Hugging Face's logo

Portx
/
v2-trial-4-gemma-4bit

Image-Text-to-Text
Transformers
Safetensors
English
gemma3n
text-generation-inference
unsloth
conversational
4-bit precision
bitsandbytes
Model card Files
xet
Community
Portx commited on
Commit
01e7e4e
·
verified ·
1 Parent(s): f0884b8

(Trained with Unsloth)

Browse files
Files changed (14) hide show
  1. .gitattributes +1 -0
  2. chat_template.jinja +49 -0
  3. config.json +519 -0
  4. generation_config.json +10 -0
  5. model-00001-of-00003.safetensors +3 -0
  6. model-00002-of-00003.safetensors +3 -0
  7. model-00003-of-00003.safetensors +3 -0
  8. model.safetensors.index.json +0 -0
  9. preprocessor_config.json +51 -0
  10. processor_config.json +5 -0
  11. special_tokens_map.json +36 -0
  12. tokenizer.json +3 -0
  13. tokenizer.model +3 -0
  14. 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.jinja ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'audio' -%}
33
+ {{ '<audio_soft_token>' }}
34
+ {%- elif item['type'] == 'image' -%}
35
+ {{ '<image_soft_token>' }}
36
+ {%- elif item['type'] == 'text' -%}
37
+ {{ item['text'] | trim }}
38
+ {%- endif -%}
39
+ {%- endfor -%}
40
+ {%- else -%}
41
+ {{ raise_exception("Invalid content type") }}
42
+ {%- endif -%}
43
+ {{ '<end_of_turn>
44
+ ' }}
45
+ {%- endfor -%}
46
+ {%- if add_generation_prompt -%}
47
+ {{'<start_of_turn>model
48
+ '}}
49
+ {%- endif -%}
config.json ADDED
@@ -0,0 +1,519 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3nForConditionalGeneration"
4
+ ],
5
+ "audio_config": {
6
+ "conf_attention_chunk_size": 12,
7
+ "conf_attention_context_left": 13,
8
+ "conf_attention_context_right": 0,
9
+ "conf_attention_logit_cap": 50.0,
10
+ "conf_conv_kernel_size": 5,
11
+ "conf_num_attention_heads": 8,
12
+ "conf_num_hidden_layers": 12,
13
+ "conf_reduction_factor": 4,
14
+ "conf_residual_weight": 0.5,
15
+ "gradient_clipping": 10000000000.0,
16
+ "hidden_size": 1536,
17
+ "input_feat_size": 128,
18
+ "model_type": "gemma3n_audio",
19
+ "rms_norm_eps": 1e-06,
20
+ "sscp_conv_channel_size": [
21
+ 128,
22
+ 32
23
+ ],
24
+ "sscp_conv_group_norm_eps": 0.001,
25
+ "sscp_conv_kernel_size": [
26
+ [
27
+ 3,
28
+ 3
29
+ ],
30
+ [
31
+ 3,
32
+ 3
33
+ ]
34
+ ],
35
+ "sscp_conv_stride_size": [
36
+ [
37
+ 2,
38
+ 2
39
+ ],
40
+ [
41
+ 2,
42
+ 2
43
+ ]
44
+ ],
45
+ "torch_dtype": "bfloat16",
46
+ "vocab_offset": 262272,
47
+ "vocab_size": 128
48
+ },
49
+ "audio_soft_tokens_per_image": 188,
50
+ "audio_token_id": 262273,
51
+ "boa_token_id": 256000,
52
+ "boi_token_id": 255999,
53
+ "bos_token_id": 2,
54
+ "eoa_token_id": 262272,
55
+ "eoi_token_id": 262144,
56
+ "eos_token_id": 1,
57
+ "image_token_id": 262145,
58
+ "initializer_range": 0.02,
59
+ "model_type": "gemma3n",
60
+ "pad_token_id": 0,
61
+ "quantization_config": {
62
+ "bnb_4bit_compute_dtype": "bfloat16",
63
+ "bnb_4bit_quant_type": "nf4",
64
+ "bnb_4bit_use_double_quant": true,
65
+ "llm_int8_enable_fp32_cpu_offload": false,
66
+ "llm_int8_has_fp16_weight": false,
67
+ "llm_int8_skip_modules": [
68
+ "model.language_model.layers.0.altup.correction_coefs",
69
+ "model.language_model.layers.0.altup.prediction_coefs",
70
+ "model.language_model.layers.0.altup.modality_router",
71
+ "model.language_model.layers.0.per_layer_projection",
72
+ "model.language_model.layers.1.altup.correction_coefs",
73
+ "model.language_model.layers.1.altup.prediction_coefs",
74
+ "model.language_model.layers.1.altup.modality_router",
75
+ "model.language_model.layers.1.per_layer_projection",
76
+ "model.language_model.layers.2.altup.correction_coefs",
77
+ "model.language_model.layers.2.altup.prediction_coefs",
78
+ "model.language_model.layers.2.altup.modality_router",
79
+ "model.language_model.layers.2.per_layer_projection",
80
+ "model.language_model.layers.3.altup.correction_coefs",
81
+ "model.language_model.layers.3.altup.prediction_coefs",
82
+ "model.language_model.layers.3.altup.modality_router",
83
+ "model.language_model.layers.3.per_layer_projection",
84
+ "model.language_model.layers.4.altup.correction_coefs",
85
+ "model.language_model.layers.4.altup.prediction_coefs",
86
+ "model.language_model.layers.4.altup.modality_router",
87
+ "model.language_model.layers.4.per_layer_projection",
88
+ "model.language_model.layers.5.altup.correction_coefs",
89
+ "model.language_model.layers.5.altup.prediction_coefs",
90
+ "model.language_model.layers.5.altup.modality_router",
91
+ "model.language_model.layers.5.per_layer_projection",
92
+ "model.language_model.layers.6.altup.correction_coefs",
93
+ "model.language_model.layers.6.altup.prediction_coefs",
94
+ "model.language_model.layers.6.altup.modality_router",
95
+ "model.language_model.layers.6.per_layer_projection",
96
+ "model.language_model.layers.7.altup.correction_coefs",
97
+ "model.language_model.layers.7.altup.prediction_coefs",
98
+ "model.language_model.layers.7.altup.modality_router",
99
+ "model.language_model.layers.7.per_layer_projection",
100
+ "model.language_model.layers.8.altup.correction_coefs",
101
+ "model.language_model.layers.8.altup.prediction_coefs",
102
+ "model.language_model.layers.8.altup.modality_router",
103
+ "model.language_model.layers.8.per_layer_projection",
104
+ "model.language_model.layers.9.altup.correction_coefs",
105
+ "model.language_model.layers.9.altup.prediction_coefs",
106
+ "model.language_model.layers.9.altup.modality_router",
107
+ "model.language_model.layers.9.per_layer_projection",
108
+ "model.language_model.layers.10.altup.correction_coefs",
109
+ "model.language_model.layers.10.altup.prediction_coefs",
110
+ "model.language_model.layers.10.altup.modality_router",
111
+ "model.language_model.layers.10.per_layer_projection",
112
+ "model.language_model.layers.11.altup.correction_coefs",
113
+ "model.language_model.layers.11.altup.prediction_coefs",
114
+ "model.language_model.layers.11.altup.modality_router",
115
+ "model.language_model.layers.11.per_layer_projection",
116
+ "model.language_model.layers.12.altup.correction_coefs",
117
+ "model.language_model.layers.12.altup.prediction_coefs",
118
+ "model.language_model.layers.12.altup.modality_router",
119
+ "model.language_model.layers.12.per_layer_projection",
120
+ "model.language_model.layers.13.altup.correction_coefs",
121
+ "model.language_model.layers.13.altup.prediction_coefs",
122
+ "model.language_model.layers.13.altup.modality_router",
123
+ "model.language_model.layers.13.per_layer_projection",
124
+ "model.language_model.layers.14.altup.correction_coefs",
125
+ "model.language_model.layers.14.altup.prediction_coefs",
126
+ "model.language_model.layers.14.altup.modality_router",
127
+ "model.language_model.layers.14.per_layer_projection",
128
+ "model.language_model.layers.15.altup.correction_coefs",
129
+ "model.language_model.layers.15.altup.prediction_coefs",
130
+ "model.language_model.layers.15.altup.modality_router",
131
+ "model.language_model.layers.15.per_layer_projection",
132
+ "model.language_model.layers.16.altup.correction_coefs",
133
+ "model.language_model.layers.16.altup.prediction_coefs",
134
+ "model.language_model.layers.16.altup.modality_router",
135
+ "model.language_model.layers.16.per_layer_projection",
136
+ "model.language_model.layers.17.altup.correction_coefs",
137
+ "model.language_model.layers.17.altup.prediction_coefs",
138
+ "model.language_model.layers.17.altup.modality_router",
139
+ "model.language_model.layers.17.per_layer_projection",
140
+ "model.language_model.layers.18.altup.correction_coefs",
141
+ "model.language_model.layers.18.altup.prediction_coefs",
142
+ "model.language_model.layers.18.altup.modality_router",
143
+ "model.language_model.layers.18.per_layer_projection",
144
+ "model.language_model.layers.19.altup.correction_coefs",
145
+ "model.language_model.layers.19.altup.prediction_coefs",
146
+ "model.language_model.layers.19.altup.modality_router",
147
+ "model.language_model.layers.19.per_layer_projection",
148
+ "model.language_model.layers.20.altup.correction_coefs",
149
+ "model.language_model.layers.20.altup.prediction_coefs",
150
+ "model.language_model.layers.20.altup.modality_router",
151
+ "model.language_model.layers.20.per_layer_projection",
152
+ "model.language_model.layers.21.altup.correction_coefs",
153
+ "model.language_model.layers.21.altup.prediction_coefs",
154
+ "model.language_model.layers.21.altup.modality_router",
155
+ "model.language_model.layers.21.per_layer_projection",
156
+ "model.language_model.layers.22.altup.correction_coefs",
157
+ "model.language_model.layers.22.altup.prediction_coefs",
158
+ "model.language_model.layers.22.altup.modality_router",
159
+ "model.language_model.layers.22.per_layer_projection",
160
+ "model.language_model.layers.23.altup.correction_coefs",
161
+ "model.language_model.layers.23.altup.prediction_coefs",
162
+ "model.language_model.layers.23.altup.modality_router",
163
+ "model.language_model.layers.23.per_layer_projection",
164
+ "model.language_model.layers.24.altup.correction_coefs",
165
+ "model.language_model.layers.24.altup.prediction_coefs",
166
+ "model.language_model.layers.24.altup.modality_router",
167
+ "model.language_model.layers.24.per_layer_projection",
168
+ "model.language_model.layers.25.altup.correction_coefs",
169
+ "model.language_model.layers.25.altup.prediction_coefs",
170
+ "model.language_model.layers.25.altup.modality_router",
171
+ "model.language_model.layers.25.per_layer_projection",
172
+ "model.language_model.layers.26.altup.correction_coefs",
173
+ "model.language_model.layers.26.altup.prediction_coefs",
174
+ "model.language_model.layers.26.altup.modality_router",
175
+ "model.language_model.layers.26.per_layer_projection",
176
+ "model.language_model.layers.27.altup.correction_coefs",
177
+ "model.language_model.layers.27.altup.prediction_coefs",
178
+ "model.language_model.layers.27.altup.modality_router",
179
+ "model.language_model.layers.27.per_layer_projection",
180
+ "model.language_model.layers.28.altup.correction_coefs",
181
+ "model.language_model.layers.28.altup.prediction_coefs",
182
+ "model.language_model.layers.28.altup.modality_router",
183
+ "model.language_model.layers.28.per_layer_projection",
184
+ "model.language_model.layers.29.altup.correction_coefs",
185
+ "model.language_model.layers.29.altup.prediction_coefs",
186
+ "model.language_model.layers.29.altup.modality_router",
