Hugging Face's logo

CocoRoF
/
POLAR_gemma-3-12b-4bit-dpo

Image-Text-to-Text
Transformers
Safetensors
gemma3
Generated from Trainer
trl
dpo
conversational
text-generation-inference
4-bit precision
bitsandbytes
Model card Files Community
CocoRoF commited on
Commit
cc18537
·
verified ·
1 Parent(s): 0b2a45f

gemma-12b-4bit-dpo Done

Browse files
Files changed (6) hide show
  1. config.json +195 -0
  2. generation_config.json +13 -0
  3. model-00001-of-00003.safetensors +3 -0
  4. model-00002-of-00003.safetensors +3 -0
  5. model-00003-of-00003.safetensors +3 -0
  6. model.safetensors.index.json +0 -0
config.json ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "eoi_token_index": 256000,
8
+ "eos_token_id": 106,
9
+ "id2label": {
10
+ "0": "LABEL_0",
11
+ "1": "LABEL_1",
12
+ "2": "LABEL_2",
13
+ "3": "LABEL_3",
14
+ "4": "LABEL_4",
15
+ "5": "LABEL_5",
16
+ "6": "LABEL_6",
17
+ "7": "LABEL_7",
18
+ "8": "LABEL_8",
19
+ "9": "LABEL_9",
20
+ "10": "LABEL_10",
21
+ "11": "LABEL_11",
22
+ "12": "LABEL_12",
23
+ "13": "LABEL_13",
24
+ "14": "LABEL_14",
25
+ "15": "LABEL_15",
26
+ "16": "LABEL_16"
27
+ },
28
+ "image_token_index": 262144,
29
+ "initializer_range": 0.02,
30
+ "label2id": {
31
+ "LABEL_0": 0,
32
+ "LABEL_1": 1,
33
+ "LABEL_10": 10,
34
+ "LABEL_11": 11,
35
+ "LABEL_12": 12,
36
+ "LABEL_13": 13,
37
+ "LABEL_14": 14,
38
+ "LABEL_15": 15,
39
+ "LABEL_16": 16,
40
+ "LABEL_2": 2,
41
+ "LABEL_3": 3,
42
+ "LABEL_4": 4,
43
+ "LABEL_5": 5,
44
+ "LABEL_6": 6,
45
+ "LABEL_7": 7,
46
+ "LABEL_8": 8,
47
+ "LABEL_9": 9
48
+ },
49
+ "mm_tokens_per_image": 256,
50
+ "model_type": "gemma3",
51
+ "pad_token_id": 0,
52
+ "quantization_config": {
53
+ "_load_in_4bit": true,
54
+ "_load_in_8bit": false,
55
+ "bnb_4bit_compute_dtype": "bfloat16",
56
+ "bnb_4bit_quant_storage": "uint8",
57
+ "bnb_4bit_quant_type": "nf4",
58
+ "bnb_4bit_use_double_quant": true,
59
+ "llm_int8_enable_fp32_cpu_offload": false,
60
+ "llm_int8_has_fp16_weight": false,
61
+ "llm_int8_skip_modules": [
62
+ "lm_head",
63
+ "multi_modal_projector",
64
+ "merger",
65
+ "modality_projection",
66
+ "language_model.model.layers.14.mlp",
67
+ "language_model.model.layers.2.self_attn",
68
+ "language_model.model.layers.12.mlp",
69
+ "language_model.model.layers.8.self_attn",
70
+ "language_model.model.layers.6.self_attn",
71
+ "language_model.model.layers.13.mlp",
72
+ "language_model.model.layers.1.mlp",
73
+ "language_model.model.layers.10.mlp",
74
+ "language_model.model.layers.16.mlp",
75
+ "language_model.model.layers.5.self_attn",
76
+ "language_model.model.layers.2.mlp",
77
+ "language_model.model.layers.4.mlp",
78
+ "language_model.model.layers.5.mlp",
79
+ "language_model.model.layers.9.mlp",
80
+ "language_model.model.layers.7.self_attn",
81
+ "language_model.model.layers.6.mlp",
82
+ "language_model.model.layers.8.mlp",
83
+ "vision_tower.vision_model.encoder.layers.23.self_attn",
84
+ "language_model.model.layers.7.mlp",
85
+ "language_model.model.layers.3.self_attn",
86
+ "language_model.model.layers.11.mlp",
87
+ "language_model.model.layers.3.mlp",
88
+ "vision_tower.vision_model.encoder.layers.19.self_attn",
89
+ "vision_tower.vision_model.encoder.layers.25.self_attn",
90
+ "vision_tower.vision_model.encoder.layers.22.self_attn",
91
+ "vision_tower.vision_model.encoder.layers.21.self_attn",
92
+ "vision_tower.vision_model.encoder.layers.18.self_attn",
93
+ "vision_tower.vision_model.encoder.layers.17.mlp",
94
+ "vision_tower.vision_model.encoder.layers.24.self_attn",
95
+ "vision_tower.vision_model.encoder.layers.24.mlp",
96
+ "vision_tower.vision_model.encoder.layers.26.self_attn",
97
+ "vision_tower.vision_model.encoder.layers.20.mlp",
98
+ "vision_tower.vision_model.encoder.layers.20.self_attn",
99
+ "vision_tower.vision_model.encoder.layers.23.mlp",
100
+ "vision_tower.vision_model.encoder.layers.19.mlp",
101
+ "vision_tower.vision_model.encoder.layers.15.self_attn",
102
+ "vision_tower.vision_model.encoder.layers.14.mlp",
103
+ "vision_tower.vision_model.encoder.layers.13.mlp",
104
+ "vision_tower.vision_model.encoder.layers.21.mlp",
105
+ "vision_tower.vision_model.encoder.layers.16.mlp",
106
+ "vision_tower.vision_model.encoder.layers.25.mlp",
107
+ "vision_tower.vision_model.encoder.layers.15.mlp",
108
+ "vision_tower.vision_model.encoder.layers.14.self_attn",
109
+ "vision_tower.vision_model.encoder.layers.16.self_attn",
110
+ "vision_tower.vision_model.encoder.layers.10.mlp",
111
+ "vision_tower.vision_model.encoder.layers.17.self_attn",
112
+ "vision_tower.vision_model.encoder.layers.9.mlp",
113
+ "vision_tower.vision_model.encoder.layers.11.mlp",
114
+ "vision_tower.vision_model.encoder.layers.12.mlp",
115
+ "vision_tower.vision_model.encoder.layers.10.self_attn",
116
+ "vision_tower.vision_model.encoder.layers.22.mlp",
117
+ "vision_tower.vision_model.encoder.layers.13.self_attn",
118
+ "vision_tower.vision_model.encoder.layers.18.mlp",
119
+ "vision_tower.vision_model.encoder.layers.12.self_attn",
120
+ "vision_tower.vision_model.encoder.layers.7.mlp",
121
+ "vision_tower.vision_model.encoder.layers.6.mlp",
122
+ "vision_tower.vision_model.encoder.layers.8.self_attn",
123
+ "vision_tower.vision_model.encoder.layers.9.self_attn",
124
+ "vision_tower.vision_model.encoder.layers.4.mlp",
125
+ "vision_tower.vision_model.encoder.layers.5.mlp",
126
+ "vision_tower.vision_model.encoder.layers.8.mlp",
127
+ "vision_tower.vision_model.encoder.layers.1.mlp",
128
+ "vision_tower.vision_model.encoder.layers.6.self_attn",
129
+ "vision_tower.vision_model.encoder.layers.4.self_attn",
130
+ "vision_tower.vision_model.encoder.layers.11.self_attn",
131
+ "vision_tower.vision_model.encoder.layers.7.self_attn",
132
+ "vision_tower.vision_model.encoder.layers.0.self_attn",
133
+ "vision_tower.vision_model.encoder.layers.3.mlp",
134
+ "vision_tower.vision_model.encoder.layers.1.self_attn",
135
+ "vision_tower.vision_model.encoder.layers.2.mlp",
136
+ "vision_tower.vision_model.encoder.layers.5.self_attn",
137
+ "vision_tower.vision_model.encoder.layers.3.self_attn",
138
+ "vision_tower.vision_model.encoder.layers.0.mlp",
139
+ "vision_tower.vision_model.encoder.layers.26.mlp",
140
+ "vision_tower.vision_model.encoder.layers.2.self_attn"
141
+ ],
142
+ "llm_int8_threshold": 6.0,
143
+ "load_in_4bit": true,
144
+ "load_in_8bit": false,
145
+ "quant_method": "bitsandbytes"
146
+ },
147
+ "text_config": {
148
+ "attention_bias": false,
149
+ "attention_dropout": 0.0,
150
+ "attn_logit_softcapping": null,
151
+ "cache_implementation": "hybrid",
152
+ "final_logit_softcapping": null,
153
+ "head_dim": 256,
154
+ "hidden_activation": "gelu_pytorch_tanh",
155
+ "hidden_size": 3840,
156
+ "initializer_range": 0.02,
157
+ "intermediate_size": 15360,
158
+ "max_position_embeddings": 131072,
159
+ "model_type": "gemma3_text",
160
+ "num_attention_heads": 16,
161
+ "num_hidden_layers": 48,
162
+ "num_key_value_heads": 8,
163
+ "query_pre_attn_scalar": 256,
164
+ "rms_norm_eps": 1e-06,
165
+ "rope_local_base_freq": 10000.0,
166
+ "rope_scaling": {
167
+ "factor": 8.0,
168
+ "rope_type": "linear"
169
+ },
170
+ "rope_theta": 1000000.0,
171
+ "sliding_window": 1024,
172
+ "sliding_window_pattern": 6,
173
+ "torch_dtype": "bfloat16",
174
+ "use_cache": true,
175
+ "vocab_size": 262208
176
+ },
177
+ "torch_dtype": "bfloat16",
178
+ "transformers_version": "4.50.0",
179
+ "unsloth_fixed": true,
180
+ "vision_config": {
181
+ "attention_dropout": 0.0,
182
+ "hidden_act": "gelu_pytorch_tanh",
183
+ "hidden_size": 1152,
184
+ "image_size": 896,
185
+ "intermediate_size": 4304,
186
+ "layer_norm_eps": 1e-06,
187
+ "model_type": "siglip_vision_model",
188
+ "num_attention_heads": 16,
189
+ "num_channels": 3,
190
+ "num_hidden_layers": 27,
191
+ "patch_size": 14,
192
+ "torch_dtype": "bfloat16",
193
+ "vision_use_head": false
194
+ }
195
+ }
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": "4.50.0"
13
+ }
model-00001-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5356852c408b6bc10f2be8d54d4a38b9b13979edd9c3d8af9c623172615cf4d8
3
+ size 4993943756
model-00002-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0628399a506065af477313eeac211023be351ce7201bbe8a224b15e36e8c809f
3
+ size 4975697510
model-00003-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2536d75523ea6dd742a1dc0d716ea52901d792371e8e22a3bae20ad69a9443ad
3
+ size 4811940920
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff