Eldar Kurtic commited on
Commit
ffdad8a
·
unverified ·
1 Parent(s): a17cf32
.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": "{{- bos_token }}\n{%- if custom_tools is defined %}\n {%- set tools = custom_tools %}\n{%- endif %}\n{%- if not tools_in_user_message is defined %}\n {%- set tools_in_user_message = true %}\n{%- endif %}\n{%- if not date_string is defined %}\n {%- if strftime_now is defined %}\n {%- set date_string = strftime_now(\"%d %b %Y\") %}\n {%- else %}\n {%- set date_string = \"26 Jul 2024\" %}\n {%- endif %}\n{%- endif %}\n{%- if not tools is defined %}\n {%- set tools = none %}\n{%- endif %}\n\n{#- This block extracts the system message, so we can slot it into the right place. #}\n{%- if messages[0]['role'] == 'system' %} \n {%- if messages[0]['content'] is string %}\n {%- set system_message = messages[0]['content']|trim %}\n {%- else %}\n {#- FIXME: The processor requires an array, always. #}\n {%- set system_message = messages[0]['content'][0]['text']|trim %}\n {%- endif %}\n {%- set messages = messages[1:] %}\n {%- set user_supplied_system_message = true %}\n{%- else %}\n {%- set system_message = \"\" %}\n {%- set user_supplied_system_message = false %}\n{%- endif %}\n\n{#- System message if the user supplied one #}\n{%- if user_supplied_system_message %}\n {{- \"<|header_start|>system<|header_end|>\\n\\n\" }}\n {%- if tools is not none %}\n {{- \"Environment: ipython\\n\" }}\n {%- endif %}\n {%- if tools is not none and not tools_in_user_message %}\n {{- \"You have access to the following functions. To call a function, please respond with JSON for a function call.\" }}\n {{- 'Respond in the format {\"name\": function name, \"parameters\": dictionary of argument name and its value}.' }}\n {{- \"Do not use variables.\\n\\n\" }}\n {%- for t in tools %}\n {{- t | tojson(indent=4) }}\n {{- \"\\n\\n\" }}\n {%- endfor %}\n {%- endif %}\n {{- system_message }}\n {{- \"<|eot|>\" }}\n{%- endif %}\n\n{#- Custom tools are passed in a user message with some extra guidance #}\n{%- if tools_in_user_message and not tools is none %}\n {#- Extract the first user message so we can plug it in here #}\n {%- if messages | length != 0 %}\n {%- set first_user_message = messages[0]['content']|trim %}\n {%- set messages = messages[1:] %}\n {%- else %}\n {{- raise_exception(\"Cannot put tools in the first user message when there's no first user message!\") }}\n{%- endif %}\n {{- '<|header_start|>user<|header_end|>\\n\\n' -}}\n {{- \"Given the following functions, please respond with a JSON for a function call \" }}\n {{- \"with its proper arguments that best answers the given prompt.\\n\\n\" }}\n {{- 'Respond in the format {\"name\": function name, \"parameters\": dictionary of argument name and its value}.' }}\n {{- \"Do not use variables.\\n\\n\" }}\n {%- for t in tools %}\n {{- t | tojson(indent=4) }}\n {{- \"\\n\\n\" }}\n {%- endfor %}\n {{- first_user_message + \"<|eot|>\"}}\n{%- endif %}\n\n{%- for message in messages %}\n {%- if not (message.role == 'ipython' or message.role == 'tool' or 'tool_calls' in message) %}\n {{- '<|header_start|>' + message['role'] + '<|header_end|>\\n\\n' }}\n {%- if message['content'] is string %}\n {{- message['content'] }}\n {%- else %}\n {%- for content in message['content'] %}\n {%- if content['type'] == 'image' %}\n {{- '<|image|>' }}\n {%- elif content['type'] == 'text' %}\n {{- content['text'] }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- \"<|eot|>\" }}\n {%- elif 'tool_calls' in message and message.tool_calls|length > 0 %}\n {{- '<|header_start|>assistant<|header_end|>\\n\\n' -}}\n {{- '<|python_start|>' }}\n {%- if message['content'] is string %}\n {{- message['content'] }}\n {%- else %}\n {%- for content in message['content'] %}\n {%- if content['type'] == 'image' %}\n {{- '<|image|>' }}\n {%- elif content['type'] == 'text' %}\n {{- content['text'] }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '<|python_end|>' }}\n {%- for tool_call in message.tool_calls %}\n {{- '{\"name\": \"' + tool_call.function.name + '\", ' }}\n {{- '\"parameters\": ' }}\n {{- tool_call.function.arguments | tojson }}\n {{- \"}\" }}\n {%- endfor %}\n {{- \"<|eot|>\" }}\n {%- elif message.role == \"tool\" or message.role == \"ipython\" %}\n {{- \"<|header_start|>ipython<|header_end|>\\n\\n\" }}\n {%- if message.content is mapping or message.content is iterable %}\n {{- message.content | tojson }}\n {%- else %}\n {{- message.content }}\n {%- endif %}\n {{- \"<|eot|>\" }}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|header_start|>assistant<|header_end|>\\n\\n' }}\n{%- endif %}\n"
3
+ }
config.json ADDED
@@ -0,0 +1,714 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Llama4ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 200080,
6
+ "eoi_token_index": 200081,
7
+ "image_token_index": 200092,
8
+ "model_type": "llama4",
9
+ "text_config": {
10
+ "_attn_implementation_autoset": true,
11
+ "attention_bias": false,
12
+ "attention_chunk_size": 8192,
13
+ "attention_dropout": 0.0,
14
+ "bos_token_id": 200000,
15
+ "eos_token_id": [
16
+ 200001,
17
+ 200007,
18
+ 200008
19
+ ],
20
+ "for_llm_compressor": true,
21
+ "head_dim": 128,
22
+ "hidden_act": "silu",
23
+ "hidden_size": 5120,
24
+ "initializer_range": 0.02,
25
+ "interleave_moe_layer_step": 1,
26
+ "intermediate_size": 8192,
27
+ "intermediate_size_mlp": 16384,
28
+ "max_position_embeddings": 10485760,
29
+ "model_type": "llama4_text",
30
+ "no_rope_layers": [],
31
+ "num_attention_heads": 40,
32
+ "num_experts_per_tok": 1,
33
+ "num_hidden_layers": 48,
34
+ "num_key_value_heads": 8,
35
+ "num_local_experts": 16,
36
+ "output_router_logits": false,
37
+ "pad_token_id": 200018,
38
+ "rms_norm_eps": 1e-05,
39
+ "rope_scaling": {
40
+ "factor": 16.0,
41
+ "high_freq_factor": 1.0,
42
+ "low_freq_factor": 1.0,
43
+ "original_max_position_embeddings": 8192,
44
+ "rope_type": "llama3"
45
+ },
46
+ "rope_theta": 500000.0,
47
+ "router_aux_loss_coef": 0.001,
48
+ "router_jitter_noise": 0.0,
49
+ "torch_dtype": "bfloat16",
50
+ "use_cache": true,
51
+ "use_qk_norm": true,
52
+ "vocab_size": 202048
53
+ },
54
+ "torch_dtype": "bfloat16",
55
+ "transformers_version": "4.51.0.dev0",
56
+ "vision_config": {
57
+ "_attn_implementation_autoset": true,
58
+ "attention_dropout": 0.0,
59
+ "hidden_act": "gelu",
60
+ "hidden_size": 1408,
61
+ "image_size": 336,
62
+ "initializer_range": 0.02,
63
+ "intermediate_size": 5632,
64
+ "model_type": "llama4_vision_model",
65
+ "multi_modal_projector_bias": false,
66
+ "norm_eps": 1e-05,
67
+ "num_attention_heads": 16,
68
+ "num_channels": 3,
69
+ "num_hidden_layers": 34,
70
+ "patch_size": 14,
71
+ "pixel_shuffle_ratio": 0.5,
72
+ "projector_dropout": 0.0,
73
+ "projector_input_dim": 4096,
74
+ "projector_output_dim": 4096,
75
+ "rope_theta": 10000,
76
+ "vision_feature_layer": -1,
77
+ "vision_feature_select_strategy": "default",
78
+ "vision_output_dim": 4096
79
+ },
80
+ "quantization_config": {
81
+ "config_groups": {
82
+ "group_0": {
83
+ "input_activations": {
84
+ "actorder": null,
85
+ "block_structure": null,
86
+ "dynamic": true,
87
+ "group_size": null,
88
+ "num_bits": 8,
89
+ "observer": null,
90
+ "observer_kwargs": {},
91
+ "strategy": "token",
92
+ "symmetric": true,
93
+ "type": "float"
94
+ },
95
+ "output_activations": null,
96
+ "targets": [
97
+ "Linear"
98
+ ],
99
+ "weights": {
100
+ "actorder": null,
101
+ "block_structure": null,
102
+ "dynamic": false,
103
+ "group_size": null,
104
+ "num_bits": 8,
105
+ "observer": "mse",
106
+ "observer_kwargs": {},
107
+ "strategy": "channel",
108
+ "symmetric": true,
109
+ "type": "float"
110
+ }
111
+ }
112
+ },
113
+ "format": "float-quantized",
114
+ "global_compression_ratio": null,
115
+ "ignore": [
116
+ "vision_model.patch_embedding.linear",
117
+ "vision_model.model.layers.0.self_attn.q_proj",
118
+ "vision_model.model.layers.0.self_attn.k_proj",
119
+ "vision_model.model.layers.0.self_attn.v_proj",
120
+ "vision_model.model.layers.0.self_attn.o_proj",
121
+ "vision_model.model.layers.0.mlp.fc1",
122
+ "vision_model.model.layers.0.mlp.fc2",
123
+ "vision_model.model.layers.1.self_attn.q_proj",
124
+ "vision_model.model.layers.1.self_attn.k_proj",
125
+ "vision_model.model.layers.1.self_attn.v_proj",
126
+ "vision_model.model.layers.1.self_attn.o_proj",
127
+ "vision_model.model.layers.1.mlp.fc1",
128
+ "vision_model.model.layers.1.mlp.fc2",
129
+ "vision_model.model.layers.2.self_attn.q_proj",
130
+ "vision_model.model.layers.2.self_attn.k_proj",
131
+ "vision_model.model.layers.2.self_attn.v_proj",
132
+ "vision_model.model.layers.2.self_attn.o_proj",
133
+ "vision_model.model.layers.2.mlp.fc1",
134
+ "vision_model.model.layers.2.mlp.fc2",
135
+ "vision_model.model.layers.3.self_attn.q_proj",
136
+ "vision_model.model.layers.3.self_attn.k_proj",
137
+ "vision_model.model.layers.3.self_attn.v_proj",
138
+ "vision_model.model.layers.3.self_attn.o_proj",
139
+ "vision_model.model.layers.3.mlp.fc1",
140
+ "vision_model.model.layers.3.mlp.fc2",
141
+ "vision_model.model.layers.4.self_attn.q_proj",
142
+ "vision_model.model.layers.4.self_attn.k_proj",
143
+ "vision_model.model.layers.4.self_attn.v_proj",
144
+ "vision_model.model.layers.4.self_attn.o_proj",
145
+ "vision_model.model.layers.4.mlp.fc1",
146
+ "vision_model.model.layers.4.mlp.fc2",
147
+ "vision_model.model.layers.5.self_attn.q_proj",
148
+ "vision_model.model.layers.5.self_attn.k_proj",
149
+ "vision_model.model.layers.5.self_attn.v_proj",
150
+ "vision_model.model.layers.5.self_attn.o_proj",
151
+ "vision_model.model.layers.5.mlp.fc1",
152
+ "vision_model.model.layers.5.mlp.fc2",
153
+ "vision_model.model.layers.6.self_attn.q_proj",
154
+ "vision_model.model.layers.6.self_attn.k_proj",
155
+ "vision_model.model.layers.6.self_attn.v_proj",
156
+ "vision_model.model.layers.6.self_attn.o_proj",
157
+ "vision_model.model.layers.6.mlp.fc1",
158
+ "vision_model.model.layers.6.mlp.fc2",
159
+ "vision_model.model.layers.7.self_attn.q_proj",
160
+ "vision_model.model.layers.7.self_attn.k_proj",
161
+ "vision_model.model.layers.7.self_attn.v_proj",
162
+ "vision_model.model.layers.7.self_attn.o_proj",
163
+ "vision_model.model.layers.7.mlp.fc1",
164
+ "vision_model.model.layers.7.mlp.fc2",
165
+ "vision_model.model.layers.8.self_attn.q_proj",
166
+ "vision_model.model.layers.8.self_attn.k_proj",
167
+ "vision_model.model.layers.8.self_attn.v_proj",
168
+ "vision_model.model.layers.8.self_attn.o_proj",
169
+ "vision_model.model.layers.8.mlp.fc1",
170
+ "vision_model.model.layers.8.mlp.fc2",
171
+ "vision_model.model.layers.9.self_attn.q_proj",
172
+ "vision_model.model.layers.9.self_attn.k_proj",
173
+ "vision_model.model.layers.9.self_attn.v_proj",
174
+ "vision_model.model.layers.9.self_attn.o_proj",
175
+ "vision_model.model.layers.9.mlp.fc1",
176
+ "vision_model.model.layers.9.mlp.fc2",
177
+ "vision_model.model.layers.10.self_attn.q_proj",
178
+ "vision_model.model.layers.10.self_attn.k_proj",
179
+ "vision_model.model.layers.10.self_attn.v_proj",
180
+ "vision_model.model.layers.10.self_attn.o_proj",
181
+ "vision_model.model.layers.10.mlp.fc1",
182
+ "vision_model.model.layers.10.mlp.fc2",
183