187
+ "model.language_model.layers.29.per_layer_projection",
188
+ "model.language_model.layers.30.altup.correction_coefs",
189
+ "model.language_model.layers.30.altup.prediction_coefs",
190
+ "model.language_model.layers.30.altup.modality_router",
191
+ "model.language_model.layers.30.per_layer_projection",
192
+ "model.language_model.layers.31.altup.correction_coefs",
193
+ "model.language_model.layers.31.altup.prediction_coefs",
194
+ "model.language_model.layers.31.altup.modality_router",
195
+ "model.language_model.layers.31.per_layer_projection",
196
+ "model.language_model.layers.32.altup.correction_coefs",
197
+ "model.language_model.layers.32.altup.prediction_coefs",
198
+ "model.language_model.layers.32.altup.modality_router",
199
+ "model.language_model.layers.32.per_layer_projection",
200
+ "model.language_model.layers.33.altup.correction_coefs",
201
+ "model.language_model.layers.33.altup.prediction_coefs",
202
+ "model.language_model.layers.33.altup.modality_router",
203
+ "model.language_model.layers.33.per_layer_projection",
204
+ "model.language_model.layers.34.altup.correction_coefs",
205
+ "model.language_model.layers.34.altup.prediction_coefs",
206
+ "model.language_model.layers.34.altup.modality_router",
207
+ "model.language_model.layers.34.per_layer_projection",
208
+ "model.language_model.altup_projections.0",
209
+ "model.language_model.altup_projections.1",
210
+ "model.language_model.altup_projections.2",
211
+ "model.language_model.altup_unembed_projections.0",
212
+ "model.language_model.altup_unembed_projections.1",
213
+ "model.language_model.altup_unembed_projections.2",
214
+ "model.audio_tower.subsample_conv_projection.input_proj_linear",
215
+ "model.audio_tower.conformer.0.ffw_layer_start.ffw_layer_1",
216
+ "model.audio_tower.conformer.0.ffw_layer_start.ffw_layer_2",
217
+ "model.audio_tower.conformer.0.attention.attn.relative_position_embedding.pos_proj",
218
+ "model.audio_tower.conformer.0.attention.attn.q_proj",
219
+ "model.audio_tower.conformer.0.attention.attn.k_proj",
220
+ "model.audio_tower.conformer.0.attention.attn.v_proj",
221
+ "model.audio_tower.conformer.0.attention.post",
222
+ "model.audio_tower.conformer.0.lconv1d.linear_start",
223
+ "model.audio_tower.conformer.0.lconv1d.linear_end",
224
+ "model.audio_tower.conformer.0.ffw_layer_end.ffw_layer_1",
225
+ "model.audio_tower.conformer.0.ffw_layer_end.ffw_layer_2",
226
+ "model.audio_tower.conformer.1.ffw_layer_start.ffw_layer_1",
227
+ "model.audio_tower.conformer.1.ffw_layer_start.ffw_layer_2",
228
+ "model.audio_tower.conformer.1.attention.attn.relative_position_embedding.pos_proj",
229
+ "model.audio_tower.conformer.1.attention.attn.q_proj",
230
+ "model.audio_tower.conformer.1.attention.attn.k_proj",
231
+ "model.audio_tower.conformer.1.attention.attn.v_proj",
232
+ "model.audio_tower.conformer.1.attention.post",
233
+ "model.audio_tower.conformer.1.lconv1d.linear_start",
234
+ "model.audio_tower.conformer.1.lconv1d.linear_end",
235
+ "model.audio_tower.conformer.1.ffw_layer_end.ffw_layer_1",
236
+ "model.audio_tower.conformer.1.ffw_layer_end.ffw_layer_2",
237
+ "model.audio_tower.conformer.2.ffw_layer_start.ffw_layer_1",
238
+ "model.audio_tower.conformer.2.ffw_layer_start.ffw_layer_2",
239
+ "model.audio_tower.conformer.2.attention.attn.relative_position_embedding.pos_proj",
240
+ "model.audio_tower.conformer.2.attention.attn.q_proj",
241
+ "model.audio_tower.conformer.2.attention.attn.k_proj",
242
+ "model.audio_tower.conformer.2.attention.attn.v_proj",
243
+ "model.audio_tower.conformer.2.attention.post",
244
+ "model.audio_tower.conformer.2.lconv1d.linear_start",
245
+ "model.audio_tower.conformer.2.lconv1d.linear_end",
246
+ "model.audio_tower.conformer.2.ffw_layer_end.ffw_layer_1",
247
+ "model.audio_tower.conformer.2.ffw_layer_end.ffw_layer_2",
248
+ "model.audio_tower.conformer.3.ffw_layer_start.ffw_layer_1",
249
+ "model.audio_tower.conformer.3.ffw_layer_start.ffw_layer_2",
250
+ "model.audio_tower.conformer.3.attention.attn.relative_position_embedding.pos_proj",
251
+ "model.audio_tower.conformer.3.attention.attn.q_proj",
252
+ "model.audio_tower.conformer.3.attention.attn.k_proj",
253
+ "model.audio_tower.conformer.3.attention.attn.v_proj",
254
+ "model.audio_tower.conformer.3.attention.post",
255
+ "model.audio_tower.conformer.3.lconv1d.linear_start",
256
+ "model.audio_tower.conformer.3.lconv1d.linear_end",
257
+ "model.audio_tower.conformer.3.ffw_layer_end.ffw_layer_1",
258
+ "model.audio_tower.conformer.3.ffw_layer_end.ffw_layer_2",
259
+ "model.audio_tower.conformer.4.ffw_layer_start.ffw_layer_1",
260
+ "model.audio_tower.conformer.4.ffw_layer_start.ffw_layer_2",
261
+ "model.audio_tower.conformer.4.attention.attn.relative_position_embedding.pos_proj",
262
+ "model.audio_tower.conformer.4.attention.attn.q_proj",
263
+ "model.audio_tower.conformer.4.attention.attn.k_proj",
264
+ "model.audio_tower.conformer.4.attention.attn.v_proj",
265
+ "model.audio_tower.conformer.4.attention.post",
266
+ "model.audio_tower.conformer.4.lconv1d.linear_start",
267
+ "model.audio_tower.conformer.4.lconv1d.linear_end",
268
+ "model.audio_tower.conformer.4.ffw_layer_end.ffw_layer_1",
269
+ "model.audio_tower.conformer.4.ffw_layer_end.ffw_layer_2",
270
+ "model.audio_tower.conformer.5.ffw_layer_start.ffw_layer_1",
271
+ "model.audio_tower.conformer.5.ffw_layer_start.ffw_layer_2",
272
+ "model.audio_tower.conformer.5.attention.attn.relative_position_embedding.pos_proj",
273
+ "model.audio_tower.conformer.5.attention.attn.q_proj",
274
+ "model.audio_tower.conformer.5.attention.attn.k_proj",
275
+ "model.audio_tower.conformer.5.attention.attn.v_proj",
276
+ "model.audio_tower.conformer.5.attention.post",
277
+ "model.audio_tower.conformer.5.lconv1d.linear_start",
278
+ "model.audio_tower.conformer.5.lconv1d.linear_end",
279
+ "model.audio_tower.conformer.5.ffw_layer_end.ffw_layer_1",
280
+ "model.audio_tower.conformer.5.ffw_layer_end.ffw_layer_2",
281
+ "model.audio_tower.conformer.6.ffw_layer_start.ffw_layer_1",
282
+ "model.audio_tower.conformer.6.ffw_layer_start.ffw_layer_2",
283
+ "model.audio_tower.conformer.6.attention.attn.relative_position_embedding.pos_proj",
284
+ "model.audio_tower.conformer.6.attention.attn.q_proj",
285
+ "model.audio_tower.conformer.6.attention.attn.k_proj",
286
+ "model.audio_tower.conformer.6.attention.attn.v_proj",
287
+ "model.audio_tower.conformer.6.attention.post",
288
+ "model.audio_tower.conformer.6.lconv1d.linear_start",
289
+ "model.audio_tower.conformer.6.lconv1d.linear_end",
290
+ "model.audio_tower.conformer.6.ffw_layer_end.ffw_layer_1",
291
+ "model.audio_tower.conformer.6.ffw_layer_end.ffw_layer_2",
292
+ "model.audio_tower.conformer.7.ffw_layer_start.ffw_layer_1",
293
+ "model.audio_tower.conformer.7.ffw_layer_start.ffw_layer_2",
294
+ "model.audio_tower.conformer.7.attention.attn.relative_position_embedding.pos_proj",
295
+ "model.audio_tower.conformer.7.attention.attn.q_proj",
296
+ "model.audio_tower.conformer.7.attention.attn.k_proj",
297
+ "model.audio_tower.conformer.7.attention.attn.v_proj",
298
+ "model.audio_tower.conformer.7.attention.post",
299
+ "model.audio_tower.conformer.7.lconv1d.linear_start",
300
+ "model.audio_tower.conformer.7.lconv1d.linear_end",
301
+ "model.audio_tower.conformer.7.ffw_layer_end.ffw_layer_1",
302
+ "model.audio_tower.conformer.7.ffw_layer_end.ffw_layer_2",
303
+ "model.audio_tower.conformer.8.ffw_layer_start.ffw_layer_1",
304
+ "model.audio_tower.conformer.8.ffw_layer_start.ffw_layer_2",
305
+ "model.audio_tower.conformer.8.attention.attn.relative_position_embedding.pos_proj",
306
+ "model.audio_tower.conformer.8.attention.attn.q_proj",
307
+ "model.audio_tower.conformer.8.attention.attn.k_proj",
308
+ "model.audio_tower.conformer.8.attention.attn.v_proj",
309
+ "model.audio_tower.conformer.8.attention.post",
310
+ "model.audio_tower.conformer.8.lconv1d.linear_start",
311
+ "model.audio_tower.conformer.8.lconv1d.linear_end",
312
+ "model.audio_tower.conformer.8.ffw_layer_end.ffw_layer_1",
313
+ "model.audio_tower.conformer.8.ffw_layer_end.ffw_layer_2",
314
+ "model.audio_tower.conformer.9.ffw_layer_start.ffw_layer_1",
315
+ "model.audio_tower.conformer.9.ffw_layer_start.ffw_layer_2",
316
+ "model.audio_tower.conformer.9.attention.attn.relative_position_embedding.pos_proj",
317
+ "model.audio_tower.conformer.9.attention.attn.q_proj",
318
+ "model.audio_tower.conformer.9.attention.attn.k_proj",
319
+ "model.audio_tower.conformer.9.attention.attn.v_proj",
320
+ "model.audio_tower.conformer.9.attention.post",
321
+ "model.audio_tower.conformer.9.lconv1d.linear_start",
322
+ "model.audio_tower.conformer.9.lconv1d.linear_end",
323
+ "model.audio_tower.conformer.9.ffw_layer_end.ffw_layer_1",
324
+ "model.audio_tower.conformer.9.ffw_layer_end.ffw_layer_2",
325
+ "model.audio_tower.conformer.10.ffw_layer_start.ffw_layer_1",
326
+ "model.audio_tower.conformer.10.ffw_layer_start.ffw_layer_2",
327
+ "model.audio_tower.conformer.10.attention.attn.relative_position_embedding.pos_proj",
328
+ "model.audio_tower.conformer.10.attention.attn.q_proj",
329
+ "model.audio_tower.conformer.10.attention.attn.k_proj",
330
+ "model.audio_tower.conformer.10.attention.attn.v_proj",
331
+ "model.audio_tower.conformer.10.attention.post",
332
+ "model.audio_tower.conformer.10.lconv1d.linear_start",
333
+ "model.audio_tower.conformer.10.lconv1d.linear_end",
334
+ "model.audio_tower.conformer.10.ffw_layer_end.ffw_layer_1",
335
+ "model.audio_tower.conformer.10.ffw_layer_end.ffw_layer_2",
336
+ "model.audio_tower.conformer.11.ffw_layer_start.ffw_layer_1",
337
+ "model.audio_tower.conformer.11.ffw_layer_start.ffw_layer_2",
338
+ "model.audio_tower.conformer.11.attention.attn.relative_position_embedding.pos_proj",
339
+ "model.audio_tower.conformer.11.attention.attn.q_proj",
340
+ "model.audio_tower.conformer.11.attention.attn.k_proj",
341
+ "model.audio_tower.conformer.11.attention.attn.v_proj",
342
+ "model.audio_tower.conformer.11.attention.post",
343
+ "model.audio_tower.conformer.11.lconv1d.linear_start",
344
+ "model.audio_tower.conformer.11.lconv1d.linear_end",
345
+ "model.audio_tower.conformer.11.ffw_layer_end.ffw_layer_1",
346
+ "model.audio_tower.conformer.11.ffw_layer_end.ffw_layer_2",
347
+ "model.embed_vision.embedding_projection",
348
+ "model.embed_audio.embedding_projection",
349
+ "lm_head"
350
+ ],
351
+ "llm_int8_threshold": 6.0,
352
+ "load_in_4bit": true,
353
+ "load_in_8bit": false,
354
+ "quant_method": "bitsandbytes"
355
+ },
356
+ "text_config": {
357
+ "activation_sparsity_pattern": [
358
+ 0.95,
359
+ 0.95,
360
+ 0.95,
361
+ 0.95,
362
+ 0.95,
363
+ 0.95,
364
+ 0.95,
365
+ 0.95,
366
+ 0.95,
367
+ 0.95,
368
+ 0.0,
369
+ 0.0,
370
+ 0.0,
371
+ 0.0,
372
+ 0.0,
373
+ 0.0,
374
+ 0.0,
375
+ 0.0,
376
+ 0.0,
377
+ 0.0,
378
+ 0.0,
379
+ 0.0,
380
+ 0.0,
381
+ 0.0,
382
+ 0.0,
383
+ 0.0,
384
+ 0.0,
385
+ 0.0,
386
+ 0.0,
387
+ 0.0,
388
+ 0.0,
389
+ 0.0,
390
+ 0.0,
391
+ 0.0,
392
+ 0.0
393
+ ],
394
+ "altup_active_idx": 0,
395
+ "altup_coef_clip": 120.0,
396
+ "altup_correct_scale": true,
397
+ "altup_num_inputs": 4,
398
+ "attention_bias": false,
399
+ "attention_dropout": 0.0,
400
+ "final_logit_softcapping": 30.0,
401
+ "head_dim": 256,
402
+ "hidden_activation": "gelu_pytorch_tanh",
403
+ "hidden_size": 2048,
404
+ "hidden_size_per_layer_input": 256,
405
+ "initializer_range": 0.02,
406
+ "intermediate_size": [
407
+ 16384,
408
+ 16384,
409
+ 16384,
410
+ 16384,
411
+ 16384,
412
+ 16384,
413
+ 16384,
414
+ 16384,
415
+ 16384,
416
+ 16384,
417
+ 16384,
418
+ 16384,
419
+ 16384,
420
+ 16384,
421
+ 16384,
422
+ 16384,
423
+ 16384,
424
+ 16384,
425
+ 16384,
426
+ 16384,
427
+ 16384,
428
+ 16384,
429
+ 16384,
430
+ 16384,
431
+ 16384,
432
+ 16384,
433
+ 16384,
434
+ 16384,
435
+ 16384,
436
+ 16384,
437
+ 16384,
438
+ 16384,
439
+ 16384,
440
+ 16384,
441
+ 16384
442
+ ],
443
+ "laurel_rank": 64,
444
+ "layer_types": [
445
+ "sliding_attention",
446
+ "sliding_attention",
447
+ "sliding_attention",
448
+ "sliding_attention",
449
+ "full_attention",
450
+ "sliding_attention",
451
+ "sliding_attention",
452
+ "sliding_attention",
453
+ "sliding_attention",
454
+ "full_attention",
455
+ "sliding_attention",
456
+ "sliding_attention",
457
+ "sliding_attention",
458
+ "sliding_attention",
459
+ "full_attention",
460
+ "sliding_attention",
461
+ "sliding_attention",
462
+ "sliding_attention",
463
+ "sliding_attention",
464
+ "full_attention",
465
+ "sliding_attention",
466
+ "sliding_attention",
467
+ "sliding_attention",
468
+ "sliding_attention",
469
+ "full_attention",
470
+ "sliding_attention",
471
+ "sliding_attention",
472
+ "sliding_attention",
473
+ "sliding_attention",
474
+ "full_attention",
475
+ "sliding_attention",
476
+ "sliding_attention",
477
+ "sliding_attention",
478
+ "sliding_attention",
479
+ "full_attention"
480
+ ],
481
+ "max_position_embeddings": 32768,
482
+ "model_type": "gemma3n_text",
483
+ "num_attention_heads": 8,
484
+ "num_hidden_layers": 35,
485
+ "num_key_value_heads": 2,
486
+ "num_kv_shared_layers": 15,
487
+ "rms_norm_eps": 1e-06,
488
+ "rope_local_base_freq": 10000.0,
489
+ "rope_scaling": null,
490
+ "rope_theta": 1000000.0,
491
+ "sliding_window": 512,
492
+ "torch_dtype": "bfloat16",
493
+ "use_cache": true,
494
+ "vocab_size": 262400,
495
+ "vocab_size_per_layer_input": 262144
496
+ },
497
+ "torch_dtype": "bfloat16",
498
+ "transformers_version": "4.55.4",
499
+ "unsloth_fixed": true,
500
+ "unsloth_version": "2025.9.8",
501
+ "vision_config": {
502
+ "architecture": "mobilenetv5_300m_enc",
503
+ "do_pooling": false,
504
+ "hidden_size": 2048,
505
+ "initializer_range": 0.02,
506
+ "label_names": [
507
+ "LABEL_0",
508
+ "LABEL_1"
509
+ ],
510
+ "model_args": null,
511
+ "model_type": "gemma3n_vision",
512
+ "num_classes": 2,
513
+ "rms_norm_eps": 1e-06,
514
+ "torch_dtype": "bfloat16",
515
+ "vocab_offset": 262144,
516
+ "vocab_size": 128
517
+ },
518
+ "vision_soft_tokens_per_image": 256
519
+ }
generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "cache_implementation": "hybrid",
4
+ "do_sample": true,
5
+ "eos_token_id": 1,
6
+ "pad_token_id": 0,
7
+ "top_k": 64,
8
+ "top_p": 0.95,
9
+ "transformers_version": "4.55.4"
10
+ }
model-00001-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:efe7066086c27000adf6dd9c08a6da9ac23b018bc5c48ea063bd422507659690
3
+ size 3724566017
model-00002-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5780cf8716796c1bf5fcf330c359e0cd11655f001f94f10897dc1731002b661f
3
+ size 4987292124
model-00003-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:475750e6524864db3c0ffcd47f3dcb22fe6ee0e12dd5b5279c9d705db9e4eb5c
3
+ size 2223613264
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
preprocessor_config.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "crop_size": null,
3
+ "data_format": "channels_first",
4
+ "default_to_square": false,
5
+ "device": null,
6
+ "disable_grouping": null,
7
+ "dither": 0.0,
8
+ "do_center_crop": null,
9
+ "do_convert_rgb": null,
10
+ "do_normalize": false,
11
+ "do_rescale": true,
12
+ "do_resize": true,
13
+ "feature_size": 128,
14
+ "fft_length": 1024,
15
+ "fft_overdrive": true,
16
+ "frame_length": 512,
17
+ "hop_length": 160,
18
+ "image_mean": [
19
+ 0.5,
20
+ 0.5,
21
+ 0.5
22
+ ],
23
+ "image_processor_type": "SiglipImageProcessorFast",
24
+ "image_seq_length": 256,
25
+ "image_std": [
26
+ 0.5,
27
+ 0.5,
28
+ 0.5
29
+ ],
30
+ "input_data_format": null,
31
+ "input_scale_factor": 1.0,
32
+ "max_frequency": 7600.0,
33
+ "mel_floor": 1e-05,
34
+ "min_frequency": 125.0,
35
+ "padding_side": "left",
36
+ "padding_value": 0.0,
37
+ "per_bin_mean": null,
38
+ "per_bin_stddev": null,
39
+ "preemphasis": 0.97,
40
+ "preemphasis_htk_flavor": true,
41
+ "processor_class": "Gemma3nProcessor",
42
+ "resample": 2,
43
+ "rescale_factor": 0.00392156862745098,
44
+ "return_attention_mask": true,
45
+ "return_tensors": null,
46
+ "sampling_rate": 16000,
47
+ "size": {
48
+ "height": 768,
49
+ "width": 768
50
+ }
51
+ }
processor_config.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "audio_seq_length": 188,
3
+ "image_seq_length": 256,
4
+ "processor_class": "Gemma3nProcessor"
5
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_token": "<audio_soft_token>",
3
+ "boa_token": "<start_of_audio>",
4
+ "boi_token": "<start_of_image>",
5
+ "bos_token": {
6
+ "content": "<bos>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false
11
+ },
12
+ "eoa_token": "<end_of_audio>",
13
+ "eoi_token": "<end_of_image>",
14
+ "eos_token": {
15
+ "content": "<eos>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false
20
+ },
21
+ "image_token": "<image_soft_token>",
22
+ "pad_token": {
23
+ "content": "<pad>",
24
+ "lstrip": false,
25
+ "normalized": false,
26
+ "rstrip": false,
27
+ "single_word": false
28
+ },
29
+ "unk_token": {
30
+ "content": "<unk>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false
35
+ }
36
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34e3f436416642ff2b4cba0602975a87bd8766c9f3b1851dde39479073bafd08
3
+ size 33442807
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea5f0cc48abfbfc04d14562270a32e02149a3e7035f368cc5a462786f4a59961
3
+ size 4696020
tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff