376,
229
+ "initializer_range": 0.02,
230
  "intermediate_size": 21504,
231
+ "layer_types": [
232
+ "sliding_attention",
233
+ "sliding_attention",
234
+ "sliding_attention",
235
+ "sliding_attention",
236
+ "sliding_attention",
237
+ "full_attention",
238
+ "sliding_attention",
239
+ "sliding_attention",
240
+ "sliding_attention",
241
+ "sliding_attention",
242
+ "sliding_attention",
243
+ "full_attention",
244
+ "sliding_attention",
245
+ "sliding_attention",
246
+ "sliding_attention",
247
+ "sliding_attention",
248
+ "sliding_attention",
249
+ "full_attention",
250
+ "sliding_attention",
251
+ "sliding_attention",
252
+ "sliding_attention",
253
+ "sliding_attention",
254
+ "sliding_attention",
255
+ "full_attention",
256
+ "sliding_attention",
257
+ "sliding_attention",
258
+ "sliding_attention",
259
+ "sliding_attention",
260
+ "sliding_attention",
261
+ "full_attention",
262
+ "sliding_attention",
263
+ "sliding_attention",
264
+ "sliding_attention",
265
+ "sliding_attention",
266
+ "sliding_attention",
267
+ "full_attention",
268
+ "sliding_attention",
269
+ "sliding_attention",
270
+ "sliding_attention",
271
+ "sliding_attention",
272
+ "sliding_attention",
273
+ "full_attention",
274
+ "sliding_attention",
275
+ "sliding_attention",
276
+ "sliding_attention",
277
+ "sliding_attention",
278
+ "sliding_attention",
279
+ "full_attention",
280
+ "sliding_attention",
281
+ "sliding_attention",
282
+ "sliding_attention",
283
+ "sliding_attention",
284
+ "sliding_attention",
285
+ "full_attention",
286
+ "sliding_attention",
287
+ "sliding_attention",
288
+ "sliding_attention",
289
+ "sliding_attention",
290
+ "sliding_attention",
291
+ "full_attention",
292
+ "sliding_attention",
293
+ "sliding_attention"
294
+ ],
295
+ "max_position_embeddings": 131072,
296
  "model_type": "gemma3_text",
297
  "num_attention_heads": 32,
298
  "num_hidden_layers": 62,
299
  "num_key_value_heads": 16,
300
  "query_pre_attn_scalar": 168,
301
+ "rms_norm_eps": 1e-06,
302
+ "rope_local_base_freq": 10000.0,
303
  "rope_scaling": {
304
  "factor": 8.0,
305
  "rope_type": "linear"
306
  },
307
+ "rope_theta": 1000000.0,
308
+ "sliding_window": 1024,
309
+ "sliding_window_pattern": 6,
310
+ "use_cache": true,
311
+ "vocab_size": 262208
312
  },
313
  "torch_dtype": "bfloat16",
314
+ "transformers_version": "4.52.4",
315
  "vision_config": {
316
+ "attention_dropout": 0.0,
317
+ "hidden_act": "gelu_pytorch_tanh",
318
  "hidden_size": 1152,
319
  "image_size": 896,
320
  "intermediate_size": 4304,
321
+ "layer_norm_eps": 1e-06,
322
  "model_type": "siglip_vision_model",
323
  "num_attention_heads": 16,
324
+ "num_channels": 3,
325
  "num_hidden_layers": 27,
326
  "patch_size": 14,
327
  "vision_use_head": false