+ "vision_model.model.layers.11.self_attn.q_proj",
184
+ "vision_model.model.layers.11.self_attn.k_proj",
185
+ "vision_model.model.layers.11.self_attn.v_proj",
186
+ "vision_model.model.layers.11.self_attn.o_proj",
187
+ "vision_model.model.layers.11.mlp.fc1",
188
+ "vision_model.model.layers.11.mlp.fc2",
189
+ "vision_model.model.layers.12.self_attn.q_proj",
190
+ "vision_model.model.layers.12.self_attn.k_proj",
191
+ "vision_model.model.layers.12.self_attn.v_proj",
192
+ "vision_model.model.layers.12.self_attn.o_proj",
193
+ "vision_model.model.layers.12.mlp.fc1",
194
+ "vision_model.model.layers.12.mlp.fc2",
195
+ "vision_model.model.layers.13.self_attn.q_proj",
196
+ "vision_model.model.layers.13.self_attn.k_proj",
197
+ "vision_model.model.layers.13.self_attn.v_proj",
198
+ "vision_model.model.layers.13.self_attn.o_proj",
199
+ "vision_model.model.layers.13.mlp.fc1",
200
+ "vision_model.model.layers.13.mlp.fc2",
201
+ "vision_model.model.layers.14.self_attn.q_proj",
202
+ "vision_model.model.layers.14.self_attn.k_proj",
203
+ "vision_model.model.layers.14.self_attn.v_proj",
204
+ "vision_model.model.layers.14.self_attn.o_proj",
205
+ "vision_model.model.layers.14.mlp.fc1",
206
+ "vision_model.model.layers.14.mlp.fc2",
207
+ "vision_model.model.layers.15.self_attn.q_proj",
208
+ "vision_model.model.layers.15.self_attn.k_proj",
209
+ "vision_model.model.layers.15.self_attn.v_proj",
210
+ "vision_model.model.layers.15.self_attn.o_proj",
211
+ "vision_model.model.layers.15.mlp.fc1",
212
+ "vision_model.model.layers.15.mlp.fc2",
213
+ "vision_model.model.layers.16.self_attn.q_proj",
214
+ "vision_model.model.layers.16.self_attn.k_proj",
215
+ "vision_model.model.layers.16.self_attn.v_proj",
216
+ "vision_model.model.layers.16.self_attn.o_proj",
217
+ "vision_model.model.layers.16.mlp.fc1",
218
+ "vision_model.model.layers.16.mlp.fc2",
219
+ "vision_model.model.layers.17.self_attn.q_proj",
220
+ "vision_model.model.layers.17.self_attn.k_proj",
221
+ "vision_model.model.layers.17.self_attn.v_proj",
222
+ "vision_model.model.layers.17.self_attn.o_proj",
223
+ "vision_model.model.layers.17.mlp.fc1",
224
+ "vision_model.model.layers.17.mlp.fc2",
225
+ "vision_model.model.layers.18.self_attn.q_proj",
226
+ "vision_model.model.layers.18.self_attn.k_proj",
227
+ "vision_model.model.layers.18.self_attn.v_proj",
228
+ "vision_model.model.layers.18.self_attn.o_proj",
229
+ "vision_model.model.layers.18.mlp.fc1",
230
+ "vision_model.model.layers.18.mlp.fc2",
231
+ "vision_model.model.layers.19.self_attn.q_proj",
232
+ "vision_model.model.layers.19.self_attn.k_proj",
233
+ "vision_model.model.layers.19.self_attn.v_proj",
234
+ "vision_model.model.layers.19.self_attn.o_proj",
235
+ "vision_model.model.layers.19.mlp.fc1",
236
+ "vision_model.model.layers.19.mlp.fc2",
237
+ "vision_model.model.layers.20.self_attn.q_proj",
238
+ "vision_model.model.layers.20.self_attn.k_proj",
239
+ "vision_model.model.layers.20.self_attn.v_proj",
240
+ "vision_model.model.layers.20.self_attn.o_proj",
241
+ "vision_model.model.layers.20.mlp.fc1",
242
+ "vision_model.model.layers.20.mlp.fc2",
243
+ "vision_model.model.layers.21.self_attn.q_proj",
244
+ "vision_model.model.layers.21.self_attn.k_proj",
245
+ "vision_model.model.layers.21.self_attn.v_proj",
246
+ "vision_model.model.layers.21.self_attn.o_proj",
247
+ "vision_model.model.layers.21.mlp.fc1",
248
+ "vision_model.model.layers.21.mlp.fc2",
249
+ "vision_model.model.layers.22.self_attn.q_proj",
250
+ "vision_model.model.layers.22.self_attn.k_proj",
251
+ "vision_model.model.layers.22.self_attn.v_proj",
252
+ "vision_model.model.layers.22.self_attn.o_proj",
253
+ "vision_model.model.layers.22.mlp.fc1",
254
+ "vision_model.model.layers.22.mlp.fc2",
255
+ "vision_model.model.layers.23.self_attn.q_proj",
256
+ "vision_model.model.layers.23.self_attn.k_proj",
257
+ "vision_model.model.layers.23.self_attn.v_proj",
258
+ "vision_model.model.layers.23.self_attn.o_proj",
259
+ "vision_model.model.layers.23.mlp.fc1",
260
+ "vision_model.model.layers.23.mlp.fc2",
261
+ "vision_model.model.layers.24.self_attn.q_proj",
262
+ "vision_model.model.layers.24.self_attn.k_proj",
263
+ "vision_model.model.layers.24.self_attn.v_proj",
264
+ "vision_model.model.layers.24.self_attn.o_proj",
265
+ "vision_model.model.layers.24.mlp.fc1",
266
+ "vision_model.model.layers.24.mlp.fc2",
267
+ "vision_model.model.layers.25.self_attn.q_proj",
268
+ "vision_model.model.layers.25.self_attn.k_proj",
269
+ "vision_model.model.layers.25.self_attn.v_proj",
270
+ "vision_model.model.layers.25.self_attn.o_proj",
271
+ "vision_model.model.layers.25.mlp.fc1",
272
+ "vision_model.model.layers.25.mlp.fc2",
273
+ "vision_model.model.layers.26.self_attn.q_proj",
274
+ "vision_model.model.layers.26.self_attn.k_proj",
275
+ "vision_model.model.layers.26.self_attn.v_proj",
276
+ "vision_model.model.layers.26.self_attn.o_proj",
277
+ "vision_model.model.layers.26.mlp.fc1",
278
+ "vision_model.model.layers.26.mlp.fc2",
279
+ "vision_model.model.layers.27.self_attn.q_proj",
280
+ "vision_model.model.layers.27.self_attn.k_proj",
281
+ "vision_model.model.layers.27.self_attn.v_proj",
282
+ "vision_model.model.layers.27.self_attn.o_proj",
283
+ "vision_model.model.layers.27.mlp.fc1",
284
+ "vision_model.model.layers.27.mlp.fc2",
285
+ "vision_model.model.layers.28.self_attn.q_proj",
286
+ "vision_model.model.layers.28.self_attn.k_proj",
287
+ "vision_model.model.layers.28.self_attn.v_proj",
288
+ "vision_model.model.layers.28.self_attn.o_proj",
289
+ "vision_model.model.layers.28.mlp.fc1",
290
+ "vision_model.model.layers.28.mlp.fc2",
291
+ "vision_model.model.layers.29.self_attn.q_proj",
292
+ "vision_model.model.layers.29.self_attn.k_proj",
293
+ "vision_model.model.layers.29.self_attn.v_proj",
294
+ "vision_model.model.layers.29.self_attn.o_proj",
295
+ "vision_model.model.layers.29.mlp.fc1",
296
+ "vision_model.model.layers.29.mlp.fc2",
297
+ "vision_model.model.layers.30.self_attn.q_proj",
298
+ "vision_model.model.layers.30.self_attn.k_proj",
299
+ "vision_model.model.layers.30.self_attn.v_proj",
300
+ "vision_model.model.layers.30.self_attn.o_proj",
301
+ "vision_model.model.layers.30.mlp.fc1",
302
+ "vision_model.model.layers.30.mlp.fc2",
303
+ "vision_model.model.layers.31.self_attn.q_proj",
304
+ "vision_model.model.layers.31.self_attn.k_proj",
305
+ "vision_model.model.layers.31.self_attn.v_proj",
306
+ "vision_model.model.layers.31.self_attn.o_proj",
307
+ "vision_model.model.layers.31.mlp.fc1",
308
+ "vision_model.model.layers.31.mlp.fc2",
309
+ "vision_model.model.layers.32.self_attn.q_proj",
310
+ "vision_model.model.layers.32.self_attn.k_proj",
311
+ "vision_model.model.layers.32.self_attn.v_proj",
312
+ "vision_model.model.layers.32.self_attn.o_proj",
313
+ "vision_model.model.layers.32.mlp.fc1",
314
+ "vision_model.model.layers.32.mlp.fc2",
315
+ "vision_model.model.layers.33.self_attn.q_proj",
316
+ "vision_model.model.layers.33.self_attn.k_proj",
317
+ "vision_model.model.layers.33.self_attn.v_proj",
318
+ "vision_model.model.layers.33.self_attn.o_proj",
319
+ "vision_model.model.layers.33.mlp.fc1",
320
+ "vision_model.model.layers.33.mlp.fc2",
321
+ "vision_model.vision_adapter.mlp.fc1",
322
+ "vision_model.vision_adapter.mlp.fc2",
323
+ "multi_modal_projector.linear_1",
324
+ "language_model.model.layers.0.self_attn.q_proj",
325
+ "language_model.model.layers.0.self_attn.k_proj",
326
+ "language_model.model.layers.0.self_attn.v_proj",
327
+ "language_model.model.layers.0.self_attn.o_proj",
328
+ "language_model.model.layers.0.feed_forward.router",
329
+ "language_model.model.layers.0.feed_forward.shared_expert.gate_proj",
330
+ "language_model.model.layers.0.feed_forward.shared_expert.up_proj",
331
+ "language_model.model.layers.0.feed_forward.shared_expert.down_proj",
332
+ "language_model.model.layers.1.self_attn.q_proj",
333
+ "language_model.model.layers.1.self_attn.k_proj",
334
+ "language_model.model.layers.1.self_attn.v_proj",
335
+ "language_model.model.layers.1.self_attn.o_proj",
336
+ "language_model.model.layers.1.feed_forward.router",
337
+ "language_model.model.layers.1.feed_forward.shared_expert.gate_proj",
338
+ "language_model.model.layers.1.feed_forward.shared_expert.up_proj",
339
+ "language_model.model.layers.1.feed_forward.shared_expert.down_proj",
340
+ "language_model.model.layers.2.self_attn.q_proj",
341
+ "language_model.model.layers.2.self_attn.k_proj",
342
+ "language_model.model.layers.2.self_attn.v_proj",
343
+ "language_model.model.layers.2.self_attn.o_proj",
344
+ "language_model.model.layers.2.feed_forward.router",
345
+ "language_model.model.layers.2.feed_forward.shared_expert.gate_proj",
346
+ "language_model.model.layers.2.feed_forward.shared_expert.up_proj",
347
+ "language_model.model.layers.2.feed_forward.shared_expert.down_proj",
348
+ "language_model.model.layers.3.self_attn.q_proj",
349
+ "language_model.model.layers.3.self_attn.k_proj",
350
+ "language_model.model.layers.3.self_attn.v_proj",
351
+ "language_model.model.layers.3.self_attn.o_proj",
352
+ "language_model.model.layers.3.feed_forward.router",
353
+ "language_model.model.layers.3.feed_forward.shared_expert.gate_proj",
354
+ "language_model.model.layers.3.feed_forward.shared_expert.up_proj",
355
+ "language_model.model.layers.3.feed_forward.shared_expert.down_proj",
356
+ "language_model.model.layers.4.self_attn.q_proj",
357
+ "language_model.model.layers.4.self_attn.k_proj",
358
+ "language_model.model.layers.4.self_attn.v_proj",
359
+ "language_model.model.layers.4.self_attn.o_proj",
360
+ "language_model.model.layers.4.feed_forward.router",
361
+ "language_model.model.layers.4.feed_forward.shared_expert.gate_proj",
362
+ "language_model.model.layers.4.feed_forward.shared_expert.up_proj",
363
+ "language_model.model.layers.4.feed_forward.shared_expert.down_proj",
364
+ "language_model.model.layers.5.self_attn.q_proj",
365
+ "language_model.model.layers.5.self_attn.k_proj",
366
+ "language_model.model.layers.5.self_attn.v_proj",
367
+ "language_model.model.layers.5.self_attn.o_proj",
368
+ "language_model.model.layers.5.feed_forward.router",
369
+ "language_model.model.layers.5.feed_forward.shared_expert.gate_proj",
370
+ "language_model.model.layers.5.feed_forward.shared_expert.up_proj",
371
+ "language_model.model.layers.5.feed_forward.shared_expert.down_proj",
372
+ "language_model.model.layers.6.self_attn.q_proj",
373
+ "language_model.model.layers.6.self_attn.k_proj",
374
+ "language_model.model.layers.6.self_attn.v_proj",
375
+ "language_model.model.layers.6.self_attn.o_proj",
376
+ "language_model.model.layers.6.feed_forward.router",
377
+ "language_model.model.layers.6.feed_forward.shared_expert.gate_proj",
378
+ "language_model.model.layers.6.feed_forward.shared_expert.up_proj",
379
+ "language_model.model.layers.6.feed_forward.shared_expert.down_proj",
380
+ "language_model.model.layers.7.self_attn.q_proj",
381
+ "language_model.model.layers.7.self_attn.k_proj",
382
+ "language_model.model.layers.7.self_attn.v_proj",
383
+ "language_model.model.layers.7.self_attn.o_proj",
384
+ "language_model.model.layers.7.feed_forward.router",
385
+ "language_model.model.layers.7.feed_forward.shared_expert.gate_proj",
386
+ "language_model.model.layers.7.feed_forward.shared_expert.up_proj",
387
+ "language_model.model.layers.7.feed_forward.shared_expert.down_proj",
388
+ "language_model.model.layers.8.self_attn.q_proj",
389
+ "language_model.model.layers.8.self_attn.k_proj",
390
+ "language_model.model.layers.8.self_attn.v_proj",
391
+ "language_model.model.layers.8.self_attn.o_proj",
392
+ "language_model.model.layers.8.feed_forward.router",
393
+ "language_model.model.layers.8.feed_forward.shared_expert.gate_proj",
394
+ "language_model.model.layers.8.feed_forward.shared_expert.up_proj",
395
+ "language_model.model.layers.8.feed_forward.shared_expert.down_proj",
396
+ "language_model.model.layers.9.self_attn.q_proj",
397
+ "language_model.model.layers.9.self_attn.k_proj",
398
+ "language_model.model.layers.9.self_attn.v_proj",
399
+ "language_model.model.layers.9.self_attn.o_proj",
400
+ "language_model.model.layers.9.feed_forward.router",
401
+ "language_model.model.layers.9.feed_forward.shared_expert.gate_proj",
402
+ "language_model.model.layers.9.feed_forward.shared_expert.up_proj",
403
+ "language_model.model.layers.9.feed_forward.shared_expert.down_proj",
404
+ "language_model.model.layers.10.self_attn.q_proj",
405
+ "language_model.model.layers.10.self_attn.k_proj",
406
+ "language_model.model.layers.10.self_attn.v_proj",
407
+ "language_model.model.layers.10.self_attn.o_proj",
408
+ "language_model.model.layers.10.feed_forward.router",
409
+ "language_model.model.layers.10.feed_forward.shared_expert.gate_proj",
410
+ "language_model.model.layers.10.feed_forward.shared_expert.up_proj",
411
+ "language_model.model.layers.10.feed_forward.shared_expert.down_proj",
412
+ "language_model.model.layers.11.self_attn.q_proj",
413
+ "language_model.model.layers.11.self_attn.k_proj",
414
+ "language_model.model.layers.11.self_attn.v_proj",
415
+ "language_model.model.layers.11.self_attn.o_proj",
416
+ "language_model.model.layers.11.feed_forward.router",
417
+ "language_model.model.layers.11.feed_forward.shared_expert.gate_proj",
418
+ "language_model.model.layers.11.feed_forward.shared_expert.up_proj",
419
+ "language_model.model.layers.11.feed_forward.shared_expert.down_proj",
420
+ "language_model.model.layers.12.self_attn.q_proj",
421
+ "language_model.model.layers.12.self_attn.k_proj",
422
+ "language_model.model.layers.12.self_attn.v_proj",
423
+ "language_model.model.layers.12.self_attn.o_proj",
424
+ "language_model.model.layers.12.feed_forward.router",
425
+ "language_model.model.layers.12.feed_forward.shared_expert.gate_proj",
426
+ "language_model.model.layers.12.feed_forward.shared_expert.up_proj",
427
+ "language_model.model.layers.12.feed_forward.shared_expert.down_proj",
428
+ "language_model.model.layers.13.self_attn.q_proj",
429
+ "language_model.model.layers.13.self_attn.k_proj",
430
+ "language_model.model.layers.13.self_attn.v_proj",
431
+ "language_model.model.layers.13.self_attn.o_proj",
432
+ "language_model.model.layers.13.feed_forward.router",
433
+ "language_model.model.layers.13.feed_forward.shared_expert.gate_proj",
434
+ "language_model.model.layers.13.feed_forward.shared_expert.up_proj",
435
+ "language_model.model.layers.13.feed_forward.shared_expert.down_proj",
436
+ "language_model.model.layers.14.self_attn.q_proj",
437
+ "language_model.model.layers.14.self_attn.k_proj",
438
+ "language_model.model.layers.14.self_attn.v_proj",
439
+ "language_model.model.layers.14.self_attn.o_proj",
440
+ "language_model.model.layers.14.feed_forward.router",
441
+ "language_model.model.layers.14.feed_forward.shared_expert.gate_proj",
442
+ "language_model.model.layers.14.feed_forward.shared_expert.up_proj",
443
+ "language_model.model.layers.14.feed_forward.shared_expert.down_proj",
444
+ "language_model.model.layers.15.self_attn.q_proj",
445
+ "language_model.model.layers.15.self_attn.k_proj",
446
+ "language_model.model.layers.15.self_attn.v_proj",
447
+ "language_model.model.layers.15.self_attn.o_proj",
448
+ "language_model.model.layers.15.feed_forward.router",
449
+ "language_model.model.layers.15.feed_forward.shared_expert.gate_proj",
450
+ "language_model.model.layers.15.feed_forward.shared_expert.up_proj",
451
+ "language_model.model.layers.15.feed_forward.shared_expert.down_proj",
452
+ "language_model.model.layers.16.self_attn.q_proj",
453
+ "language_model.model.layers.16.self_attn.k_proj",
454
+ "language_model.model.layers.16.self_attn.v_proj",
455
+ "language_model.model.layers.16.self_attn.o_proj",
456
+ "language_model.model.layers.16.feed_forward.router",
457
+ "language_model.model.layers.16.feed_forward.shared_expert.gate_proj",
458
+ "language_model.model.layers.16.feed_forward.shared_expert.up_proj",
459
+ "language_model.model.layers.16.feed_forward.shared_expert.down_proj",
460
+ "language_model.model.layers.17.self_attn.q_proj",
461
+ "language_model.model.layers.17.self_attn.k_proj",
462
+ "language_model.model.layers.17.self_attn.v_proj",
463
+ "language_model.model.layers.17.self_attn.o_proj",
464
+ "language_model.model.layers.17.feed_forward.router",
465
+ "language_model.model.layers.17.feed_forward.shared_expert.gate_proj",
466
+ "language_model.model.layers.17.feed_forward.shared_expert.up_proj",
467
+ "language_model.model.layers.17.feed_forward.shared_expert.down_proj",
468
+ "language_model.model.layers.18.self_attn.q_proj",
469
+ "language_model.model.layers.18.self_attn.k_proj",
470
+ "language_model.model.layers.18.self_attn.v_proj",
471
+ "language_model.model.layers.18.self_attn.o_proj",
472
+ "language_model.model.layers.18.feed_forward.router",
473
+ "language_model.model.layers.18.feed_forward.shared_expert.gate_proj",
474
+ "language_model.model.layers.18.feed_forward.shared_expert.up_proj",
475
+ "language_model.model.layers.18.feed_forward.shared_expert.down_proj",
476
+ "language_model.model.layers.19.self_attn.q_proj",
477
+ "language_model.model.layers.19.self_attn.k_proj",
478
+ "language_model.model.layers.19.self_attn.v_proj",
479
+ "language_model.model.layers.19.self_attn.o_proj",
480
+ "language_model.model.layers.19.feed_forward.router",
481
+ "language_model.model.layers.19.feed_forward.shared_expert.gate_proj",
482
+ "language_model.model.layers.19.feed_forward.shared_expert.up_proj",
483
+ "language_model.model.layers.19.feed_forward.shared_expert.down_proj",
484
+ "language_model.model.layers.20.self_attn.q_proj",
485
+ "language_model.model.layers.20.self_attn.k_proj",
486
+ "language_model.model.layers.20.self_attn.v_proj",
487
+ "language_model.model.layers.20.self_attn.o_proj",
488
+ "language_model.model.layers.20.feed_forward.router",
489
+ "language_model.model.layers.20.feed_forward.shared_expert.gate_proj",
490
+ "language_model.model.layers.20.feed_forward.shared_expert.up_proj",
491
+ "language_model.model.layers.20.feed_forward.shared_expert.down_proj",
492
+ "language_model.model.layers.21.self_attn.q_proj",
493
+ "language_model.model.layers.21.self_attn.k_proj",
494
+ "language_model.model.layers.21.self_attn.v_proj",
495
+ "language_model.model.layers.21.self_attn.o_proj",
496
+ "language_model.model.layers.21.feed_forward.router",
497
+ "language_model.model.layers.21.feed_forward.shared_expert.gate_proj",
498
+ "language_model.model.layers.21.feed_forward.shared_expert.up_proj",
499
+ "language_model.model.layers.21.feed_forward.shared_expert.down_proj",
500
+ "language_model.model.layers.22.self_attn.q_proj",
501
+ "language_model.model.layers.22.self_attn.k_proj",
502
+ "language_model.model.layers.22.self_attn.v_proj",
503
+ "language_model.model.layers.22.self_attn.o_proj",
504
+ "language_model.model.layers.22.feed_forward.router",
505
+ "language_model.model.layers.22.feed_forward.shared_expert.gate_proj",
506
+ "language_model.model.layers.22.feed_forward.shared_expert.up_proj",
507
+ "language_model.model.layers.22.feed_forward.shared_expert.down_proj",
508
+ "language_model.model.layers.23.self_attn.q_proj",
509
+ "language_model.model.layers.23.self_attn.k_proj",
510
+ "language_model.model.layers.23.self_attn.v_proj",
511
+ "language_model.model.layers.23.self_attn.o_proj",
512
+ "language_model.model.layers.23.feed_forward.router",
513
+ "language_model.model.layers.23.feed_forward.shared_expert.gate_proj",
514
+ "language_model.model.layers.23.feed_forward.shared_expert.up_proj",
515
+ "language_model.model.layers.23.feed_forward.shared_expert.down_proj",
516
+ "language_model.model.layers.24.self_attn.q_proj",
517
+ "language_model.model.layers.24.self_attn.k_proj",
518
+ "language_model.model.layers.24.self_attn.v_proj",
519
+ "language_model.model.layers.24.self_attn.o_proj",
520
+ "language_model.model.layers.24.feed_forward.router",
521
+ "language_model.model.layers.24.feed_forward.shared_expert.gate_proj",
522
+ "language_model.model.layers.24.feed_forward.shared_expert.up_proj",
523
+ "language_model.model.layers.24.feed_forward.shared_expert.down_proj",
524
+ "language_model.model.layers.25.self_attn.q_proj",
525
+ "language_model.model.layers.25.self_attn.k_proj",
526
+ "language_model.model.layers.25.self_attn.v_proj",
527
+ "language_model.model.layers.25.self_attn.o_proj",
528
+ "language_model.model.layers.25.feed_forward.router",
529
+ "language_model.model.layers.25.feed_forward.shared_expert.gate_proj",
530
+ "language_model.model.layers.25.feed_forward.shared_expert.up_proj",
531
+ "language_model.model.layers.25.feed_forward.shared_expert.down_proj",
532
+ "language_model.model.layers.26.self_attn.q_proj",
533
+ "language_model.model.layers.26.self_attn.k_proj",
534
+ "language_model.model.layers.26.self_attn.v_proj",
535
+ "language_model.model.layers.26.self_attn.o_proj",
536
+ "language_model.model.layers.26.feed_forward.router",
537
+ "language_model.model.layers.26.feed_forward.shared_expert.gate_proj",
538
+ "language_model.model.layers.26.feed_forward.shared_expert.up_proj",
539
+ "language_model.model.layers.26.feed_forward.shared_expert.down_proj",
540
+ "language_model.model.layers.27.self_attn.q_proj",
541
+ "language_model.model.layers.27.self_attn.k_proj",
542
+ "language_model.model.layers.27.self_attn.v_proj",
543
+ "language_model.model.layers.27.self_attn.o_proj",
544
+ "language_model.model.layers.27.feed_forward.router",
545
+ "language_model.model.layers.27.feed_forward.shared_expert.gate_proj",
546
+ "language_model.model.layers.27.feed_forward.shared_expert.up_proj",
547
+ "language_model.model.layers.27.feed_forward.shared_expert.down_proj",
548
+ "language_model.model.layers.28.self_attn.q_proj",
549
+ "language_model.model.layers.28.self_attn.k_proj",
550
+ "language_model.model.layers.28.self_attn.v_proj",
551
+ "language_model.model.layers.28.self_attn.o_proj",
552
+ "language_model.model.layers.28.feed_forward.router",
553
+ "language_model.model.layers.28.feed_forward.shared_expert.gate_proj",
554
+ "language_model.model.layers.28.feed_forward.shared_expert.up_proj",
555
+ "language_model.model.layers.28.feed_forward.shared_expert.down_proj",
556
+ "language_model.model.layers.29.self_attn.q_proj",
557
+ "language_model.model.layers.29.self_attn.k_proj",
558
+ "language_model.model.layers.29.self_attn.v_proj",
559
+ "language_model.model.layers.29.self_attn.o_proj",
560
+ "language_model.model.layers.29.feed_forward.router",
561
+ "language_model.model.layers.29.feed_forward.shared_expert.gate_proj",
562
+ "language_model.model.layers.29.feed_forward.shared_expert.up_proj",
563
+ "language_model.model.layers.29.feed_forward.shared_expert.down_proj",
564
+ "language_model.model.layers.30.self_attn.q_proj",
565
+ "language_model.model.layers.30.self_attn.k_proj",
566
+ "language_model.model.layers.30.self_attn.v_proj",
567
+ "language_model.model.layers.30.self_attn.o_proj",
568
+ "language_model.model.layers.30.feed_forward.router",
569
+ "language_model.model.layers.30.feed_forward.shared_expert.gate_proj",
570
+ "language_model.model.layers.30.feed_forward.shared_expert.up_proj",
571
+ "language_model.model.layers.30.feed_forward.shared_expert.down_proj",
572
+ "language_model.model.layers.31.self_attn.q_proj",
573
+ "language_model.model.layers.31.self_attn.k_proj",
574
+ "language_model.model.layers.31.self_attn.v_proj",
575
+ "language_model.model.layers.31.self_attn.o_proj",
576
+ "language_model.model.layers.31.feed_forward.router",
577
+ "language_model.model.layers.31.feed_forward.shared_expert.gate_proj",
578
+ "language_model.model.layers.31.feed_forward.shared_expert.up_proj",
579
+ "language_model.model.layers.31.feed_forward.shared_expert.down_proj",
580
+ "language_model.model.layers.32.self_attn.q_proj",
581
+ "language_model.model.layers.32.self_attn.k_proj",
582
+ "language_model.model.layers.32.self_attn.v_proj",
583
+ "language_model.model.layers.32.self_attn.o_proj",
584
+ "language_model.model.layers.32.feed_forward.router",
585
+ "language_model.model.layers.32.feed_forward.shared_expert.gate_proj",
586
+ "language_model.model.layers.32.feed_forward.shared_expert.up_proj",
587
+ "language_model.model.layers.32.feed_forward.shared_expert.down_proj",
588
+ "language_model.model.layers.33.self_attn.q_proj",
589
+ "language_model.model.layers.33.self_attn.k_proj",
590
+ "language_model.model.layers.33.self_attn.v_proj",
591
+ "language_model.model.layers.33.self_attn.o_proj",
592
+ "language_model.model.layers.33.feed_forward.router",
593
+ "language_model.model.layers.33.feed_forward.shared_expert.gate_proj",
594
+ "language_model.model.layers.33.feed_forward.shared_expert.up_proj",
595
+ "language_model.model.layers.33.feed_forward.shared_expert.down_proj",
596
+ "language_model.model.layers.34.self_attn.q_proj",
597
+ "language_model.model.layers.34.self_attn.k_proj",
598
+ "language_model.model.layers.34.self_attn.v_proj",
599
+ "language_model.model.layers.34.self_attn.o_proj",
600
+ "language_model.model.layers.34.feed_forward.router",
601
+ "language_model.model.layers.34.feed_forward.shared_expert.gate_proj",
602
+ "language_model.model.layers.34.feed_forward.shared_expert.up_proj",
603
+ "language_model.model.layers.34.feed_forward.shared_expert.down_proj",
604
+ "language_model.model.layers.35.self_attn.q_proj",
605
+ "language_model.model.layers.35.self_attn.k_proj",
606
+ "language_model.model.layers.35.self_attn.v_proj",
607
+ "language_model.model.layers.35.self_attn.o_proj",
608
+ "language_model.model.layers.35.feed_forward.router",
609
+ "language_model.model.layers.35.feed_forward.shared_expert.gate_proj",
610
+ "language_model.model.layers.35.feed_forward.shared_expert.up_proj",
611
+ "language_model.model.layers.35.feed_forward.shared_expert.down_proj",
612
+ "language_model.model.layers.36.self_attn.q_proj",
613
+ "language_model.model.layers.36.self_attn.k_proj",
614
+ "language_model.model.layers.36.self_attn.v_proj",
615
+ "language_model.model.layers.36.self_attn.o_proj",
616
+ "language_model.model.layers.36.feed_forward.router",
617
+ "language_model.model.layers.36.feed_forward.shared_expert.gate_proj",
618
+ "language_model.model.layers.36.feed_forward.shared_expert.up_proj",
619
+ "language_model.model.layers.36.feed_forward.shared_expert.down_proj",
620
+ "language_model.model.layers.37.self_attn.q_proj",
621
+ "language_model.model.layers.37.self_attn.k_proj",
622
+ "language_model.model.layers.37.self_attn.v_proj",
623
+ "language_model.model.layers.37.self_attn.o_proj",
624
+ "language_model.model.layers.37.feed_forward.router",
625
+ "language_model.model.layers.37.feed_forward.shared_expert.gate_proj",
626
+ "language_model.model.layers.37.feed_forward.shared_expert.up_proj",
627
+ "language_model.model.layers.37.feed_forward.shared_expert.down_proj",
628
+ "language_model.model.layers.38.self_attn.q_proj",
629
+ "language_model.model.layers.38.self_attn.k_proj",
630
+ "language_model.model.layers.38.self_attn.v_proj",
631
+ "language_model.model.layers.38.self_attn.o_proj",
632
+ "language_model.model.layers.38.feed_forward.router",
633
+ "language_model.model.layers.38.feed_forward.shared_expert.gate_proj",
634
+ "language_model.model.layers.38.feed_forward.shared_expert.up_proj",
635
+ "language_model.model.layers.38.feed_forward.shared_expert.down_proj",
636
+ "language_model.model.layers.39.self_attn.q_proj",
637
+ "language_model.model.layers.39.self_attn.k_proj",
638
+ "language_model.model.layers.39.self_attn.v_proj",
639
+ "language_model.model.layers.39.self_attn.o_proj",
640
+ "language_model.model.layers.39.feed_forward.router",
641
+ "language_model.model.layers.39.feed_forward.shared_expert.gate_proj",
642
+ "language_model.model.layers.39.feed_forward.shared_expert.up_proj",
643
+ "language_model.model.layers.39.feed_forward.shared_expert.down_proj",
644
+ "language_model.model.layers.40.self_attn.q_proj",
645
+ "language_model.model.layers.40.self_attn.k_proj",
646
+ "language_model.model.layers.40.self_attn.v_proj",
647
+ "language_model.model.layers.40.self_attn.o_proj",
648
+ "language_model.model.layers.40.feed_forward.router",
649
+ "language_model.model.layers.40.feed_forward.shared_expert.gate_proj",
650
+ "language_model.model.layers.40.feed_forward.shared_expert.up_proj",
651
+ "language_model.model.layers.40.feed_forward.shared_expert.down_proj",
652
+ "language_model.model.layers.41.self_attn.q_proj",
653
+ "language_model.model.layers.41.self_attn.k_proj",
654
+ "language_model.model.layers.41.self_attn.v_proj",
655
+ "language_model.model.layers.41.self_attn.o_proj",
656
+ "language_model.model.layers.41.feed_forward.router",
657
+ "language_model.model.layers.41.feed_forward.shared_expert.gate_proj",
658
+ "language_model.model.layers.41.feed_forward.shared_expert.up_proj",
659
+ "language_model.model.layers.41.feed_forward.shared_expert.down_proj",
660
+ "language_model.model.layers.42.self_attn.q_proj",
661
+ "language_model.model.layers.42.self_attn.k_proj",
662
+ "language_model.model.layers.42.self_attn.v_proj",
663
+ "language_model.model.layers.42.self_attn.o_proj",
664
+ "language_model.model.layers.42.feed_forward.router",
665
+ "language_model.model.layers.42.feed_forward.shared_expert.gate_proj",
666
+ "language_model.model.layers.42.feed_forward.shared_expert.up_proj",
667
+ "language_model.model.layers.42.feed_forward.shared_expert.down_proj",
668
+ "language_model.model.layers.43.self_attn.q_proj",
669
+ "language_model.model.layers.43.self_attn.k_proj",
670
+ "language_model.model.layers.43.self_attn.v_proj",
671
+ "language_model.model.layers.43.self_attn.o_proj",
672
+ "language_model.model.layers.43.feed_forward.router",
673
+ "language_model.model.layers.43.feed_forward.shared_expert.gate_proj",
674
+ "language_model.model.layers.43.feed_forward.shared_expert.up_proj",
675
+ "language_model.model.layers.43.feed_forward.shared_expert.down_proj",
676
+ "language_model.model.layers.44.self_attn.q_proj",
677
+ "language_model.model.layers.44.self_attn.k_proj",
678
+ "language_model.model.layers.44.self_attn.v_proj",
679
+ "language_model.model.layers.44.self_attn.o_proj",
680
+ "language_model.model.layers.44.feed_forward.router",
681
+ "language_model.model.layers.44.feed_forward.shared_expert.gate_proj",
682
+ "language_model.model.layers.44.feed_forward.shared_expert.up_proj",
683
+ "language_model.model.layers.44.feed_forward.shared_expert.down_proj",
684
+ "language_model.model.layers.45.self_attn.q_proj",
685
+ "language_model.model.layers.45.self_attn.k_proj",
686
+ "language_model.model.layers.45.self_attn.v_proj",
687
+ "language_model.model.layers.45.self_attn.o_proj",
688
+ "language_model.model.layers.45.feed_forward.router",
689
+ "language_model.model.layers.45.feed_forward.shared_expert.gate_proj",
690
+ "language_model.model.layers.45.feed_forward.shared_expert.up_proj",
691
+ "language_model.model.layers.45.feed_forward.shared_expert.down_proj",
692
+ "language_model.model.layers.46.self_attn.q_proj",
693
+ "language_model.model.layers.46.self_attn.k_proj",
694
+ "language_model.model.layers.46.self_attn.v_proj",
695
+ "language_model.model.layers.46.self_attn.o_proj",
696
+ "language_model.model.layers.46.feed_forward.router",
697
+ "language_model.model.layers.46.feed_forward.shared_expert.gate_proj",
698
+ "language_model.model.layers.46.feed_forward.shared_expert.up_proj",
699
+ "language_model.model.layers.46.feed_forward.shared_expert.down_proj",
700
+ "language_model.model.layers.47.self_attn.q_proj",
701
+ "language_model.model.layers.47.self_attn.k_proj",
702
+ "language_model.model.layers.47.self_attn.v_proj",
703
+ "language_model.model.layers.47.self_attn.o_proj",
704
+ "language_model.model.layers.47.feed_forward.router",
705
+ "language_model.model.layers.47.feed_forward.shared_expert.gate_proj",
706
+ "language_model.model.layers.47.feed_forward.shared_expert.up_proj",
707
+ "language_model.model.layers.47.feed_forward.shared_expert.down_proj",
708
+ "language_model.lm_head"
709
+ ],
710
+ "kv_cache_scheme": null,
711
+ "quant_method": "compressed-tensors",
712
+ "quantization_status": "compressed"
713
+ }
714
+ }
config.json.original ADDED
@@ -0,0 +1,817 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Llama4ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 200080,
6
+ "eoi_token_index": 200081,
7
+ "image_token_index": 200092,
8
+ "model_type": "llama4",
9
+ "quantization_config": {
10
+ "config_groups": {
11
+ "group_0": {
12
+ "input_activations": {
13
+ "actorder": null,
14
+ "block_structure": null,
15
+ "dynamic": true,
16
+ "group_size": null,
17
+ "num_bits": 8,
18
+ "observer": null,
19
+ "observer_kwargs": {},
20
+ "strategy": "token",
21
+ "symmetric": true,
22
+ "type": "float"
23
+ },
24
+ "output_activations": null,
25
+ "targets": [
26
+ "Linear"
27
+ ],
28
+ "weights": {
29
+ "actorder": null,
30
+ "block_structure": null,
31
+ "dynamic": false,
32
+ "group_size": null,
33
+ "num_bits": 8,
34
+ "observer": "mse",
35
+ "observer_kwargs": {},
36
+ "strategy": "channel",
37
+ "symmetric": true,
38
+ "type": "float"
39
+ }
40
+ }
41
+ },
42
+ "format": "float-quantized",
43
+ "global_compression_ratio": null,
44
+ "ignore": [
45
+ "vision_model.patch_embedding.linear",
46
+ "vision_model.model.layers.0.self_attn.q_proj",
47
+ "vision_model.model.layers.0.self_attn.k_proj",
48
+ "vision_model.model.layers.0.self_attn.v_proj",
49
+ "vision_model.model.layers.0.self_attn.o_proj",
50
+ "vision_model.model.layers.0.mlp.fc1",
51
+ "vision_model.model.layers.0.mlp.fc2",
52
+ "vision_model.model.layers.1.self_attn.q_proj",
53
+ "vision_model.model.layers.1.self_attn.k_proj",
54
+ "vision_model.model.layers.1.self_attn.v_proj",
55
+ "vision_model.model.layers.1.self_attn.o_proj",
56
+ "vision_model.model.layers.1.mlp.fc1",
57
+ "vision_model.model.layers.1.mlp.fc2",
58
+ "vision_model.model.layers.2.self_attn.q_proj",
59
+ "vision_model.model.layers.2.self_attn.k_proj",
60
+ "vision_model.model.layers.2.self_attn.v_proj",
61
+ "vision_model.model.layers.2.self_attn.o_proj",
62
+ "vision_model.model.layers.2.mlp.fc1",
63
+ "vision_model.model.layers.2.mlp.fc2",
64
+ "vision_model.model.layers.3.self_attn.q_proj",
65
+ "vision_model.model.layers.3.self_attn.k_proj",
66
+ "vision_model.model.layers.3.self_attn.v_proj",
67
+ "vision_model.model.layers.3.self_attn.o_proj",
68
+ "vision_model.model.layers.3.mlp.fc1",
69
+ "vision_model.model.layers.3.mlp.fc2",
70
+ "vision_model.model.layers.4.self_attn.q_proj",
71
+ "vision_model.model.layers.4.self_attn.k_proj",
72
+ "vision_model.model.layers.4.self_attn.v_proj",
73
+ "vision_model.model.layers.4.self_attn.o_proj",
74
+ "vision_model.model.layers.4.mlp.fc1",
75
+ "vision_model.model.layers.4.mlp.fc2",
76
+ "vision_model.model.layers.5.self_attn.q_proj",
77
+ "vision_model.model.layers.5.self_attn.k_proj",
78
+ "vision_model.model.layers.5.self_attn.v_proj",
79
+ "vision_model.model.layers.5.self_attn.o_proj",
80
+ "vision_model.model.layers.5.mlp.fc1",
81
+ "vision_model.model.layers.5.mlp.fc2",
82
+ "vision_model.model.layers.6.self_attn.q_proj",
83
+ "vision_model.model.layers.6.self_attn.k_proj",
84
+ "vision_model.model.layers.6.self_attn.v_proj",
85
+ "vision_model.model.layers.6.self_attn.o_proj",
86
+ "vision_model.model.layers.6.mlp.fc1",
87
+ "vision_model.model.layers.6.mlp.fc2",
88
+ "vision_model.model.layers.7.self_attn.q_proj",
89
+ "vision_model.model.layers.7.self_attn.k_proj",
90
+ "vision_model.model.layers.7.self_attn.v_proj",
91
+ "vision_model.model.layers.7.self_attn.o_proj",
92
+ "vision_model.model.layers.7.mlp.fc1",
93
+ "vision_model.model.layers.7.mlp.fc2",
94
+ "vision_model.model.layers.8.self_attn.q_proj",
95
+ "vision_model.model.layers.8.self_attn.k_proj",
96
+ "vision_model.model.layers.8.self_attn.v_proj",
97
+ "vision_model.model.layers.8.self_attn.o_proj",
98
+ "vision_model.model.layers.8.mlp.fc1",
99
+ "vision_model.model.layers.8.mlp.fc2",
100
+ "vision_model.model.layers.9.self_attn.q_proj",
101
+ "vision_model.model.layers.9.self_attn.k_proj",
102
+ "vision_model.model.layers.9.self_attn.v_proj",
103
+ "vision_model.model.layers.9.self_attn.o_proj",
104
+ "vision_model.model.layers.9.mlp.fc1",
105
+ "vision_model.model.layers.9.mlp.fc2",
106
+ "vision_model.model.layers.10.self_attn.q_proj",
107
+ "vision_model.model.layers.10.self_attn.k_proj",
108
+ "vision_model.model.layers.10.self_attn.v_proj",
109
+ "vision_model.model.layers.10.self_attn.o_proj",
110
+ "vision_model.model.layers.10.mlp.fc1",
111
+ "vision_model.model.layers.10.mlp.fc2",
112
+ "vision_model.model.layers.11.self_attn.q_proj",
113
+ "vision_model.model.layers.11.self_attn.k_proj",
114
+ "vision_model.model.layers.11.self_attn.v_proj",
115
+ "vision_model.model.layers.11.self_attn.o_proj",
116
+ "vision_model.model.layers.11.mlp.fc1",
117
+ "vision_model.model.layers.11.mlp.fc2",
118
+ "vision_model.model.layers.12.self_attn.q_proj",
119
+ "vision_model.model.layers.12.self_attn.k_proj",
120
+ "vision_model.model.layers.12.self_attn.v_proj",
121
+ "vision_model.model.layers.12.self_attn.o_proj",
122
+ "vision_model.model.layers.12.mlp.fc1",
123
+ "vision_model.model.layers.12.mlp.fc2",
124
+ "vision_model.model.layers.13.self_attn.q_proj",
125
+ "vision_model.model.layers.13.self_attn.k_proj",
126
+ "vision_model.model.layers.13.self_attn.v_proj",
127
+ "vision_model.model.layers.13.self_attn.o_proj",
128
+ "vision_model.model.layers.13.mlp.fc1",
129
+ "vision_model.model.layers.13.mlp.fc2",
130
+ "vision_model.model.layers.14.self_attn.q_proj",
131
+ "vision_model.model.layers.14.self_attn.k_proj",
132
+ "vision_model.model.layers.14.self_attn.v_proj",
133
+ "vision_model.model.layers.14.self_attn.o_proj",
134
+ "vision_model.model.layers.14.mlp.fc1",
135
+ "vision_model.model.layers.14.mlp.fc2",
136
+ "vision_model.model.layers.15.self_attn.q_proj",
137
+ "vision_model.model.layers.15.self_attn.k_proj",
138
+ "vision_model.model.layers.15.self_attn.v_proj",
139
+ "vision_model.model.layers.15.self_attn.o_proj",
140
+ "vision_model.model.layers.15.mlp.fc1",
141
+ "vision_model.model.layers.15.mlp.fc2",
142
+ "vision_model.model.layers.16.self_attn.q_proj",
143
+ "vision_model.model.layers.16.self_attn.k_proj",
144
+ "vision_model.model.layers.16.self_attn.v_proj",
145
+ "vision_model.model.layers.16.self_attn.o_proj",
146
+ "vision_model.model.layers.16.mlp.fc1",
147
+ "vision_model.model.layers.16.mlp.fc2",
148
+ "vision_model.model.layers.17.self_attn.q_proj",
149
+ "vision_model.model.layers.17.self_attn.k_proj",
150
+ "vision_model.model.layers.17.self_attn.v_proj",
151
+ "vision_model.model.layers.17.self_attn.o_proj",
152
+ "vision_model.model.layers.17.mlp.fc1",
153
+ "vision_model.model.layers.17.mlp.fc2",
154
+ "vision_model.model.layers.18.self_attn.q_proj",
155
+ "vision_model.model.layers.18.self_attn.k_proj",
156
+ "vision_model.model.layers.18.self_attn.v_proj",
157
+ "vision_model.model.layers.18.self_attn.o_proj",
158
+ "vision_model.model.layers.18.mlp.fc1",
159
+ "vision_model.model.layers.18.mlp.fc2",
160
+ "vision_model.model.layers.19.self_attn.q_proj",
161
+ "vision_model.model.layers.19.self_attn.k_proj",
162
+ "vision_model.model.layers.19.self_attn.v_proj",
163
+ "vision_model.model.layers.19.self_attn.o_proj",
164
+ "vision_model.model.layers.19.mlp.fc1",
165
+ "vision_model.model.layers.19.mlp.fc2",
166
+ "vision_model.model.layers.20.self_attn.q_proj",
167
+ "vision_model.model.layers.20.self_attn.k_proj",
168
+ "vision_model.model.layers.20.self_attn.v_proj",
169
+ "vision_model.model.layers.20.self_attn.o_proj",
170
+ "vision_model.model.layers.20.mlp.fc1",
171
+ "vision_model.model.layers.20.mlp.fc2",
172
+ "vision_model.model.layers.21.self_attn.q_proj",
173
+ "vision_model.model.layers.21.self_attn.k_proj",
174
+ "vision_model.model.layers.21.self_attn.v_proj",
175
+ "vision_model.model.layers.21.self_attn.o_proj",
176
+ "vision_model.model.layers.21.mlp.fc1",
177
+ "vision_model.model.layers.21.mlp.fc2",
178
+ "vision_model.model.layers.22.self_attn.q_proj",
179
+ "vision_model.model.layers.22.self_attn.k_proj",
180
+ "vision_model.model.layers.22.self_attn.v_proj",
181
+ "vision_model.model.layers.22.self_attn.o_proj",
182
+ "vision_model.model.layers.22.mlp.fc1",
183
+ "vision_model.model.layers.22.mlp.fc2",
184
+ "vision_model.model.layers.23.self_attn.q_proj",
185
+ "vision_model.model.layers.23.self_attn.k_proj",
186
+ "vision_model.model.layers.23.self_attn.v_proj",
187
+ "vision_model.model.layers.23.self_attn.o_proj",
188
+ "vision_model.model.layers.23.mlp.fc1",
189
+ "vision_model.model.layers.23.mlp.fc2",
190
+ "vision_model.model.layers.24.self_attn.q_proj",
191
+ "vision_model.model.layers.24.self_attn.k_proj",
192
+ "vision_model.model.layers.24.self_attn.v_proj",
193
+ "vision_model.model.layers.24.self_attn.o_proj",
194
+ "vision_model.model.layers.24.mlp.fc1",
195
+ "vision_model.model.layers.24.mlp.fc2",
196
+ "vision_model.model.layers.25.self_attn.q_proj",
197
+ "vision_model.model.layers.25.self_attn.k_proj",
198
+ "vision_model.model.layers.25.self_attn.v_proj",
199
+ "vision_model.model.layers.25.self_attn.o_proj",
200
+ "vision_model.model.layers.25.mlp.fc1",
201
+ "vision_model.model.layers.25.mlp.fc2",
202
+ "vision_model.model.layers.26.self_attn.q_proj",
203
+ "vision_model.model.layers.26.self_attn.k_proj",
204
+ "vision_model.model.layers.26.self_attn.v_proj",
205
+ "vision_model.model.layers.26.self_attn.o_proj",
206
+ "vision_model.model.layers.26.mlp.fc1",
207
+ "vision_model.model.layers.26.mlp.fc2",
208
+ "vision_model.model.layers.27.self_attn.q_proj",
209
+ "vision_model.model.layers.27.self_attn.k_proj",
210
+ "vision_model.model.layers.27.self_attn.v_proj",
211
+ "vision_model.model.layers.27.self_attn.o_proj",
212
+ "vision_model.model.layers.27.mlp.fc1",
213
+ "vision_model.model.layers.27.mlp.fc2",
214
+ "vision_model.model.layers.28.self_attn.q_proj",
215
+ "vision_model.model.layers.28.self_attn.k_proj",
216
+ "vision_model.model.layers.28.self_attn.v_proj",
217
+ "vision_model.model.layers.28.self_attn.o_proj",
218
+ "vision_model.model.layers.28.mlp.fc1",
219
+ "vision_model.model.layers.28.mlp.fc2",
220
+ "vision_model.model.layers.29.self_attn.q_proj",
221
+ "vision_model.model.layers.29.self_attn.k_proj",
222
+ "vision_model.model.layers.29.self_attn.v_proj",
223
+ "vision_model.model.layers.29.self_attn.o_proj",
224
+ "vision_model.model.layers.29.mlp.fc1",
225
+ "vision_model.model.layers.29.mlp.fc2",
226
+ "vision_model.model.layers.30.self_attn.q_proj",
227
+ "vision_model.model.layers.30.self_attn.k_proj",
228
+ "vision_model.model.layers.30.self_attn.v_proj",
229
+ "vision_model.model.layers.30.self_attn.o_proj",
230
+ "vision_model.model.layers.30.mlp.fc1",
231
+ "vision_model.model.layers.30.mlp.fc2",
232
+ "vision_model.model.layers.31.self_attn.q_proj",
233
+ "vision_model.model.layers.31.self_attn.k_proj",
234
+ "vision_model.model.layers.31.self_attn.v_proj",
235
+ "vision_model.model.layers.31.self_attn.o_proj",
236
+ "vision_model.model.layers.31.mlp.fc1",
237
+ "vision_model.model.layers.31.mlp.fc2",
238
+ "vision_model.model.layers.32.self_attn.q_proj",
239
+ "vision_model.model.layers.32.self_attn.k_proj",
240
+ "vision_model.model.layers.32.self_attn.v_proj",
241
+ "vision_model.model.layers.32.self_attn.o_proj",
242
+ "vision_model.model.layers.32.mlp.fc1",
243
+ "vision_model.model.layers.32.mlp.fc2",
244
+ "vision_model.model.layers.33.self_attn.q_proj",
245
+ "vision_model.model.layers.33.self_attn.k_proj",
246
+ "vision_model.model.layers.33.self_attn.v_proj",
247
+ "vision_model.model.layers.33.self_attn.o_proj",
248
+ "vision_model.model.layers.33.mlp.fc1",
249
+ "vision_model.model.layers.33.mlp.fc2",
250
+ "vision_model.vision_adapter.mlp.fc1",
251
+ "vision_model.vision_adapter.mlp.fc2",
252
+ "multi_modal_projector.linear_1",
253
+ "language_model.model.layers.0.self_attn.q_proj",
254
+ "language_model.model.layers.0.self_attn.k_proj",
255
+ "language_model.model.layers.0.self_attn.v_proj",
256
+ "language_model.model.layers.0.self_attn.o_proj",
257
+ "language_model.model.layers.0.feed_forward.router",
258
+ "language_model.model.layers.0.feed_forward.shared_expert.gate_proj",
259
+ "language_model.model.layers.0.feed_forward.shared_expert.up_proj",
260
+ "language_model.model.layers.0.feed_forward.shared_expert.down_proj",
261
+ "language_model.model.layers.1.self_attn.q_proj",
262
+ "language_model.model.layers.1.self_attn.k_proj",
263
+ "language_model.model.layers.1.self_attn.v_proj",
264
+ "language_model.model.layers.1.self_attn.o_proj",
265
+ "language_model.model.layers.1.feed_forward.router",
266
+ "language_model.model.layers.1.feed_forward.shared_expert.gate_proj",
267
+ "language_model.model.layers.1.feed_forward.shared_expert.up_proj",
268
+ "language_model.model.layers.1.feed_forward.shared_expert.down_proj",
269
+ "language_model.model.layers.2.self_attn.q_proj",
270
+ "language_model.model.layers.2.self_attn.k_proj",
271
+ "language_model.model.layers.2.self_attn.v_proj",
272
+ "language_model.model.layers.2.self_attn.o_proj",
273
+ "language_model.model.layers.2.feed_forward.router",
274
+ "language_model.model.layers.2.feed_forward.shared_expert.gate_proj",
275
+ "language_model.model.layers.2.feed_forward.shared_expert.up_proj",
276
+ "language_model.model.layers.2.feed_forward.shared_expert.down_proj",
277
+ "language_model.model.layers.3.self_attn.q_proj",
278
+ "language_model.model.layers.3.self_attn.k_proj",
279
+ "language_model.model.layers.3.self_attn.v_proj",
280
+ "language_model.model.layers.3.self_attn.o_proj",
281
+ "language_model.model.layers.3.feed_forward.router",
282
+ "language_model.model.layers.3.feed_forward.shared_expert.gate_proj",
283
+ "language_model.model.layers.3.feed_forward.shared_expert.up_proj",
284
+ "language_model.model.layers.3.feed_forward.shared_expert.down_proj",
285
+ "language_model.model.layers.4.self_attn.q_proj",
286
+ "language_model.model.layers.4.self_attn.k_proj",
287
+ "language_model.model.layers.4.self_attn.v_proj",
288
+ "language_model.model.layers.4.self_attn.o_proj",
289
+ "language_model.model.layers.4.feed_forward.router",
290
+ "language_model.model.layers.4.feed_forward.shared_expert.gate_proj",
291
+ "language_model.model.layers.4.feed_forward.shared_expert.up_proj",
292
+ "language_model.model.layers.4.feed_forward.shared_expert.down_proj",
293
+ "language_model.model.layers.5.self_attn.q_proj",
294
+ "language_model.model.layers.5.self_attn.k_proj",
295
+ "language_model.model.layers.5.self_attn.v_proj",
296
+ "language_model.model.layers.5.self_attn.o_proj",
297
+ "language_model.model.layers.5.feed_forward.router",
298
+ "language_model.model.layers.5.feed_forward.shared_expert.gate_proj",
299
+ "language_model.model.layers.5.feed_forward.shared_expert.up_proj",
300
+ "language_model.model.layers.5.feed_forward.shared_expert.down_proj",
301
+ "language_model.model.layers.6.self_attn.q_proj",
302
+ "language_model.model.layers.6.self_attn.k_proj",
303
+ "language_model.model.layers.6.self_attn.v_proj",
304
+ "language_model.model.layers.6.self_attn.o_proj",
305
+ "language_model.model.layers.6.feed_forward.router",
306
+ "language_model.model.layers.6.feed_forward.shared_expert.gate_proj",
307
+ "language_model.model.layers.6.feed_forward.shared_expert.up_proj",
308
+ "language_model.model.layers.6.feed_forward.shared_expert.down_proj",
309
+ "language_model.model.layers.7.self_attn.q_proj",
310
+ "language_model.model.layers.7.self_attn.k_proj",
311
+ "language_model.model.layers.7.self_attn.v_proj",
312
+ "language_model.model.layers.7.self_attn.o_proj",
313
+ "language_model.model.layers.7.feed_forward.router",
314
+ "language_model.model.layers.7.feed_forward.shared_expert.gate_proj",
315
+ "language_model.model.layers.7.feed_forward.shared_expert.up_proj",
316
+ "language_model.model.layers.7.feed_forward.shared_expert.down_proj",
317
+ "language_model.model.layers.8.self_attn.q_proj",
318
+ "language_model.model.layers.8.self_attn.k_proj",
319
+ "language_model.model.layers.8.self_attn.v_proj",
320
+ "language_model.model.layers.8.self_attn.o_proj",
321
+ "language_model.model.layers.8.feed_forward.router",
322
+ "language_model.model.layers.8.feed_forward.shared_expert.gate_proj",
323
+ "language_model.model.layers.8.feed_forward.shared_expert.up_proj",
324
+ "language_model.model.layers.8.feed_forward.shared_expert.down_proj",
325
+ "language_model.model.layers.9.self_attn.q_proj",
326
+ "language_model.model.layers.9.self_attn.k_proj",
327
+ "language_model.model.layers.9.self_attn.v_proj",
328
+ "language_model.model.layers.9.self_attn.o_proj",
329
+ "language_model.model.layers.9.feed_forward.router",
330
+ "language_model.model.layers.9.feed_forward.shared_expert.gate_proj",
331
+ "language_model.model.layers.9.feed_forward.shared_expert.up_proj",
332
+ "language_model.model.layers.9.feed_forward.shared_expert.down_proj",
333
+ "language_model.model.layers.10.self_attn.q_proj",
334
+ "language_model.model.layers.10.self_attn.k_proj",
335
+ "language_model.model.layers.10.self_attn.v_proj",
336
+ "language_model.model.layers.10.self_attn.o_proj",
337
+ "language_model.model.layers.10.feed_forward.router",
338
+ "language_model.model.layers.10.feed_forward.shared_expert.gate_proj",
339
+ "language_model.model.layers.10.feed_forward.shared_expert.up_proj",
340
+ "language_model.model.layers.10.feed_forward.shared_expert.down_proj",
341
+ "language_model.model.layers.11.self_attn.q_proj",
342
+ "language_model.model.layers.11.self_attn.k_proj",
343
+ "language_model.model.layers.11.self_attn.v_proj",
344
+ "language_model.model.layers.11.self_attn.o_proj",
345
+ "language_model.model.layers.11.feed_forward.router",
346
+ "language_model.model.layers.11.feed_forward.shared_expert.gate_proj",
347
+ "language_model.model.layers.11.feed_forward.shared_expert.up_proj",
348
+ "language_model.model.layers.11.feed_forward.shared_expert.down_proj",
349
+ "language_model.model.layers.12.self_attn.q_proj",
350
+ "language_model.model.layers.12.self_attn.k_proj",
351
+ "language_model.model.layers.12.self_attn.v_proj",
352
+ "language_model.model.layers.12.self_attn.o_proj",
353
+ "language_model.model.layers.12.feed_forward.router",
354
+ "language_model.model.layers.12.feed_forward.shared_expert.gate_proj",
355
+ "language_model.model.layers.12.feed_forward.shared_expert.up_proj",
356
+ "language_model.model.layers.12.feed_forward.shared_expert.down_proj",
357
+ "language_model.model.layers.13.self_attn.q_proj",
358
+ "language_model.model.layers.13.self_attn.k_proj",
359
+ "language_model.model.layers.13.self_attn.v_proj",
360
+ "language_model.model.layers.13.self_attn.o_proj",
361
+ "language_model.model.layers.13.feed_forward.router",
362
+ "language_model.model.layers.13.feed_forward.shared_expert.gate_proj",
363
+ "language_model.model.layers.13.feed_forward.shared_expert.up_proj",
364
+ "language_model.model.layers.13.feed_forward.shared_expert.down_proj",
365
+ "language_model.model.layers.14.self_attn.q_proj",
366
+ "language_model.model.layers.14.self_attn.k_proj",
367
+ "language_model.model.layers.14.self_attn.v_proj",
368
+ "language_model.model.layers.14.self_attn.o_proj",
369
+ "language_model.model.layers.14.feed_forward.router",
370
+ "language_model.model.layers.14.feed_forward.shared_expert.gate_proj",
371
+ "language_model.model.layers.14.feed_forward.shared_expert.up_proj",
372
+ "language_model.model.layers.14.feed_forward.shared_expert.down_proj",
373
+ "language_model.model.layers.15.self_attn.q_proj",
374
+ "language_model.model.layers.15.self_attn.k_proj",
375
+ "language_model.model.layers.15.self_attn.v_proj",
376
+ "language_model.model.layers.15.self_attn.o_proj",
377
+ "language_model.model.layers.15.feed_forward.router",
378
+ "language_model.model.layers.15.feed_forward.shared_expert.gate_proj",
379
+ "language_model.model.layers.15.feed_forward.shared_expert.up_proj",
380
+ "language_model.model.layers.15.feed_forward.shared_expert.down_proj",
381
+ "language_model.model.layers.16.self_attn.q_proj",
382
+ "language_model.model.layers.16.self_attn.k_proj",
383
+ "language_model.model.layers.16.self_attn.v_proj",
384
+ "language_model.model.layers.16.self_attn.o_proj",
385
+ "language_model.model.layers.16.feed_forward.router",
386
+ "language_model.model.layers.16.feed_forward.shared_expert.gate_proj",
387
+ "language_model.model.layers.16.feed_forward.shared_expert.up_proj",
388
+ "language_model.model.layers.16.feed_forward.shared_expert.down_proj",
389
+ "language_model.model.layers.17.self_attn.q_proj",
390
+ "language_model.model.layers.17.self_attn.k_proj",
391
+ "language_model.model.layers.17.self_attn.v_proj",
392
+ "language_model.model.layers.17.self_attn.o_proj",
393
+ "language_model.model.layers.17.feed_forward.router",
394
+ "language_model.model.layers.17.feed_forward.shared_expert.gate_proj",
395
+ "language_model.model.layers.17.feed_forward.shared_expert.up_proj",
396
+ "language_model.model.layers.17.feed_forward.shared_expert.down_proj",
397
+ "language_model.model.layers.18.self_attn.q_proj",
398
+ "language_model.model.layers.18.self_attn.k_proj",
399
+ "language_model.model.layers.18.self_attn.v_proj",
400
+ "language_model.model.layers.18.self_attn.o_proj",
401
+ "language_model.model.layers.18.feed_forward.router",
402
+ "language_model.model.layers.18.feed_forward.shared_expert.gate_proj",
403
+ "language_model.model.layers.18.feed_forward.shared_expert.up_proj",
404
+ "language_model.model.layers.18.feed_forward.shared_expert.down_proj",
405
+ "language_model.model.layers.19.self_attn.q_proj",
406
+ "language_model.model.layers.19.self_attn.k_proj",
407
+ "language_model.model.layers.19.self_attn.v_proj",
408
+ "language_model.model.layers.19.self_attn.o_proj",
409
+ "language_model.model.layers.19.feed_forward.router",
410
+ "language_model.model.layers.19.feed_forward.shared_expert.gate_proj",
411
+ "language_model.model.layers.19.feed_forward.shared_expert.up_proj",
412
+ "language_model.model.layers.19.feed_forward.shared_expert.down_proj",
413
+ "language_model.model.layers.20.self_attn.q_proj",
414
+ "language_model.model.layers.20.self_attn.k_proj",
415
+ "language_model.model.layers.20.self_attn.v_proj",
416
+ "language_model.model.layers.20.self_attn.o_proj",
417
+ "language_model.model.layers.20.feed_forward.router",
418
+ "language_model.model.layers.20.feed_forward.shared_expert.gate_proj",
419
+ "language_model.model.layers.20.feed_forward.shared_expert.up_proj",
420
+ "language_model.model.layers.20.feed_forward.shared_expert.down_proj",
421
+ "language_model.model.layers.21.self_attn.q_proj",
422
+ "language_model.model.layers.21.self_attn.k_proj",
423
+ "language_model.model.layers.21.self_attn.v_proj",
424
+ "language_model.model.layers.21.self_attn.o_proj",
425
+ "language_model.model.layers.21.feed_forward.router",
426
+ "language_model.model.layers.21.feed_forward.shared_expert.gate_proj",
427
+ "language_model.model.layers.21.feed_forward.shared_expert.up_proj",
428
+ "language_model.model.layers.21.feed_forward.shared_expert.down_proj",
429
+ "language_model.model.layers.22.self_attn.q_proj",
430
+ "language_model.model.layers.22.self_attn.k_proj",
431
+ "language_model.model.layers.22.self_attn.v_proj",
432
+ "language_model.model.layers.22.self_attn.o_proj",
433
+ "language_model.model.layers.22.feed_forward.router",
434
+ "language_model.model.layers.22.feed_forward.shared_expert.gate_proj",
435
+ "language_model.model.layers.22.feed_forward.shared_expert.up_proj",
436
+ "language_model.model.layers.22.feed_forward.shared_expert.down_proj",
437
+ "language_model.model.layers.23.self_attn.q_proj",
438
+ "language_model.model.layers.23.self_attn.k_proj",
439
+ "language_model.model.layers.23.self_attn.v_proj",
440
+ "language_model.model.layers.23.self_attn.o_proj",
441
+ "language_model.model.layers.23.feed_forward.router",
442
+ "language_model.model.layers.23.feed_forward.shared_expert.gate_proj",
443
+ "language_model.model.layers.23.feed_forward.shared_expert.up_proj",
444
+ "language_model.model.layers.23.feed_forward.shared_expert.down_proj",
445
+ "language_model.model.layers.24.self_attn.q_proj",
446
+ "language_model.model.layers.24.self_attn.k_proj",
447
+ "language_model.model.layers.24.self_attn.v_proj",
448
+ "language_model.model.layers.24.self_attn.o_proj",
449
+ "language_model.model.layers.24.feed_forward.router",
450
+ "language_model.model.layers.24.feed_forward.shared_expert.gate_proj",
451
+ "language_model.model.layers.24.feed_forward.shared_expert.up_proj",
452
+ "language_model.model.layers.24.feed_forward.shared_expert.down_proj",
453
+ "language_model.model.layers.25.self_attn.q_proj",
454
+ "language_model.model.layers.25.self_attn.k_proj",
455
+ "language_model.model.layers.25.self_attn.v_proj",
456
+ "language_model.model.layers.25.self_attn.o_proj",
457
+ "language_model.model.layers.25.feed_forward.router",
458
+ "language_model.model.layers.25.feed_forward.shared_expert.gate_proj",
459
+ "language_model.model.layers.25.feed_forward.shared_expert.up_proj",
460
+ "language_model.model.layers.25.feed_forward.shared_expert.down_proj",
461
+ "language_model.model.layers.26.self_attn.q_proj",
462
+ "language_model.model.layers.26.self_attn.k_proj",
463
+ "language_model.model.layers.26.self_attn.v_proj",
464
+ "language_model.model.layers.26.self_attn.o_proj",
465
+ "language_model.model.layers.26.feed_forward.router",
466
+ "language_model.model.layers.26.feed_forward.shared_expert.gate_proj",
467
+ "language_model.model.layers.26.feed_forward.shared_expert.up_proj",
468
+ "language_model.model.layers.26.feed_forward.shared_expert.down_proj",
469
+ "language_model.model.layers.27.self_attn.q_proj",
470
+ "language_model.model.layers.27.self_attn.k_proj",
471
+ "language_model.model.layers.27.self_attn.v_proj",
472
+ "language_model.model.layers.27.self_attn.o_proj",
473
+ "language_model.model.layers.27.feed_forward.router",
474
+ "language_model.model.layers.27.feed_forward.shared_expert.gate_proj",
475
+ "language_model.model.layers.27.feed_forward.shared_expert.up_proj",
476
+ "language_model.model.layers.27.feed_forward.shared_expert.down_proj",
477
+ "language_model.model.layers.28.self_attn.q_proj",
478
+ "language_model.model.layers.28.self_attn.k_proj",
479
+ "language_model.model.layers.28.self_attn.v_proj",
480
+ "language_model.model.layers.28.self_attn.o_proj",
481
+ "language_model.model.layers.28.feed_forward.router",
482
+ "language_model.model.layers.28.feed_forward.shared_expert.gate_proj",
483
+ "language_model.model.layers.28.feed_forward.shared_expert.up_proj",
484
+ "language_model.model.layers.28.feed_forward.shared_expert.down_proj",
485
+ "language_model.model.layers.29.self_attn.q_proj",
486
+ "language_model.model.layers.29.self_attn.k_proj",
487
+ "language_model.model.layers.29.self_attn.v_proj",
488
+ "language_model.model.layers.29.self_attn.o_proj",
489
+ "language_model.model.layers.29.feed_forward.router",
490
+ "language_model.model.layers.29.feed_forward.shared_expert.gate_proj",
491
+ "language_model.model.layers.29.feed_forward.shared_expert.up_proj",
492
+ "language_model.model.layers.29.feed_forward.shared_expert.down_proj",
493
+ "language_model.model.layers.30.self_attn.q_proj",
494
+ "language_model.model.layers.30.self_attn.k_proj",
495
+ "language_model.model.layers.30.self_attn.v_proj",
496
+ "language_model.model.layers.30.self_attn.o_proj",
497
+ "language_model.model.layers.30.feed_forward.router",
498
+ "language_model.model.layers.30.feed_forward.shared_expert.gate_proj",
499
+ "language_model.model.layers.30.feed_forward.shared_expert.up_proj",
500
+ "language_model.model.layers.30.feed_forward.shared_expert.down_proj",
501
+ "language_model.model.layers.31.self_attn.q_proj",
502
+ "language_model.model.layers.31.self_attn.k_proj",
503
+ "language_model.model.layers.31.self_attn.v_proj",
504
+ "language_model.model.layers.31.self_attn.o_proj",
505
+ "language_model.model.layers.31.feed_forward.router",
506
+ "language_model.model.layers.31.feed_forward.shared_expert.gate_proj",
507
+ "language_model.model.layers.31.feed_forward.shared_expert.up_proj",
508
+ "language_model.model.layers.31.feed_forward.shared_expert.down_proj",
509
+ "language_model.model.layers.32.self_attn.q_proj",
510
+ "language_model.model.layers.32.self_attn.k_proj",
511
+ "language_model.model.layers.32.self_attn.v_proj",
512
+ "language_model.model.layers.32.self_attn.o_proj",
513
+ "language_model.model.layers.32.feed_forward.router",
514
+ "language_model.model.layers.32.feed_forward.shared_expert.gate_proj",
515
+ "language_model.model.layers.32.feed_forward.shared_expert.up_proj",
516
+ "language_model.model.layers.32.feed_forward.shared_expert.down_proj",
517
+ "language_model.model.layers.33.self_attn.q_proj",
518
+ "language_model.model.layers.33.self_attn.k_proj",
519
+ "language_model.model.layers.33.self_attn.v_proj",
520
+ "language_model.model.layers.33.self_attn.o_proj",
521
+ "language_model.model.layers.33.feed_forward.router",
522
+ "language_model.model.layers.33.feed_forward.shared_expert.gate_proj",
523
+ "language_model.model.layers.33.feed_forward.shared_expert.up_proj",
524
+ "language_model.model.layers.33.feed_forward.shared_expert.down_proj",
525
+ "language_model.model.layers.34.self_attn.q_proj",
526
+ "language_model.model.layers.34.self_attn.k_proj",
527
+ "language_model.model.layers.34.self_attn.v_proj",
528
+ "language_model.model.layers.34.self_attn.o_proj",
529
+ "language_model.model.layers.34.feed_forward.router",
530
+ "language_model.model.layers.34.feed_forward.shared_expert.gate_proj",
531
+ "language_model.model.layers.34.feed_forward.shared_expert.up_proj",
532
+ "language_model.model.layers.34.feed_forward.shared_expert.down_proj",
533
+ "language_model.model.layers.35.self_attn.q_proj",
534
+ "language_model.model.layers.35.self_attn.k_proj",
535
+ "language_model.model.layers.35.self_attn.v_proj",
536
+ "language_model.model.layers.35.self_attn.o_proj",
537
+ "language_model.model.layers.35.feed_forward.router",
538
+ "language_model.model.layers.35.feed_forward.shared_expert.gate_proj",
539
+ "language_model.model.layers.35.feed_forward.shared_expert.up_proj",
540
+ "language_model.model.layers.35.feed_forward.shared_expert.down_proj",
541
+ "language_model.model.layers.36.self_attn.q_proj",
542
+ "language_model.model.layers.36.self_attn.k_proj",
543
+ "language_model.model.layers.36.self_attn.v_proj",
544
+ "language_model.model.layers.36.self_attn.o_proj",
545
+ "language_model.model.layers.36.feed_forward.router",
546
+ "language_model.model.layers.36.feed_forward.shared_expert.gate_proj",
547
+ "language_model.model.layers.36.feed_forward.shared_expert.up_proj",
548
+ "language_model.model.layers.36.feed_forward.shared_expert.down_proj",
549
+ "language_model.model.layers.37.self_attn.q_proj",
550
+ "language_model.model.layers.37.self_attn.k_proj",
551
+ "language_model.model.layers.37.self_attn.v_proj",
552
+ "language_model.model.layers.37.self_attn.o_proj",
553
+ "language_model.model.layers.37.feed_forward.router",
554
+ "language_model.model.layers.37.feed_forward.shared_expert.gate_proj",
555
+ "language_model.model.layers.37.feed_forward.shared_expert.up_proj",
556
+ "language_model.model.layers.37.feed_forward.shared_expert.down_proj",
557
+ "language_model.model.layers.38.self_attn.q_proj",
558
+ "language_model.model.layers.38.self_attn.k_proj",
559
+ "language_model.model.layers.38.self_attn.v_proj",
560
+ "language_model.model.layers.38.self_attn.o_proj",
561
+ "language_model.model.layers.38.feed_forward.router",
562
+ "language_model.model.layers.38.feed_forward.shared_expert.gate_proj",
563
+ "language_model.model.layers.38.feed_forward.shared_expert.up_proj",
564
+ "language_model.model.layers.38.feed_forward.shared_expert.down_proj",
565
+ "language_model.model.layers.39.self_attn.q_proj",
566
+ "language_model.model.layers.39.self_attn.k_proj",
567
+ "language_model.model.layers.39.self_attn.v_proj",
568
+ "language_model.model.layers.39.self_attn.o_proj",
569
+ "language_model.model.layers.39.feed_forward.router",
570
+ "language_model.model.layers.39.feed_forward.shared_expert.gate_proj",
571
+ "language_model.model.layers.39.feed_forward.shared_expert.up_proj",
572
+ "language_model.model.layers.39.feed_forward.shared_expert.down_proj",
573
+ "language_model.model.layers.40.self_attn.q_proj",
574
+ "language_model.model.layers.40.self_attn.k_proj",
575
+ "language_model.model.layers.40.self_attn.v_proj",
576
+ "language_model.model.layers.40.self_attn.o_proj",
577
+ "language_model.model.layers.40.feed_forward.router",
578
+ "language_model.model.layers.40.feed_forward.shared_expert.gate_proj",
579
+ "language_model.model.layers.40.feed_forward.shared_expert.up_proj",
580
+ "language_model.model.layers.40.feed_forward.shared_expert.down_proj",
581
+ "language_model.model.layers.41.self_attn.q_proj",
582
+ "language_model.model.layers.41.self_attn.k_proj",
583
+ "language_model.model.layers.41.self_attn.v_proj",
584
+ "language_model.model.layers.41.self_attn.o_proj",
585
+ "language_model.model.layers.41.feed_forward.router",
586
+ "language_model.model.layers.41.feed_forward.shared_expert.gate_proj",
587
+ "language_model.model.layers.41.feed_forward.shared_expert.up_proj",
588
+ "language_model.model.layers.41.feed_forward.shared_expert.down_proj",
589
+ "language_model.model.layers.42.self_attn.q_proj",
590
+ "language_model.model.layers.42.self_attn.k_proj",
591
+ "language_model.model.layers.42.self_attn.v_proj",
592
+ "language_model.model.layers.42.self_attn.o_proj",
593
+ "language_model.model.layers.42.feed_forward.router",
594
+ "language_model.model.layers.42.feed_forward.shared_expert.gate_proj",
595
+ "language_model.model.layers.42.feed_forward.shared_expert.up_proj",
596
+ "language_model.model.layers.42.feed_forward.shared_expert.down_proj",
597
+ "language_model.model.layers.43.self_attn.q_proj",
598
+ "language_model.model.layers.43.self_attn.k_proj",
599
+ "language_model.model.layers.43.self_attn.v_proj",
600
+ "language_model.model.layers.43.self_attn.o_proj",
601
+ "language_model.model.layers.43.feed_forward.router",
602
+ "language_model.model.layers.43.feed_forward.shared_expert.gate_proj",
603
+ "language_model.model.layers.43.feed_forward.shared_expert.up_proj",
604
+ "language_model.model.layers.43.feed_forward.shared_expert.down_proj",
605
+ "language_model.model.layers.44.self_attn.q_proj",
606
+ "language_model.model.layers.44.self_attn.k_proj",
607
+ "language_model.model.layers.44.self_attn.v_proj",
608
+ "language_model.model.layers.44.self_attn.o_proj",
609
+ "language_model.model.layers.44.feed_forward.router",
610
+ "language_model.model.layers.44.feed_forward.shared_expert.gate_proj",
611
+ "language_model.model.layers.44.feed_forward.shared_expert.up_proj",
612
+ "language_model.model.layers.44.feed_forward.shared_expert.down_proj",
613
+ "language_model.model.layers.45.self_attn.q_proj",
614
+ "language_model.model.layers.45.self_attn.k_proj",
615
+ "language_model.model.layers.45.self_attn.v_proj",
616
+ "language_model.model.layers.45.self_attn.o_proj",
617
+ "language_model.model.layers.45.feed_forward.router",
618
+ "language_model.model.layers.45.feed_forward.shared_expert.gate_proj",
619
+ "language_model.model.layers.45.feed_forward.shared_expert.up_proj",
620
+ "language_model.model.layers.45.feed_forward.shared_expert.down_proj",
621
+ "language_model.model.layers.46.self_attn.q_proj",
622
+ "language_model.model.layers.46.self_attn.k_proj",
623
+ "language_model.model.layers.46.self_attn.v_proj",
624
+ "language_model.model.layers.46.self_attn.o_proj",
625
+ "language_model.model.layers.46.feed_forward.router",
626
+ "language_model.model.layers.46.feed_forward.shared_expert.gate_proj",
627
+ "language_model.model.layers.46.feed_forward.shared_expert.up_proj",
628
+ "language_model.model.layers.46.feed_forward.shared_expert.down_proj",
629
+ "language_model.model.layers.47.self_attn.q_proj",
630
+ "language_model.model.layers.47.self_attn.k_proj",
631
+ "language_model.model.layers.47.self_attn.v_proj",
632
+ "language_model.model.layers.47.self_attn.o_proj",
633
+ "language_model.model.layers.47.feed_forward.router",
634
+ "language_model.model.layers.47.feed_forward.shared_expert.gate_proj",
635
+ "language_model.model.layers.47.feed_forward.shared_expert.up_proj",
636
+ "language_model.model.layers.47.feed_forward.shared_expert.down_proj",
637
+ "language_model.lm_head"
638
+ ],
639
+ "kv_cache_scheme": null,
640
+ "quant_method": "compressed-tensors",
641
+ "quantization_status": "compressed"
642
+ },
643
+ "text_config": {
644
+ "_attn_implementation_autoset": true,
645
+ "attention_bias": false,
646
+ "attention_chunk_size": 8192,
647
+ "attention_dropout": 0.0,
648
+ "attn_scale": 0.1,
649
+ "attn_temperature_tuning": 4,
650
+ "bos_token_id": 200000,
651
+ "eos_token_id": [
652
+ 200001,
653
+ 200007,
654
+ 200008
655
+ ],
656
+ "floor_scale": 8192,
657
+ "for_llm_compressor": true,
658
+ "head_dim": 128,
659
+ "hidden_act": "silu",
660
+ "hidden_size": 5120,
661
+ "initializer_range": 0.02,
662
+ "interleave_moe_layer_step": 1,
663
+ "intermediate_size": 8192,
664
+ "intermediate_size_mlp": 16384,
665
+ "max_position_embeddings": 10485760,
666
+ "model_type": "llama4_text",
667
+ "moe_layers": [
668
+ 0,
669
+ 1,
670
+ 2,
671
+ 3,
672
+ 4,
673
+ 5,
674
+ 6,
675
+ 7,
676
+ 8,
677
+ 9,
678
+ 10,
679
+ 11,
680
+ 12,
681
+ 13,
682
+ 14,
683
+ 15,
684
+ 16,
685
+ 17,
686
+ 18,
687
+ 19,
688
+ 20,
689
+ 21,
690
+ 22,
691
+ 23,
692
+ 24,
693
+ 25,
694
+ 26,
695
+ 27,
696
+ 28,
697
+ 29,
698
+ 30,
699
+ 31,
700
+ 32,
701
+ 33,
702
+ 34,
703
+ 35,
704
+ 36,
705
+ 37,
706
+ 38,
707
+ 39,
708
+ 40,
709
+ 41,
710
+ 42,
711
+ 43,
712
+ 44,
713
+ 45,
714
+ 46,
715
+ 47
716
+ ],
717
+ "no_rope_layers": [
718
+ 1,
719
+ 1,
720
+ 1,
721
+ 0,
722
+ 1,
723
+ 1,
724
+ 1,
725
+ 0,
726
+ 1,
727
+ 1,
728
+ 1,
729
+ 0,
730
+ 1,
731
+ 1,
732
+ 1,
733
+ 0,
734
+ 1,
735
+ 1,
736
+ 1,
737
+ 0,
738
+ 1,
739
+ 1,
740
+ 1,
741
+ 0,
742
+ 1,
743
+ 1,
744
+ 1,
745
+ 0,
746
+ 1,
747
+ 1,
748
+ 1,
749
+ 0,
750
+ 1,
751
+ 1,
752
+ 1,
753
+ 0,
754
+ 1,
755
+ 1,
756
+ 1,
757
+ 0,
758
+ 1,
759
+ 1,
760
+ 1,
761
+ 0,
762
+ 1,
763
+ 1,
764
+ 1,
765
+ 0
766
+ ],
767
+ "num_attention_heads": 40,
768
+ "num_experts_per_tok": 1,
769
+ "num_hidden_layers": 48,
770
+ "num_key_value_heads": 8,
771
+ "num_local_experts": 16,
772
+ "output_router_logits": false,
773
+ "pad_token_id": 200018,
774
+ "rms_norm_eps": 1e-05,
775
+ "rope_scaling": {
776
+ "factor": 16.0,
777
+ "high_freq_factor": 1.0,
778
+ "low_freq_factor": 1.0,
779
+ "original_max_position_embeddings": 8192,
780
+ "rope_type": "llama3"
781
+ },
782
+ "rope_theta": 500000.0,
783
+ "router_aux_loss_coef": 0.001,
784
+ "router_jitter_noise": 0.0,
785
+ "torch_dtype": "bfloat16",
786
+ "use_cache": true,
787
+ "use_qk_norm": true,
788
+ "vocab_size": 202048
789
+ },
790
+ "tie_word_embeddings": false,
791
+ "torch_dtype": "bfloat16",
792
+ "transformers_version": "4.51.0.dev0",
793
+ "vision_config": {
794
+ "_attn_implementation_autoset": true,
795
+ "attention_dropout": 0.0,
796
+ "hidden_act": "gelu",
797
+ "hidden_size": 1408,
798
+ "image_size": 336,
799
+ "initializer_range": 0.02,
800
+ "intermediate_size": 5632,
801
+ "model_type": "llama4_vision_model",
802
+ "multi_modal_projector_bias": false,
803
+ "norm_eps": 1e-05,
804
+ "num_attention_heads": 16,
805
+ "num_channels": 3,
806
+ "num_hidden_layers": 34,
807
+ "patch_size": 14,
808
+ "pixel_shuffle_ratio": 0.5,
809
+ "projector_dropout": 0.0,
810
+ "projector_input_dim": 4096,
811
+ "projector_output_dim": 4096,
812
+ "rope_theta": 10000,
813
+ "vision_feature_layer": -1,
814
+ "vision_feature_select_strategy": "default",
815
+ "vision_output_dim": 4096
816
+ }
817
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 200000,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 200001,
6
+ 200007,
7
+ 200008
8
+ ],
9
+ "pad_token_id": 200018,
10
+ "temperature": 0.6,
11
+ "top_p": 0.9,
12
+ "transformers_version": "4.51.0.dev0"
13
+ }
model-00001-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36c6c6b4a70837939c51ebbc1a3682c2dfd0695e5ba8f8509d8860088191cf51
3
+ size 4987679352
model-00002-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02fc570f74e96c1fd2ee828e4ae3d69dbdaa6554cc5768be4d0e5f3fa7b7ac0a
3
+ size 4993069104
model-00003-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6fac21aa1fa6d3b65273078a19903d093998c0db46cfdca619fab70758fa5df9
3
+ size 4993075288
model-00004-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f49cb8d3ddc44a715dae3da6db66b2d729fb628259311163694fc726e58008a5
3
+ size 4993069176
model-00005-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94ced100083cd715d2bac71163717f3e204d30cd36c0eb1aa74fb3d980205004
3
+ size 4993069200
model-00006-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:90bcea4c3fb40c6f3a6975e3380a497368738ba5df686414809ebdfe431141fc
3
+ size 4993205960
model-00007-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30fa834e2583a5bba4ab7452c859ff9bb5da6ef224b4dd58273f0a445b178408
3
+ size 4951076352
model-00008-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4e7acbfde1a6a3fd546c05bee4262ac88b30a1f6e0242fb545926c7b1f705465
3
+ size 4993031936
model-00009-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe1f0f050e4c519d47d1d5ec852275c4b0e7bb2799bb3195c5e9e097a62720d7
3
+ size 4993069192
model-00010-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d6836ce9eb1d9b5b15a98e99937d5d924eff206ad290cbde08c8f6de37a41c38
3
+ size 4993069216
model-00011-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b6546b902d40930ccbee985371e412ff9e93421bb699d7ad7a2cf4da45f1c61
3
+ size 4993075392
model-00012-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3dbe728b83b6f0abda059177b9ef469eaaeef59355fb78c2efc50a94fbd998de
3
+ size 4993069280
model-00013-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c661d5d69debe21c8d31e549a203bebaeebfd4b694b114d38a571d00fec6ec0
3
+ size 4993069304
model-00014-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa24e1409f811ca22955375b79166a537b6fc5b53f1a6c418e6405be6e89bc08
3
+ size 4993075480
model-00015-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:377f9bed46d56b0e82156b7f35d2ffd2cf40621c929d62c0168f943c1a30ff42
3
+ size 4993069376
model-00016-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed39b016dc2b3f68d1b0c60a60cb4ca44b58fbcf39cf70c54696fd1d2936493f
3
+ size 4993069408
model-00017-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cfa91c9d8734d8c524b755d51abfc04dec652efcfd868e1905a28289ca31c76e
3
+ size 4993075584
model-00018-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e514d79585ee2d25edb3a8b7aeebddcdf4d3d964fd36cbb9780c3cf7ebffd573
3
+ size 4993173160
model-00019-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:397722313f499afbc6da44a0110c5f3b067d3e8aeb6807f1aba051bbfa81452e
3
+ size 4993019672
model-00020-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4406a44e8804aa9c8ea21b6b96369ca772d373bb04d078b714b7306676852272
3
+ size 4993075320
model-00021-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8444521428980f70c2d46afe868a946e5bb2c7404302d31b725c3d9cf9cb8deb
3
+ size 4993069208
model-00022-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11b10547da072ddfc27d43f73119e557479f4e829440c3b697b6121c069f99b8
3
+ size 4993069232
model-00023-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:076ae740ff6265b501bd556aad41f666e005d6d3d1d6d9e32995e8c777a4859b
3
+ size 4993075408
model-00024-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19964b0364c221a1cc8fba57d2bdc20382af117e25225887b0eab416705877e3
3
+ size 3818306024
model-00025-of-00025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f412ae287747cc86e6ee5d58600bbe5c574d6fa77f351367aae9ea3f4dbcc3a
3
+ size 2068971664
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
preprocessor_config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "crop_size": null,
3
+ "data_format": "channels_first",
4
+ "default_to_square": true,
5
+ "device": null,
6
+ "do_center_crop": null,
7
+ "do_convert_rgb": true,
8
+ "do_normalize": true,
9
+ "do_rescale": true,
10
+ "do_resize": true,
11
+ "image_mean": [
12
+ 0.5,
13
+ 0.5,
14
+ 0.5
15
+ ],
16
+ "image_processor_type": "Llama4ImageProcessorFast",
17
+ "image_std": [
18
+ 0.5,
19
+ 0.5,
20
+ 0.5
21
+ ],
22
+ "input_data_format": null,
23
+ "max_patches": 16,
24
+ "processor_class": "Llama4Processor",
25
+ "resample": 2,
26
+ "rescale_factor": 0.00392156862745098,
27
+ "resize_to_max_canvas": false,
28
+ "return_tensors": null,
29
+ "size": {
30
+ "height": 336,
31
+ "width": 336
32
+ }
33
+ }
processor_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "fake_image_token": "<|image|>",
3
+ "image_token": "<|image|>",
4
+ "patch_size": 14,
5
+ "processor_class": "Llama4Processor"
6
+ }
recipe.yaml ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ default_stage:
2
+ default_modifiers:
3
+ QuantizationModifier:
4
+ config_groups:
5
+ group_0:
6
+ targets: [Linear]
7
+ weights: {num_bits: 8, type: float, symmetric: true, strategy: channel, observer: mse}
8
+ input_activations: {num_bits: 8, type: float, symmetric: true, strategy: token,
9
+ dynamic: true, observer: null}
10
+ output_activations: null
11
+ ignore: ['re:.*lm_head', 're:.*self_attn', 're:.*router', 're:.*vision_model', 're:.*multi_modal_projector',
12
+ 're:.*shared_expert', 're:.*feed_forward.gate_proj', 're:.*feed_forward.up_proj',
13
+ 're:.*feed_forward.down_proj']
14
+ targets: [Linear]
special_tokens_map.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "bos_token": "<|begin_of_text|>",
3
+ "eos_token": "<|eot|>",
4
+ "pad_token": "<|finetune_right_pad_id|>"
5
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:172c9eb4beafc72601690da3ccfcede5c2e6806a8d5ec1fca33e22acea8023a4
3
+ size 27948578
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0bdbaf59b0762c8c807617e2d8ea51420eb1b1de266df2495be755c8e0ed6ed
3
+ size 3622230
tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff