Hugging Face's logo

jaithree
/
Astrea-R8-Chat-9B-OptiQ-4bit

Text Generation
MLX
Safetensors
English
qwen3_5
qwen3.5
chat
creative-writing
altworld
conversational
4-bit precision
Model card Files
xet
 Community
jaithree commited on
Commit
e8b92ba
·
verified ·
1 Parent(s): 9dd76e9

Upload folder using huggingface_hub

Browse files
Files changed (12) hide show
  1. .gitattributes +1 -0
  2. README.md +11 -0
  3. chat_template.jinja +154 -0
  4. config.json +2129 -0
  5. generation_config.json +11 -0
  6. model-00001-of-00002.safetensors +3 -0
  7. model-00002-of-00002.safetensors +3 -0
  8. model.safetensors.index.json +0 -0
  9. optiq/optiq_vision.safetensors +3 -0
  10. optiq_metadata.json +1008 -0
  11. tokenizer.json +3 -0
  12. tokenizer_config.json +33 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,14 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
1
  ---
2
  license: apache-2.0
3
+ base_model: Altworld/Astrea-R8-Chat-9B
4
+ pipeline_tag: text-generation
5
+ library_name: mlx
6
+ language:
7
+ - en
8
+ tags:
9
+ - qwen3.5
10
+ - chat
11
+ - creative-writing
12
+ - altworld
13
+ - mlx
14
  ---
chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is not defined or enable_thinking is false %}
150
+ {{- '<think>\n\n</think>\n\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,2129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3_5ForConditionalGeneration"
4
+ ],
5
+ "dtype": "bfloat16",
6
+ "eos_token_id": [
7
+ 248044,
8
+ 248046
9
+ ],
10
+ "image_token_id": 248056,
11
+ "model_type": "qwen3_5",
12
+ "quantization": {
13
+ "group_size": 64,
14
+ "bits": 4,
15
+ "mode": "affine",
16
+ "language_model.model.embed_tokens": {
17
+ "bits": 8,
18
+ "group_size": 64
19
+ },
20
+ "language_model.model.layers.0.linear_attn.in_proj_qkv": {
21
+ "bits": 8,
22
+ "group_size": 64
23
+ },
24
+ "language_model.model.layers.0.linear_attn.in_proj_z": {
25
+ "bits": 8,
26
+ "group_size": 64
27
+ },
28
+ "language_model.model.layers.0.linear_attn.in_proj_b": {
29
+ "bits": 8,
30
+ "group_size": 64
31
+ },
32
+ "language_model.model.layers.0.linear_attn.in_proj_a": {
33
+ "bits": 8,
34
+ "group_size": 64
35
+ },
36
+ "language_model.model.layers.0.linear_attn.out_proj": {
37
+ "bits": 8,
38
+ "group_size": 64
39
+ },
40
+ "language_model.model.layers.0.mlp.gate_proj": {
41
+ "bits": 8,
42
+ "group_size": 64
43
+ },
44
+ "language_model.model.layers.0.mlp.down_proj": {
45
+ "bits": 8,
46
+ "group_size": 64
47
+ },
48
+ "language_model.model.layers.0.mlp.up_proj": {
49
+ "bits": 8,
50
+ "group_size": 64
51
+ },
52
+ "language_model.model.layers.1.linear_attn.in_proj_qkv": {
53
+ "bits": 8,
54
+ "group_size": 64
55
+ },
56
+ "language_model.model.layers.1.linear_attn.in_proj_z": {
57
+ "bits": 8,
58
+ "group_size": 64
59
+ },
60
+ "language_model.model.layers.1.linear_attn.in_proj_b": {
61
+ "bits": 8,
62
+ "group_size": 64
63
+ },
64
+ "language_model.model.layers.1.linear_attn.in_proj_a": {
65
+ "bits": 8,
66
+ "group_size": 64
67
+ },
68
+ "language_model.model.layers.1.linear_attn.out_proj": {
69
+ "bits": 4,
70
+ "group_size": 64
71
+ },
72
+ "language_model.model.layers.1.mlp.gate_proj": {
73
+ "bits": 4,
74
+ "group_size": 64
75
+ },
76
+ "language_model.model.layers.1.mlp.down_proj": {
77
+ "bits": 4,
78
+ "group_size": 64
79
+ },
80
+ "language_model.model.layers.1.mlp.up_proj": {
81
+ "bits": 4,
82
+ "group_size": 64
83
+ },
84
+ "language_model.model.layers.2.linear_attn.in_proj_qkv": {
85
+ "bits": 8,
86
+ "group_size": 64
87
+ },
88
+ "language_model.model.layers.2.linear_attn.in_proj_z": {
89
+ "bits": 4,
90
+ "group_size": 64
91
+ },
92
+ "language_model.model.layers.2.linear_attn.in_proj_b": {
93
+ "bits": 8,
94
+ "group_size": 64
95
+ },
96
+ "language_model.model.layers.2.linear_attn.in_proj_a": {
97
+ "bits": 8,
98
+ "group_size": 64
99
+ },
100
+ "language_model.model.layers.2.linear_attn.out_proj": {
101
+ "bits": 4,
102
+ "group_size": 64
103
+ },
104
+ "language_model.model.layers.2.mlp.gate_proj": {
105
+ "bits": 4,
106
+ "group_size": 64
107
+ },
108
+ "language_model.model.layers.2.mlp.down_proj": {
109
+ "bits": 4,
110
+ "group_size": 64
111
+ },
112
+ "language_model.model.layers.2.mlp.up_proj": {
113
+ "bits": 4,
114
+ "group_size": 64
115
+ },
116
+ "language_model.model.layers.3.self_attn.q_proj": {
117
+ "bits": 4,
118
+ "group_size": 64
119
+ },
120
+ "language_model.model.layers.3.self_attn.k_proj": {
121
+ "bits": 8,
122
+ "group_size": 64
123
+ },
124
+ "language_model.model.layers.3.self_attn.v_proj": {
125
+ "bits": 8,
126
+ "group_size": 64
127
+ },
128
+ "language_model.model.layers.3.self_attn.o_proj": {
129
+ "bits": 8,
130
+ "group_size": 64
131
+ },
132
+ "language_model.model.layers.3.mlp.gate_proj": {
133
+ "bits": 4,
134
+ "group_size": 64
135
+ },
136
+ "language_model.model.layers.3.mlp.down_proj": {
137
+ "bits": 4,
138
+ "group_size": 64
139
+ },
140
+ "language_model.model.layers.3.mlp.up_proj": {
141
+ "bits": 4,
142
+ "group_size": 64
143
+ },
144
+ "language_model.model.layers.4.linear_attn.in_proj_qkv": {
145
+ "bits": 8,
146
+ "group_size": 64
147
+ },
148
+ "language_model.model.layers.4.linear_attn.in_proj_z": {
149
+ "bits": 8,
150
+ "group_size": 64
151
+ },
152
+ "language_model.model.layers.4.linear_attn.in_proj_b": {
153
+ "bits": 8,
154
+ "group_size": 64
155
+ },
156
+ "language_model.model.layers.4.linear_attn.in_proj_a": {
157
+ "bits": 8,
158
+ "group_size": 64
159
+ },
160
+ "language_model.model.layers.4.linear_attn.out_proj": {
161
+ "bits": 4,
162
+ "group_size": 64
163
+ },
164
+ "language_model.model.layers.4.mlp.gate_proj": {
165
+ "bits": 4,
166
+ "group_size": 64
167
+ },
168
+ "language_model.model.layers.4.mlp.down_proj": {
169
+ "bits": 4,
170
+ "group_size": 64
171
+ },
172
+ "language_model.model.layers.4.mlp.up_proj": {
173
+ "bits": 4,
174
+ "group_size": 64
175
+ },
176
+ "language_model.model.layers.5.linear_attn.in_proj_qkv": {
177
+ "bits": 8,
178
+ "group_size": 64
179
+ },
180
+ "language_model.model.layers.5.linear_attn.in_proj_z": {
181
+ "bits": 4,
182
+ "group_size": 64
183
+ },
184
+ "language_model.model.layers.5.linear_attn.in_proj_b": {
185
+ "bits": 8,
186
+ "group_size": 64
187
+ },
188
+ "language_model.model.layers.5.linear_attn.in_proj_a": {
189
+ "bits": 8,
190
+ "group_size": 64
191
+ },
192
+ "language_model.model.layers.5.linear_attn.out_proj": {
193
+ "bits": 8,
194
+ "group_size": 64
195
+ },
196
+ "language_model.model.layers.5.mlp.gate_proj": {
197
+ "bits": 4,
198
+ "group_size": 64
199
+ },
200
+ "language_model.model.layers.5.mlp.down_proj": {
201
+ "bits": 4,
202
+ "group_size": 64
203
+ },
204
+ "language_model.model.layers.5.mlp.up_proj": {
205
+ "bits": 4,
206
+ "group_size": 64
207
+ },
208
+ "language_model.model.layers.6.linear_attn.in_proj_qkv": {
209
+ "bits": 8,
210
+ "group_size": 64
211
+ },
212
+ "language_model.model.layers.6.linear_attn.in_proj_z": {
213
+ "bits": 8,
214
+ "group_size": 64
215
+ },
216
+ "language_model.model.layers.6.linear_attn.in_proj_b": {
217
+ "bits": 8,
218
+ "group_size": 64
219
+ },
220
+ "language_model.model.layers.6.linear_attn.in_proj_a": {
221
+ "bits": 8,
222
+ "group_size": 64
223
+ },
224
+ "language_model.model.layers.6.linear_attn.out_proj": {
225
+ "bits": 4,
226
+ "group_size": 64
227
+ },
228
+ "language_model.model.layers.6.mlp.gate_proj": {
229
+ "bits": 4,
230
+ "group_size": 64
231
+ },
232
+ "language_model.model.layers.6.mlp.down_proj": {
233
+ "bits": 4,
234
+ "group_size": 64
235
+ },
236
+ "language_model.model.layers.6.mlp.up_proj": {
237
+ "bits": 4,
238
+ "group_size": 64
239
+ },
240
+ "language_model.model.layers.7.self_attn.q_proj": {
241
+ "bits": 8,
242
+ "group_size": 64
243
+ },
244
+ "language_model.model.layers.7.self_attn.k_proj": {
245
+ "bits": 8,
246
+ "group_size": 64
247
+ },
248
+ "language_model.model.layers.7.self_attn.v_proj": {
249
+ "bits": 8,
250
+ "group_size": 64
251
+ },
252
+ "language_model.model.layers.7.self_attn.o_proj": {
253
+ "bits": 4,
254
+ "group_size": 64
255
+ },
256
+ "language_model.model.layers.7.mlp.gate_proj": {
257
+ "bits": 4,
258
+ "group_size": 64
259
+ },
260
+ "language_model.model.layers.7.mlp.down_proj": {
261
+ "bits": 4,
262
+ "group_size": 64
263
+ },
264
+ "language_model.model.layers.7.mlp.up_proj": {
265
+ "bits": 4,
266
+ "group_size": 64
267
+ },
268
+ "language_model.model.layers.8.linear_attn.in_proj_qkv": {
269
+ "bits": 8,
270
+ "group_size": 64
271
+ },
272
+ "language_model.model.layers.8.linear_attn.in_proj_z": {
273
+ "bits": 8,
274
+ "group_size": 64
275
+ },
276
+ "language_model.model.layers.8.linear_attn.in_proj_b": {
277
+ "bits": 8,
278
+ "group_size": 64
279
+ },
280
+ "language_model.model.layers.8.linear_attn.in_proj_a": {
281
+ "bits": 8,
282
+ "group_size": 64
283
+ },
284
+ "language_model.model.layers.8.linear_attn.out_proj": {
285
+ "bits": 4,
286
+ "group_size": 64
287
+ },
288
+ "language_model.model.layers.8.mlp.gate_proj": {
289
+ "bits": 4,
290
+ "group_size": 64
291
+ },
292
+ "language_model.model.layers.8.mlp.down_proj": {
293
+ "bits": 4,
294
+ "group_size": 64
295
+ },
296
+ "language_model.model.layers.8.mlp.up_proj": {
297
+ "bits": 4,
298
+ "group_size": 64
299
+ },
300
+ "language_model.model.layers.9.linear_attn.in_proj_qkv": {
301
+ "bits": 8,
302
+ "group_size": 64
303
+ },
304
+ "language_model.model.layers.9.linear_attn.in_proj_z": {
305
+ "bits": 4,
306
+ "group_size": 64
307
+ },
308
+ "language_model.model.layers.9.linear_attn.in_proj_b": {
309
+ "bits": 8,
310
+ "group_size": 64
311
+ },
312
+ "language_model.model.layers.9.linear_attn.in_proj_a": {
313
+ "bits": 8,
314
+ "group_size": 64
315
+ },
316
+ "language_model.model.layers.9.linear_attn.out_proj": {
317
+ "bits": 8,
318
+ "group_size": 64
319
+ },
320
+ "language_model.model.layers.9.mlp.gate_proj": {
321
+ "bits": 4,
322
+ "group_size": 64
323
+ },
324
+ "language_model.model.layers.9.mlp.down_proj": {
325
+ "bits": 4,
326
+ "group_size": 64
327
+ },
328
+ "language_model.model.layers.9.mlp.up_proj": {
329
+ "bits": 4,
330
+ "group_size": 64
331
+ },
332
+ "language_model.model.layers.10.linear_attn.in_proj_qkv": {
333
+ "bits": 8,
334
+ "group_size": 64
335
+ },
336
+ "language_model.model.layers.10.linear_attn.in_proj_z": {
337
+ "bits": 8,
338
+ "group_size": 64
339
+ },
340
+ "language_model.model.layers.10.linear_attn.in_proj_b": {
341
+ "bits": 8,
342
+ "group_size": 64
343
+ },
344
+ "language_model.model.layers.10.linear_attn.in_proj_a": {
345
+ "bits": 8,
346
+ "group_size": 64
347
+ },
348
+ "language_model.model.layers.10.linear_attn.out_proj": {
349
+ "bits": 4,
350
+ "group_size": 64
351
+ },
352
+ "language_model.model.layers.10.mlp.gate_proj": {
353
+ "bits": 4,
354
+ "group_size": 64
355
+ },
356
+ "language_model.model.layers.10.mlp.down_proj": {
357
+ "bits": 4,
358
+ "group_size": 64
359
+ },
360
+ "language_model.model.layers.10.mlp.up_proj": {
361
+ "bits": 4,
362
+ "group_size": 64
363
+ },
364
+ "language_model.model.layers.11.self_attn.q_proj": {
365
+ "bits": 4,
366
+ "group_size": 64
367
+ },
368
+ "language_model.model.layers.11.self_attn.k_proj": {
369
+ "bits": 4,
370
+ "group_size": 64
371
+ },
372
+ "language_model.model.layers.11.self_attn.v_proj": {
373
+ "bits": 8,
374
+ "group_size": 64
375
+ },
376
+ "language_model.model.layers.11.self_attn.o_proj": {
377
+ "bits": 8,
378
+ "group_size": 64
379
+ },
380
+ "language_model.model.layers.11.mlp.gate_proj": {
381
+ "bits": 4,
382
+ "group_size": 64
383
+ },
384
+ "language_model.model.layers.11.mlp.down_proj": {
385
+ "bits": 4,
386
+ "group_size": 64
387
+ },
388
+ "language_model.model.layers.11.mlp.up_proj": {
389
+ "bits": 4,
390
+ "group_size": 64
391
+ },
392
+ "language_model.model.layers.12.linear_attn.in_proj_qkv": {
393
+ "bits": 8,
394
+ "group_size": 64
395
+ },
396
+ "language_model.model.layers.12.linear_attn.in_proj_z": {
397
+ "bits": 8,
398
+ "group_size": 64
399
+ },
400
+ "language_model.model.layers.12.linear_attn.in_proj_b": {
401
+ "bits": 8,
402
+ "group_size": 64
403
+ },
404
+ "language_model.model.layers.12.linear_attn.in_proj_a": {
405
+ "bits": 8,
406
+ "group_size": 64
407
+ },
408
+ "language_model.model.layers.12.linear_attn.out_proj": {
409
+ "bits": 4,
410
+ "group_size": 64
411
+ },
412
+ "language_model.model.layers.12.mlp.gate_proj": {
413
+ "bits": 4,
414
+ "group_size": 64
415
+ },
416
+ "language_model.model.layers.12.mlp.down_proj": {
417
+ "bits": 4,
418
+ "group_size": 64
419
+ },
420
+ "language_model.model.layers.12.mlp.up_proj": {
421
+ "bits": 4,
422
+ "group_size": 64
423
+ },
424
+ "language_model.model.layers.13.linear_attn.in_proj_qkv": {
425
+ "bits": 8,
426
+ "group_size": 64
427
+ },
428
+ "language_model.model.layers.13.linear_attn.in_proj_z": {
429
+ "bits": 8,
430
+ "group_size": 64
431
+ },
432
+ "language_model.model.layers.13.linear_attn.in_proj_b": {
433
+ "bits": 8,
434
+ "group_size": 64
435
+ },
436
+ "language_model.model.layers.13.linear_attn.in_proj_a": {
437
+ "bits": 8,
438
+ "group_size": 64
439
+ },
440
+ "language_model.model.layers.13.linear_attn.out_proj": {
441
+ "bits": 4,
442
+ "group_size": 64
443
+ },
444
+ "language_model.model.layers.13.mlp.gate_proj": {
445
+ "bits": 4,
446
+ "group_size": 64
447
+ },
448
+ "language_model.model.layers.13.mlp.down_proj": {
449
+ "bits": 4,
450
+ "group_size": 64
451
+ },
452
+ "language_model.model.layers.13.mlp.up_proj": {
453
+ "bits": 4,
454
+ "group_size": 64
455
+ },
456
+ "language_model.model.layers.14.linear_attn.in_proj_qkv": {
457
+ "bits": 8,
458
+ "group_size": 64
459
+ },
460
+ "language_model.model.layers.14.linear_attn.in_proj_z": {
461
+ "bits": 8,
462
+ "group_size": 64
463
+ },
464
+ "language_model.model.layers.14.linear_attn.in_proj_b": {
465
+ "bits": 8,
466
+ "group_size": 64
467
+ },
468
+ "language_model.model.layers.14.linear_attn.in_proj_a": {
469
+ "bits": 8,
470
+ "group_size": 64
471
+ },
472
+ "language_model.model.layers.14.linear_attn.out_proj": {
473
+ "bits": 4,
474
+ "group_size": 64
475
+ },
476
+ "language_model.model.layers.14.mlp.gate_proj": {
477
+ "bits": 4,
478
+ "group_size": 64
479
+ },
480
+ "language_model.model.layers.14.mlp.down_proj": {
481
+ "bits": 4,
482
+ "group_size": 64
483
+ },
484
+ "language_model.model.layers.14.mlp.up_proj": {
485
+ "bits": 4,
486
+ "group_size": 64
487
+ },
488
+ "language_model.model.layers.15.self_attn.q_proj": {
489
+ "bits": 4,
490
+ "group_size": 64
491
+ },
492
+ "language_model.model.layers.15.self_attn.k_proj": {
493
+ "bits": 4,
494
+ "group_size": 64
495
+ },
496
+ "language_model.model.layers.15.self_attn.v_proj": {
497
+ "bits": 8,
498
+ "group_size": 64
499
+ },
500
+ "language_model.model.layers.15.self_attn.o_proj": {
501
+ "bits": 8,
502
+ "group_size": 64
503
+ },
504
+ "language_model.model.layers.15.mlp.gate_proj": {
505
+ "bits": 4,
506
+ "group_size": 64
507
+ },
508
+ "language_model.model.layers.15.mlp.down_proj": {
509
+ "bits": 4,
510
+ "group_size": 64
511
+ },
512
+ "language_model.model.layers.15.mlp.up_proj": {
513
+ "bits": 4,
514
+ "group_size": 64
515
+ },
516
+ "language_model.model.layers.16.linear_attn.in_proj_qkv": {
517
+ "bits": 8,
518
+ "group_size": 64
519
+ },
520
+ "language_model.model.layers.16.linear_attn.in_proj_z": {
521
+ "bits": 8,
522
+ "group_size": 64
523
+ },
524
+ "language_model.model.layers.16.linear_attn.in_proj_b": {
525
+ "bits": 8,
526
+ "group_size": 64
527
+ },
528
+ "language_model.model.layers.16.linear_attn.in_proj_a": {
529
+ "bits": 8,
530
+ "group_size": 64
531
+ },
532
+ "language_model.model.layers.16.linear_attn.out_proj": {
533
+ "bits": 4,
534
+ "group_size": 64
535
+ },
536
+ "language_model.model.layers.16.mlp.gate_proj": {
537
+ "bits": 4,
538
+ "group_size": 64
539
+ },
540
+ "language_model.model.layers.16.mlp.down_proj": {
541
+ "bits": 4,
542
+ "group_size": 64
543
+ },
544
+ "language_model.model.layers.16.mlp.up_proj": {
545
+ "bits": 4,
546
+ "group_size": 64
547
+ },
548
+ "language_model.model.layers.17.linear_attn.in_proj_qkv": {
549
+ "bits": 8,
550
+ "group_size": 64
551
+ },
552
+ "language_model.model.layers.17.linear_attn.in_proj_z": {
553
+ "bits": 8,
554
+ "group_size": 64
555
+ },
556
+ "language_model.model.layers.17.linear_attn.in_proj_b": {
557
+ "bits": 4,
558
+ "group_size": 64
559
+ },
560
+ "language_model.model.layers.17.linear_attn.in_proj_a": {
561
+ "bits": 8,
562
+ "group_size": 64
563
+ },
564
+ "language_model.model.layers.17.linear_attn.out_proj": {
565
+ "bits": 4,
566
+ "group_size": 64
567
+ },
568
+ "language_model.model.layers.17.mlp.gate_proj": {
569
+ "bits": 4,
570
+ "group_size": 64
571
+ },
572
+ "language_model.model.layers.17.mlp.down_proj": {
573
+ "bits": 4,
574
+ "group_size": 64
575
+ },
576
+ "language_model.model.layers.17.mlp.up_proj": {
577
+ "bits": 8,
578
+ "group_size": 64
579
+ },
580
+ "language_model.model.layers.18.linear_attn.in_proj_qkv": {
581
+ "bits": 8,
582
+ "group_size": 64
583
+ },
584
+ "language_model.model.layers.18.linear_attn.in_proj_z": {
585
+ "bits": 4,
586
+ "group_size": 64
587
+ },
588
+ "language_model.model.layers.18.linear_attn.in_proj_b": {
589
+ "bits": 4,
590
+ "group_size": 64
591
+ },
592
+ "language_model.model.layers.18.linear_attn.in_proj_a": {
593
+ "bits": 8,
594
+ "group_size": 64
595
+ },
596
+ "language_model.model.layers.18.linear_attn.out_proj": {
597
+ "bits": 8,
598
+ "group_size": 64
599
+ },
600
+ "language_model.model.layers.18.mlp.gate_proj": {
601
+ "bits": 4,
602
+ "group_size": 64
603
+ },
604
+ "language_model.model.layers.18.mlp.down_proj": {
605
+ "bits": 4,
606
+ "group_size": 64
607
+ },
608
+ "language_model.model.layers.18.mlp.up_proj": {
609
+ "bits": 4,
610
+ "group_size": 64
611
+ },
612
+ "language_model.model.layers.19.self_attn.q_proj": {
613
+ "bits": 4,
614
+ "group_size": 64
615
+ },
616
+ "language_model.model.layers.19.self_attn.k_proj": {
617
+ "bits": 4,
618
+ "group_size": 64
619
+ },
620
+ "language_model.model.layers.19.self_attn.v_proj": {
621
+ "bits": 8,
622
+ "group_size": 64
623
+ },
624
+ "language_model.model.layers.19.self_attn.o_proj": {
625
+ "bits": 8,
626
+ "group_size": 64
627
+ },
628
+ "language_model.model.layers.19.mlp.gate_proj": {
629
+ "bits": 4,
630
+ "group_size": 64
631
+ },
632
+ "language_model.model.layers.19.mlp.down_proj": {
633
+ "bits": 4,
634
+ "group_size": 64
635
+ },
636
+ "language_model.model.layers.19.mlp.up_proj": {
637
+ "bits": 4,
638
+ "group_size": 64
639
+ },
640
+ "language_model.model.layers.20.linear_attn.in_proj_qkv": {
641
+ "bits": 8,
642
+ "group_size": 64
643
+ },
644
+ "language_model.model.layers.20.linear_attn.in_proj_z": {
645
+ "bits": 4,
646
+ "group_size": 64
647
+ },
648
+ "language_model.model.layers.20.linear_attn.in_proj_b": {
649
+ "bits": 4,
650
+ "group_size": 64
651
+ },
652
+ "language_model.model.layers.20.linear_attn.in_proj_a": {
653
+ "bits": 4,
654
+ "group_size": 64
655
+ },
656
+ "language_model.model.layers.20.linear_attn.out_proj": {
657
+ "bits": 8,
658
+ "group_size": 64
659
+ },
660
+ "language_model.model.layers.20.mlp.gate_proj": {
661
+ "bits": 4,
662
+ "group_size": 64
663
+ },
664
+ "language_model.model.layers.20.mlp.down_proj": {
665
+ "bits": 8,
666
+ "group_size": 64
667
+ },
668
+ "language_model.model.layers.20.mlp.up_proj": {
669
+ "bits": 4,
670
+ "group_size": 64
671
+ },
672
+ "language_model.model.layers.21.linear_attn.in_proj_qkv": {
673
+ "bits": 8,
674
+ "group_size": 64
675
+ },
676
+ "language_model.model.layers.21.linear_attn.in_proj_z": {
677
+ "bits": 4,
678
+ "group_size": 64
679
+ },
680
+ "language_model.model.layers.21.linear_attn.in_proj_b": {
681
+ "bits": 4,
682
+ "group_size": 64
683
+ },
684
+ "language_model.model.layers.21.linear_attn.in_proj_a": {
685
+ "bits": 4,
686
+ "group_size": 64
687
+ },
688
+ "language_model.model.layers.21.linear_attn.out_proj": {
689
+ "bits": 8,
690
+ "group_size": 64
691
+ },
692
+ "language_model.model.layers.21.mlp.gate_proj": {
693
+ "bits": 4,
694
+ "group_size": 64
695
+ },
696
+ "language_model.model.layers.21.mlp.down_proj": {
697
+ "bits": 4,
698
+ "group_size": 64
699
+ },
700
+ "language_model.model.layers.21.mlp.up_proj": {
701
+ "bits": 4,
702
+ "group_size": 64
703
+ },
704
+ "language_model.model.layers.22.linear_attn.in_proj_qkv": {
705
+ "bits": 8,
706
+ "group_size": 64
707
+ },
708
+ "language_model.model.layers.22.linear_attn.in_proj_z": {
709
+ "bits": 4,
710
+ "group_size": 64
711
+ },
712
+ "language_model.model.layers.22.linear_attn.in_proj_b": {
713
+ "bits": 8,
714
+ "group_size": 64
715
+ },
716
+ "language_model.model.layers.22.linear_attn.in_proj_a": {
717
+ "bits": 4,
718
+ "group_size": 64
719
+ },
720
+ "language_model.model.layers.22.linear_attn.out_proj": {
721
+ "bits": 8,
722
+ "group_size": 64
723
+ },
724
+ "language_model.model.layers.22.mlp.gate_proj": {
725
+ "bits": 4,
726
+ "group_size": 64
727
+ },
728
+ "language_model.model.layers.22.mlp.down_proj": {
729
+ "bits": 4,
730
+ "group_size": 64
731
+ },
732
+ "language_model.model.layers.22.mlp.up_proj": {
733
+ "bits": 4,
734
+ "group_size": 64
735
+ },
736
+ "language_model.model.layers.23.self_attn.q_proj": {
737
+ "bits": 4,
738
+ "group_size": 64
739
+ },
740
+ "language_model.model.layers.23.self_attn.k_proj": {
741
+ "bits": 4,
742
+ "group_size": 64
743
+ },
744
+ "language_model.model.layers.23.self_attn.v_proj": {
745
+ "bits": 8,
746
+ "group_size": 64
747
+ },
748
+ "language_model.model.layers.23.self_attn.o_proj": {
749
+ "bits": 8,
750
+ "group_size": 64
751
+ },
752
+ "language_model.model.layers.23.mlp.gate_proj": {
753
+ "bits": 4,
754
+ "group_size": 64
755
+ },
756
+ "language_model.model.layers.23.mlp.down_proj": {
757
+ "bits": 8,
758
+ "group_size": 64
759
+ },
760
+ "language_model.model.layers.23.mlp.up_proj": {
761
+ "bits": 4,
762
+ "group_size": 64
763
+ },
764
+ "language_model.model.layers.24.linear_attn.in_proj_qkv": {
765
+ "bits": 8,
766
+ "group_size": 64
767
+ },
768
+ "language_model.model.layers.24.linear_attn.in_proj_z": {
769
+ "bits": 4,
770
+ "group_size": 64
771
+ },
772
+ "language_model.model.layers.24.linear_attn.in_proj_b": {
773
+ "bits": 4,
774
+ "group_size": 64
775
+ },
776
+ "language_model.model.layers.24.linear_attn.in_proj_a": {
777
+ "bits": 4,
778
+ "group_size": 64
779
+ },
780
+ "language_model.model.layers.24.linear_attn.out_proj": {
781
+ "bits": 8,
782
+ "group_size": 64
783
+ },
784
+ "language_model.model.layers.24.mlp.gate_proj": {
785
+ "bits": 4,
786
+ "group_size": 64
787
+ },
788
+ "language_model.model.layers.24.mlp.down_proj": {
789
+ "bits": 4,
790
+ "group_size": 64
791
+ },
792
+ "language_model.model.layers.24.mlp.up_proj": {
793
+ "bits": 4,
794
+ "group_size": 64
795
+ },
796
+ "language_model.model.layers.25.linear_attn.in_proj_qkv": {
797
+ "bits": 8,
798
+ "group_size": 64
799
+ },
800
+ "language_model.model.layers.25.linear_attn.in_proj_z": {
801
+ "bits": 4,
802
+ "group_size": 64
803
+ },
804
+ "language_model.model.layers.25.linear_attn.in_proj_b": {
805
+ "bits": 4,
806
+ "group_size": 64
807
+ },
808
+ "language_model.model.layers.25.linear_attn.in_proj_a": {
809
+ "bits": 4,
810
+ "group_size": 64
811
+ },
812
+ "language_model.model.layers.25.linear_attn.out_proj": {
813
+ "bits": 8,
814
+ "group_size": 64
815
+ },
816
+ "language_model.model.layers.25.mlp.gate_proj": {
817
+ "bits": 4,
818
+ "group_size": 64
819
+ },
820
+ "language_model.model.layers.25.mlp.down_proj": {
821
+ "bits": 4,
822
+ "group_size": 64
823
+ },
824
+ "language_model.model.layers.25.mlp.up_proj": {
825
+ "bits": 4,
826
+ "group_size": 64
827
+ },
828
+ "language_model.model.layers.26.linear_attn.in_proj_qkv": {
829
+ "bits": 8,
830
+ "group_size": 64
831
+ },
832
+ "language_model.model.layers.26.linear_attn.in_proj_z": {
833
+ "bits": 8,
834
+ "group_size": 64
835
+ },
836
+ "language_model.model.layers.26.linear_attn.in_proj_b": {
837
+ "bits": 4,
838
+ "group_size": 64
839
+ },
840
+ "language_model.model.layers.26.linear_attn.in_proj_a": {
841
+ "bits": 4,
842
+ "group_size": 64
843
+ },
844
+ "language_model.model.layers.26.linear_attn.out_proj": {
845
+ "bits": 4,
846
+ "group_size": 64
847
+ },
848
+ "language_model.model.layers.26.mlp.gate_proj": {
849
+ "bits": 4,
850
+ "group_size": 64
851
+ },
852
+ "language_model.model.layers.26.mlp.down_proj": {
853
+ "bits": 8,
854
+ "group_size": 64
855
+ },
856
+ "language_model.model.layers.26.mlp.up_proj": {
857
+ "bits": 4,
858
+ "group_size": 64
859
+ },
860
+ "language_model.model.layers.27.self_attn.q_proj": {
861
+ "bits": 4,
862
+ "group_size": 64
863
+ },
864
+ "language_model.model.layers.27.self_attn.k_proj": {
865
+ "bits": 4,
866
+ "group_size": 64
867
+ },
868
+ "language_model.model.layers.27.self_attn.v_proj": {
869
+ "bits": 8,
870
+ "group_size": 64
871
+ },
872
+ "language_model.model.layers.27.self_attn.o_proj": {
873
+ "bits": 8,
874
+ "group_size": 64
875
+ },
876
+ "language_model.model.layers.27.mlp.gate_proj": {
877
+ "bits": 4,
878
+ "group_size": 64
879
+ },
880
+ "language_model.model.layers.27.mlp.down_proj": {
881
+ "bits": 4,
882
+ "group_size": 64
883
+ },
884
+ "language_model.model.layers.27.mlp.up_proj": {
885
+ "bits": 4,
886
+ "group_size": 64
887
+ },
888
+ "language_model.model.layers.28.linear_attn.in_proj_qkv": {
889
+ "bits": 8,
890
+ "group_size": 64
891
+ },
892
+ "language_model.model.layers.28.linear_attn.in_proj_z": {
893
+ "bits": 4,
894
+ "group_size": 64
895
+ },
896
+ "language_model.model.layers.28.linear_attn.in_proj_b": {
897
+ "bits": 4,
898
+ "group_size": 64
899
+ },
900
+ "language_model.model.layers.28.linear_attn.in_proj_a": {
901
+ "bits": 4,
902
+ "group_size": 64
903
+ },
904
+ "language_model.model.layers.28.linear_attn.out_proj": {
905
+ "bits": 8,
906
+ "group_size": 64
907
+ },
908
+ "language_model.model.layers.28.mlp.gate_proj": {
909
+ "bits": 4,
910
+ "group_size": 64
911
+ },
912
+ "language_model.model.layers.28.mlp.down_proj": {
913
+ "bits": 4,
914
+ "group_size": 64
915
+ },
916
+ "language_model.model.layers.28.mlp.up_proj": {
917
+ "bits": 4,
918
+ "group_size": 64
919
+ },
920
+ "language_model.model.layers.29.linear_attn.in_proj_qkv": {
921
+ "bits": 8,
922
+ "group_size": 64
923
+ },
924
+ "language_model.model.layers.29.linear_attn.in_proj_z": {
925
+ "bits": 8,
926
+ "group_size": 64
927
+ },
928
+ "language_model.model.layers.29.linear_attn.in_proj_b": {
929
+ "bits": 4,
930
+ "group_size": 64
931
+ },
932
+ "language_model.model.layers.29.linear_attn.in_proj_a": {
933
+ "bits": 4,
934
+ "group_size": 64
935
+ },
936
+ "language_model.model.layers.29.linear_attn.out_proj": {
937
+ "bits": 4,
938
+ "group_size": 64
939
+ },
940
+ "language_model.model.layers.29.mlp.gate_proj": {
941
+ "bits": 4,
942
+ "group_size": 64
943
+ },
944
+ "language_model.model.layers.29.mlp.down_proj": {
945
+ "bits": 4,
946
+ "group_size": 64
947
+ },
948
+ "language_model.model.layers.29.mlp.up_proj": {
949
+ "bits": 8,
950
+ "group_size": 64
951
+ },
952
+ "language_model.model.layers.30.linear_attn.in_proj_qkv": {
953
+ "bits": 8,
954
+ "group_size": 64
955
+ },
956
+ "language_model.model.layers.30.linear_attn.in_proj_z": {
957
+ "bits": 4,
958
+ "group_size": 64
959
+ },
960
+ "language_model.model.layers.30.linear_attn.in_proj_b": {
961
+ "bits": 4,
962
+ "group_size": 64
963
+ },
964
+ "language_model.model.layers.30.linear_attn.in_proj_a": {
965
+ "bits": 4,
966
+ "group_size": 64
967
+ },
968
+ "language_model.model.layers.30.linear_attn.out_proj": {
969
+ "bits": 8,
970
+ "group_size": 64
971
+ },
972
+ "language_model.model.layers.30.mlp.gate_proj": {
973
+ "bits": 4,
974
+ "group_size": 64
975
+ },
976
+ "language_model.model.layers.30.mlp.down_proj": {
977
+ "bits": 4,
978
+ "group_size": 64
979
+ },
980
+ "language_model.model.layers.30.mlp.up_proj": {
981
+ "bits": 4,
982
+ "group_size": 64
983
+ },
984
+ "language_model.model.layers.31.self_attn.q_proj": {
985
+ "bits": 8,
986
+ "group_size": 64
987
+ },
988
+ "language_model.model.layers.31.self_attn.k_proj": {
989
+ "bits": 8,
990
+ "group_size": 64
991
+ },
992
+ "language_model.model.layers.31.self_attn.v_proj": {
993
+ "bits": 8,
994
+ "group_size": 64
995
+ },
996
+ "language_model.model.layers.31.self_attn.o_proj": {
997
+ "bits": 8,
998
+ "group_size": 64
999
+ },
1000
+ "language_model.model.layers.31.mlp.gate_proj": {
1001
+ "bits": 8,
1002
+ "group_size": 64
1003
+ },
1004
+ "language_model.model.layers.31.mlp.down_proj": {
1005
+ "bits": 8,
1006
+ "group_size": 64
1007
+ },
1008
+ "language_model.model.layers.31.mlp.up_proj": {
1009
+ "bits": 8,
1010
+ "group_size": 64
1011
+ },
1012
+ "language_model.lm_head": {
1013
+ "bits": 8,
1014
+ "group_size": 64
1015
+ }
1016
+ },
1017
+ "quantization_config": {
1018
+ "group_size": 64,
1019
+ "bits": 4,
1020
+ "mode": "affine",
1021
+ "language_model.model.embed_tokens": {
1022
+ "bits": 8,
1023
+ "group_size": 64
1024
+ },
1025
+ "language_model.model.layers.0.linear_attn.in_proj_qkv": {
1026
+ "bits": 8,
1027
+ "group_size": 64
1028
+ },
1029
+ "language_model.model.layers.0.linear_attn.in_proj_z": {
1030
+ "bits": 8,
1031
+ "group_size": 64
1032
+ },
1033
+ "language_model.model.layers.0.linear_attn.in_proj_b": {
1034
+ "bits": 8,
1035
+ "group_size": 64
1036
+ },
1037
+ "language_model.model.layers.0.linear_attn.in_proj_a": {
1038
+ "bits": 8,
1039
+ "group_size": 64
1040
+ },
1041
+ "language_model.model.layers.0.linear_attn.out_proj": {
1042
+ "bits": 8,
1043
+ "group_size": 64
1044
+ },
1045
+ "language_model.model.layers.0.mlp.gate_proj": {
1046
+ "bits": 8,
1047
+ "group_size": 64
1048
+ },
1049
+ "language_model.model.layers.0.mlp.down_proj": {
1050
+ "bits": 8,
1051
+ "group_size": 64
1052
+ },
1053
+ "language_model.model.layers.0.mlp.up_proj": {
1054
+ "bits": 8,
1055
+ "group_size": 64
1056
+ },
1057
+ "language_model.model.layers.1.linear_attn.in_proj_qkv": {
1058
+ "bits": 8,
1059
+ "group_size": 64
1060
+ },
1061
+ "language_model.model.layers.1.linear_attn.in_proj_z": {
1062
+ "bits": 8,
1063
+ "group_size": 64
1064
+ },
1065
+ "language_model.model.layers.1.linear_attn.in_proj_b": {
1066
+ "bits": 8,
1067
+ "group_size": 64
1068
+ },
1069
+ "language_model.model.layers.1.linear_attn.in_proj_a": {
1070
+ "bits": 8,
1071
+ "group_size": 64
1072
+ },
1073
+ "language_model.model.layers.1.linear_attn.out_proj": {
1074
+ "bits": 4,
1075
+ "group_size": 64
1076
+ },
1077
+ "language_model.model.layers.1.mlp.gate_proj": {
1078
+ "bits": 4,
1079
+ "group_size": 64
1080
+ },
1081
+ "language_model.model.layers.1.mlp.down_proj": {
1082
+ "bits": 4,
1083
+ "group_size": 64
1084
+ },
1085
+ "language_model.model.layers.1.mlp.up_proj": {
1086
+ "bits": 4,
1087
+ "group_size": 64
1088
+ },
1089
+ "language_model.model.layers.2.linear_attn.in_proj_qkv": {
1090
+ "bits": 8,
1091
+ "group_size": 64
1092
+ },
1093
+ "language_model.model.layers.2.linear_attn.in_proj_z": {
1094
+ "bits": 4,
1095
+ "group_size": 64
1096
+ },
1097
+ "language_model.model.layers.2.linear_attn.in_proj_b": {
1098
+ "bits": 8,
1099
+ "group_size": 64
1100
+ },
1101
+ "language_model.model.layers.2.linear_attn.in_proj_a": {
1102
+ "bits": 8,
1103
+ "group_size": 64
1104
+ },
1105
+ "language_model.model.layers.2.linear_attn.out_proj": {
1106
+ "bits": 4,
1107
+ "group_size": 64
1108
+ },
1109
+ "language_model.model.layers.2.mlp.gate_proj": {
1110
+ "bits": 4,
1111
+ "group_size": 64
1112
+ },
1113
+ "language_model.model.layers.2.mlp.down_proj": {
1114
+ "bits": 4,
1115
+ "group_size": 64
1116
+ },
1117
+ "language_model.model.layers.2.mlp.up_proj": {
1118
+ "bits": 4,
1119
+ "group_size": 64
1120
+ },
1121
+ "language_model.model.layers.3.self_attn.q_proj": {
1122
+ "bits": 4,
1123
+ "group_size": 64
1124
+ },
1125
+ "language_model.model.layers.3.self_attn.k_proj": {
1126
+ "bits": 8,
1127
+ "group_size": 64
1128
+ },
1129
+ "language_model.model.layers.3.self_attn.v_proj": {
1130
+ "bits": 8,
1131
+ "group_size": 64
1132
+ },
1133
+ "language_model.model.layers.3.self_attn.o_proj": {
1134
+ "bits": 8,
1135
+ "group_size": 64
1136
+ },
1137
+ "language_model.model.layers.3.mlp.gate_proj": {
1138
+ "bits": 4,
1139
+ "group_size": 64
1140
+ },
1141
+ "language_model.model.layers.3.mlp.down_proj": {
1142
+ "bits": 4,
1143
+ "group_size": 64
1144
+ },
1145
+ "language_model.model.layers.3.mlp.up_proj": {
1146
+ "bits": 4,
1147
+ "group_size": 64
1148
+ },
1149
+ "language_model.model.layers.4.linear_attn.in_proj_qkv": {
1150
+ "bits": 8,
1151
+ "group_size": 64
1152
+ },
1153
+ "language_model.model.layers.4.linear_attn.in_proj_z": {
1154
+ "bits": 8,
1155
+ "group_size": 64
1156
+ },
1157
+ "language_model.model.layers.4.linear_attn.in_proj_b": {
1158
+ "bits": 8,
1159
+ "group_size": 64
1160
+ },
1161
+ "language_model.model.layers.4.linear_attn.in_proj_a": {
1162
+ "bits": 8,
1163
+ "group_size": 64
1164
+ },
1165
+ "language_model.model.layers.4.linear_attn.out_proj": {
1166
+ "bits": 4,
1167
+ "group_size": 64
1168
+ },
1169
+ "language_model.model.layers.4.mlp.gate_proj": {
1170
+ "bits": 4,
1171
+ "group_size": 64
1172
+ },
1173
+ "language_model.model.layers.4.mlp.down_proj": {
1174
+ "bits": 4,
1175
+ "group_size": 64
1176
+ },
1177
+ "language_model.model.layers.4.mlp.up_proj": {
1178
+ "bits": 4,
1179
+ "group_size": 64
1180
+ },
1181
+ "language_model.model.layers.5.linear_attn.in_proj_qkv": {
1182
+ "bits": 8,
1183
+ "group_size": 64
1184
+ },
1185
+ "language_model.model.layers.5.linear_attn.in_proj_z": {
1186
+ "bits": 4,
1187
+ "group_size": 64
1188
+ },
1189
+ "language_model.model.layers.5.linear_attn.in_proj_b": {
1190
+ "bits": 8,
1191
+ "group_size": 64
1192
+ },
1193
+ "language_model.model.layers.5.linear_attn.in_proj_a": {
1194
+ "bits": 8,
1195
+ "group_size": 64
1196
+ },
1197
+ "language_model.model.layers.5.linear_attn.out_proj": {
1198
+ "bits": 8,
1199
+ "group_size": 64
1200
+ },
1201
+ "language_model.model.layers.5.mlp.gate_proj": {
1202
+ "bits": 4,
1203
+ "group_size": 64
1204
+ },
1205
+ "language_model.model.layers.5.mlp.down_proj": {
1206
+ "bits": 4,
1207
+ "group_size": 64
1208
+ },
1209
+ "language_model.model.layers.5.mlp.up_proj": {
1210
+ "bits": 4,
1211
+ "group_size": 64
1212
+ },
1213
+ "language_model.model.layers.6.linear_attn.in_proj_qkv": {
1214
+ "bits": 8,
1215
+ "group_size": 64
1216
+ },
1217
+ "language_model.model.layers.6.linear_attn.in_proj_z": {
1218
+ "bits": 8,
1219
+ "group_size": 64
1220
+ },
1221
+ "language_model.model.layers.6.linear_attn.in_proj_b": {
1222
+ "bits": 8,
1223
+ "group_size": 64
1224
+ },
1225
+ "language_model.model.layers.6.linear_attn.in_proj_a": {
1226
+ "bits": 8,
1227
+ "group_size": 64
1228
+ },
1229
+ "language_model.model.layers.6.linear_attn.out_proj": {
1230
+ "bits": 4,
1231
+ "group_size": 64
1232
+ },
1233
+ "language_model.model.layers.6.mlp.gate_proj": {
1234
+ "bits": 4,
1235
+ "group_size": 64
1236
+ },
1237
+ "language_model.model.layers.6.mlp.down_proj": {
1238
+ "bits": 4,
1239
+ "group_size": 64
1240
+ },
1241
+ "language_model.model.layers.6.mlp.up_proj": {
1242
+ "bits": 4,
1243
+ "group_size": 64
1244
+ },
1245
+ "language_model.model.layers.7.self_attn.q_proj": {
1246
+ "bits": 8,
1247
+ "group_size": 64
1248
+ },
1249
+ "language_model.model.layers.7.self_attn.k_proj": {
1250
+ "bits": 8,
1251
+ "group_size": 64
1252
+ },
1253
+ "language_model.model.layers.7.self_attn.v_proj": {
1254
+ "bits": 8,
1255
+ "group_size": 64
1256
+ },
1257
+ "language_model.model.layers.7.self_attn.o_proj": {
1258
+ "bits": 4,
1259
+ "group_size": 64
1260
+ },
1261
+ "language_model.model.layers.7.mlp.gate_proj": {
1262
+ "bits": 4,
1263
+ "group_size": 64
1264
+ },
1265
+ "language_model.model.layers.7.mlp.down_proj": {
1266
+ "bits": 4,
1267
+ "group_size": 64
1268
+ },
1269
+ "language_model.model.layers.7.mlp.up_proj": {
1270
+ "bits": 4,
1271
+ "group_size": 64
1272
+ },
1273
+ "language_model.model.layers.8.linear_attn.in_proj_qkv": {
1274
+ "bits": 8,
1275
+ "group_size": 64
1276
+ },
1277
+ "language_model.model.layers.8.linear_attn.in_proj_z": {
1278
+ "bits": 8,
1279
+ "group_size": 64
1280
+ },
1281
+ "language_model.model.layers.8.linear_attn.in_proj_b": {
1282
+ "bits": 8,
1283
+ "group_size": 64
1284
+ },
1285
+ "language_model.model.layers.8.linear_attn.in_proj_a": {
1286
+ "bits": 8,
1287
+ "group_size": 64
1288
+ },
1289
+ "language_model.model.layers.8.linear_attn.out_proj": {
1290
+ "bits": 4,
1291
+ "group_size": 64
1292
+ },
1293
+ "language_model.model.layers.8.mlp.gate_proj": {
1294
+ "bits": 4,
1295
+ "group_size": 64
1296
+ },
1297
+ "language_model.model.layers.8.mlp.down_proj": {
1298
+ "bits": 4,
1299
+ "group_size": 64
1300
+ },
1301
+ "language_model.model.layers.8.mlp.up_proj": {
1302
+ "bits": 4,
1303
+ "group_size": 64
1304
+ },
1305
+ "language_model.model.layers.9.linear_attn.in_proj_qkv": {
1306
+ "bits": 8,
1307
+ "group_size": 64
1308
+ },
1309
+ "language_model.model.layers.9.linear_attn.in_proj_z": {
1310
+ "bits": 4,
1311
+ "group_size": 64
1312
+ },
1313
+ "language_model.model.layers.9.linear_attn.in_proj_b": {
1314
+ "bits": 8,
1315
+ "group_size": 64
1316
+ },
1317
+ "language_model.model.layers.9.linear_attn.in_proj_a": {
1318
+ "bits": 8,
1319
+ "group_size": 64
1320
+ },
1321
+ "language_model.model.layers.9.linear_attn.out_proj": {
1322
+ "bits": 8,
1323
+ "group_size": 64
1324
+ },
1325
+ "language_model.model.layers.9.mlp.gate_proj": {
1326
+ "bits": 4,
1327
+ "group_size": 64
1328
+ },
1329
+ "language_model.model.layers.9.mlp.down_proj": {
1330
+ "bits": 4,
1331
+ "group_size": 64
1332
+ },
1333
+ "language_model.model.layers.9.mlp.up_proj": {
1334
+ "bits": 4,
1335
+ "group_size": 64
1336
+ },
1337
+ "language_model.model.layers.10.linear_attn.in_proj_qkv": {
1338
+ "bits": 8,
1339
+ "group_size": 64
1340
+ },
1341
+ "language_model.model.layers.10.linear_attn.in_proj_z": {
1342
+ "bits": 8,
1343
+ "group_size": 64
1344
+ },
1345
+ "language_model.model.layers.10.linear_attn.in_proj_b": {
1346
+ "bits": 8,
1347
+ "group_size": 64
1348
+ },
1349
+ "language_model.model.layers.10.linear_attn.in_proj_a": {
1350
+ "bits": 8,
1351
+ "group_size": 64
1352
+ },
1353
+ "language_model.model.layers.10.linear_attn.out_proj": {
1354
+ "bits": 4,
1355
+ "group_size": 64
1356
+ },
1357
+ "language_model.model.layers.10.mlp.gate_proj": {
1358
+ "bits": 4,
1359
+ "group_size": 64
1360
+ },
1361
+ "language_model.model.layers.10.mlp.down_proj": {
1362
+ "bits": 4,
1363
+ "group_size": 64
1364
+ },
1365
+ "language_model.model.layers.10.mlp.up_proj": {
1366
+ "bits": 4,
1367
+ "group_size": 64
1368
+ },
1369
+ "language_model.model.layers.11.self_attn.q_proj": {
1370
+ "bits": 4,
1371
+ "group_size": 64
1372
+ },
1373
+ "language_model.model.layers.11.self_attn.k_proj": {
1374
+ "bits": 4,
1375
+ "group_size": 64
1376
+ },
1377
+ "language_model.model.layers.11.self_attn.v_proj": {
1378
+ "bits": 8,
1379
+ "group_size": 64
1380
+ },
1381
+ "language_model.model.layers.11.self_attn.o_proj": {
1382
+ "bits": 8,
1383
+ "group_size": 64
1384
+ },
1385
+ "language_model.model.layers.11.mlp.gate_proj": {
1386
+ "bits": 4,
1387
+ "group_size": 64
1388
+ },
1389
+ "language_model.model.layers.11.mlp.down_proj": {
1390
+ "bits": 4,
1391
+ "group_size": 64
1392
+ },
1393
+ "language_model.model.layers.11.mlp.up_proj": {
1394
+ "bits": 4,
1395
+ "group_size": 64
1396
+ },
1397
+ "language_model.model.layers.12.linear_attn.in_proj_qkv": {
1398
+ "bits": 8,
1399
+ "group_size": 64
1400
+ },
1401
+ "language_model.model.layers.12.linear_attn.in_proj_z": {
1402
+ "bits": 8,
1403
+ "group_size": 64
1404
+ },
1405
+ "language_model.model.layers.12.linear_attn.in_proj_b": {
1406
+ "bits": 8,
1407
+ "group_size": 64
1408
+ },
1409
+ "language_model.model.layers.12.linear_attn.in_proj_a": {
1410
+ "bits": 8,
1411
+ "group_size": 64
1412
+ },
1413
+ "language_model.model.layers.12.linear_attn.out_proj": {
1414
+ "bits": 4,
1415
+ "group_size": 64
1416
+ },
1417
+ "language_model.model.layers.12.mlp.gate_proj": {
1418
+ "bits": 4,
1419
+ "group_size": 64
1420
+ },
1421
+ "language_model.model.layers.12.mlp.down_proj": {
1422
+ "bits": 4,
1423
+ "group_size": 64
1424
+ },
1425
+ "language_model.model.layers.12.mlp.up_proj": {
1426
+ "bits": 4,
1427
+ "group_size": 64
1428
+ },
1429
+ "language_model.model.layers.13.linear_attn.in_proj_qkv": {
1430
+ "bits": 8,
1431
+ "group_size": 64
1432
+ },
1433
+ "language_model.model.layers.13.linear_attn.in_proj_z": {
1434
+ "bits": 8,
1435
+ "group_size": 64
1436
+ },
1437
+ "language_model.model.layers.13.linear_attn.in_proj_b": {
1438
+ "bits": 8,
1439
+ "group_size": 64
1440
+ },
1441
+ "language_model.model.layers.13.linear_attn.in_proj_a": {
1442
+ "bits": 8,
1443
+ "group_size": 64
1444
+ },
1445
+ "language_model.model.layers.13.linear_attn.out_proj": {
1446
+ "bits": 4,
1447
+ "group_size": 64
1448
+ },
1449
+ "language_model.model.layers.13.mlp.gate_proj": {
1450
+ "bits": 4,
1451
+ "group_size": 64
1452
+ },
1453
+ "language_model.model.layers.13.mlp.down_proj": {
1454
+ "bits": 4,
1455
+ "group_size": 64
1456
+ },
1457
+ "language_model.model.layers.13.mlp.up_proj": {
1458
+ "bits": 4,
1459
+ "group_size": 64
1460
+ },
1461
+ "language_model.model.layers.14.linear_attn.in_proj_qkv": {
1462
+ "bits": 8,
1463
+ "group_size": 64
1464
+ },
1465
+ "language_model.model.layers.14.linear_attn.in_proj_z": {
1466
+ "bits": 8,
1467
+ "group_size": 64
1468
+ },
1469
+ "language_model.model.layers.14.linear_attn.in_proj_b": {
1470
+ "bits": 8,
1471
+ "group_size": 64
1472
+ },
1473
+ "language_model.model.layers.14.linear_attn.in_proj_a": {
1474
+ "bits": 8,
1475
+ "group_size": 64
1476
+ },
1477
+ "language_model.model.layers.14.linear_attn.out_proj": {
1478
+ "bits": 4,
1479
+ "group_size": 64
1480
+ },
1481
+ "language_model.model.layers.14.mlp.gate_proj": {
1482
+ "bits": 4,
1483
+ "group_size": 64
1484
+ },
1485
+ "language_model.model.layers.14.mlp.down_proj": {
1486
+ "bits": 4,
1487
+ "group_size": 64
1488
+ },
1489
+ "language_model.model.layers.14.mlp.up_proj": {
1490
+ "bits": 4,
1491
+ "group_size": 64
1492
+ },
1493
+ "language_model.model.layers.15.self_attn.q_proj": {
1494
+ "bits": 4,
1495
+ "group_size": 64
1496
+ },
1497
+ "language_model.model.layers.15.self_attn.k_proj": {
1498
+ "bits": 4,
1499
+ "group_size": 64
1500
+ },
1501
+ "language_model.model.layers.15.self_attn.v_proj": {
1502
+ "bits": 8,
1503
+ "group_size": 64
1504
+ },
1505
+ "language_model.model.layers.15.self_attn.o_proj": {
1506
+ "bits": 8,
1507
+ "group_size": 64
1508
+ },
1509
+ "language_model.model.layers.15.mlp.gate_proj": {
1510
+ "bits": 4,
1511
+ "group_size": 64
1512
+ },
1513
+ "language_model.model.layers.15.mlp.down_proj": {
1514
+ "bits": 4,
1515
+ "group_size": 64
1516
+ },
1517
+ "language_model.model.layers.15.mlp.up_proj": {
1518
+ "bits": 4,
1519
+ "group_size": 64
1520
+ },
1521
+ "language_model.model.layers.16.linear_attn.in_proj_qkv": {
1522
+ "bits": 8,
1523
+ "group_size": 64
1524
+ },
1525
+ "language_model.model.layers.16.linear_attn.in_proj_z": {
1526
+ "bits": 8,
1527
+ "group_size": 64
1528
+ },
1529
+ "language_model.model.layers.16.linear_attn.in_proj_b": {
1530
+ "bits": 8,
1531
+ "group_size": 64
1532
+ },
1533
+ "language_model.model.layers.16.linear_attn.in_proj_a": {
1534
+ "bits": 8,
1535
+ "group_size": 64
1536
+ },
1537
+ "language_model.model.layers.16.linear_attn.out_proj": {
1538
+ "bits": 4,
1539
+ "group_size": 64
1540
+ },
1541
+ "language_model.model.layers.16.mlp.gate_proj": {
1542
+ "bits": 4,
1543
+ "group_size": 64
1544
+ },
1545
+ "language_model.model.layers.16.mlp.down_proj": {
1546
+ "bits": 4,
1547
+ "group_size": 64
1548
+ },
1549
+ "language_model.model.layers.16.mlp.up_proj": {
1550
+ "bits": 4,
1551
+ "group_size": 64
1552
+ },
1553
+ "language_model.model.layers.17.linear_attn.in_proj_qkv": {
1554
+ "bits": 8,
1555
+ "group_size": 64
1556
+ },
1557
+ "language_model.model.layers.17.linear_attn.in_proj_z": {
1558
+ "bits": 8,
1559
+ "group_size": 64
1560
+ },
1561
+ "language_model.model.layers.17.linear_attn.in_proj_b": {
1562
+ "bits": 4,
1563
+ "group_size": 64
1564
+ },
1565
+ "language_model.model.layers.17.linear_attn.in_proj_a": {
1566
+ "bits": 8,
1567
+ "group_size": 64
1568
+ },
1569
+ "language_model.model.layers.17.linear_attn.out_proj": {
1570
+ "bits": 4,
1571
+ "group_size": 64
1572
+ },
1573
+ "language_model.model.layers.17.mlp.gate_proj": {
1574
+ "bits": 4,
1575
+ "group_size": 64
1576
+ },
1577
+ "language_model.model.layers.17.mlp.down_proj": {
1578
+ "bits": 4,
1579
+ "group_size": 64
1580
+ },
1581
+ "language_model.model.layers.17.mlp.up_proj": {
1582
+ "bits": 8,
1583
+ "group_size": 64
1584
+ },
1585
+ "language_model.model.layers.18.linear_attn.in_proj_qkv": {
1586
+ "bits": 8,
1587
+ "group_size": 64
1588
+ },
1589
+ "language_model.model.layers.18.linear_attn.in_proj_z": {
1590
+ "bits": 4,
1591
+ "group_size": 64
1592
+ },
1593
+ "language_model.model.layers.18.linear_attn.in_proj_b": {
1594
+ "bits": 4,
1595
+ "group_size": 64
1596
+ },
1597
+ "language_model.model.layers.18.linear_attn.in_proj_a": {
1598
+ "bits": 8,
1599
+ "group_size": 64
1600
+ },
1601
+ "language_model.model.layers.18.linear_attn.out_proj": {
1602
+ "bits": 8,
1603
+ "group_size": 64
1604
+ },
1605
+ "language_model.model.layers.18.mlp.gate_proj": {
1606
+ "bits": 4,
1607
+ "group_size": 64
1608
+ },
1609
+ "language_model.model.layers.18.mlp.down_proj": {
1610
+ "bits": 4,
1611
+ "group_size": 64
1612
+ },
1613
+ "language_model.model.layers.18.mlp.up_proj": {
1614
+ "bits": 4,
1615
+ "group_size": 64
1616
+ },
1617
+ "language_model.model.layers.19.self_attn.q_proj": {
1618
+ "bits": 4,
1619
+ "group_size": 64
1620
+ },
1621
+ "language_model.model.layers.19.self_attn.k_proj": {
1622
+ "bits": 4,
1623
+ "group_size": 64
1624
+ },
1625
+ "language_model.model.layers.19.self_attn.v_proj": {
1626
+ "bits": 8,
1627
+ "group_size": 64
1628
+ },
1629
+ "language_model.model.layers.19.self_attn.o_proj": {
1630
+ "bits": 8,
1631
+ "group_size": 64
1632
+ },
1633
+ "language_model.model.layers.19.mlp.gate_proj": {
1634
+ "bits": 4,
1635
+ "group_size": 64
1636
+ },
1637
+ "language_model.model.layers.19.mlp.down_proj": {
1638
+ "bits": 4,
1639
+ "group_size": 64
1640
+ },
1641
+ "language_model.model.layers.19.mlp.up_proj": {
1642
+ "bits": 4,
1643
+ "group_size": 64
1644
+ },
1645
+ "language_model.model.layers.20.linear_attn.in_proj_qkv": {
1646
+ "bits": 8,
1647
+ "group_size": 64
1648
+ },
1649
+ "language_model.model.layers.20.linear_attn.in_proj_z": {
1650
+ "bits": 4,
1651
+ "group_size": 64
1652
+ },
1653
+ "language_model.model.layers.20.linear_attn.in_proj_b": {
1654
+ "bits": 4,
1655
+ "group_size": 64
1656
+ },
1657
+ "language_model.model.layers.20.linear_attn.in_proj_a": {
1658
+ "bits": 4,
1659
+ "group_size": 64
1660
+ },
1661
+ "language_model.model.layers.20.linear_attn.out_proj": {
1662
+ "bits": 8,
1663
+ "group_size": 64
1664
+ },
1665
+ "language_model.model.layers.20.mlp.gate_proj": {
1666
+ "bits": 4,
1667
+ "group_size": 64
1668
+ },
1669
+ "language_model.model.layers.20.mlp.down_proj": {
1670
+ "bits": 8,
1671
+ "group_size": 64
1672
+ },
1673
+ "language_model.model.layers.20.mlp.up_proj": {
1674
+ "bits": 4,
1675
+ "group_size": 64
1676
+ },
1677
+ "language_model.model.layers.21.linear_attn.in_proj_qkv": {
1678
+ "bits": 8,
1679
+ "group_size": 64
1680
+ },
1681
+ "language_model.model.layers.21.linear_attn.in_proj_z": {
1682
+ "bits": 4,
1683
+ "group_size": 64
1684
+ },
1685
+ "language_model.model.layers.21.linear_attn.in_proj_b": {
1686
+ "bits": 4,
1687
+ "group_size": 64
1688
+ },
1689
+ "language_model.model.layers.21.linear_attn.in_proj_a": {
1690
+ "bits": 4,
1691
+ "group_size": 64
1692
+ },
1693
+ "language_model.model.layers.21.linear_attn.out_proj": {
1694
+ "bits": 8,
1695
+ "group_size": 64
1696
+ },
1697
+ "language_model.model.layers.21.mlp.gate_proj": {
1698
+ "bits": 4,
1699
+ "group_size": 64
1700
+ },
1701
+ "language_model.model.layers.21.mlp.down_proj": {
1702
+ "bits": 4,
1703
+ "group_size": 64
1704
+ },
1705
+ "language_model.model.layers.21.mlp.up_proj": {
1706
+ "bits": 4,
1707
+ "group_size": 64
1708
+ },
1709
+ "language_model.model.layers.22.linear_attn.in_proj_qkv": {
1710
+ "bits": 8,
1711
+ "group_size": 64
1712
+ },
1713
+ "language_model.model.layers.22.linear_attn.in_proj_z": {
1714
+ "bits": 4,
1715
+ "group_size": 64
1716
+ },
1717
+ "language_model.model.layers.22.linear_attn.in_proj_b": {
1718
+ "bits": 8,
1719
+ "group_size": 64
1720
+ },
1721
+ "language_model.model.layers.22.linear_attn.in_proj_a": {
1722
+ "bits": 4,
1723
+ "group_size": 64
1724
+ },
1725
+ "language_model.model.layers.22.linear_attn.out_proj": {
1726
+ "bits": 8,
1727
+ "group_size": 64
1728
+ },
1729
+ "language_model.model.layers.22.mlp.gate_proj": {
1730
+ "bits": 4,
1731
+ "group_size": 64
1732
+ },
1733
+ "language_model.model.layers.22.mlp.down_proj": {
1734
+ "bits": 4,
1735
+ "group_size": 64
1736
+ },
1737
+ "language_model.model.layers.22.mlp.up_proj": {
1738
+ "bits": 4,
1739
+ "group_size": 64
1740
+ },
1741
+ "language_model.model.layers.23.self_attn.q_proj": {
1742
+ "bits": 4,
1743
+ "group_size": 64
1744
+ },
1745
+ "language_model.model.layers.23.self_attn.k_proj": {
1746
+ "bits": 4,
1747
+ "group_size": 64
1748
+ },
1749
+ "language_model.model.layers.23.self_attn.v_proj": {
1750
+ "bits": 8,
1751
+ "group_size": 64
1752
+ },
1753
+ "language_model.model.layers.23.self_attn.o_proj": {
1754
+ "bits": 8,
1755
+ "group_size": 64
1756
+ },
1757
+ "language_model.model.layers.23.mlp.gate_proj": {
1758
+ "bits": 4,
1759
+ "group_size": 64
1760
+ },
1761
+ "language_model.model.layers.23.mlp.down_proj": {
1762
+ "bits": 8,
1763
+ "group_size": 64
1764
+ },
1765
+ "language_model.model.layers.23.mlp.up_proj": {
1766
+ "bits": 4,
1767
+ "group_size": 64
1768
+ },
1769
+ "language_model.model.layers.24.linear_attn.in_proj_qkv": {
1770
+ "bits": 8,
1771
+ "group_size": 64
1772
+ },
1773
+ "language_model.model.layers.24.linear_attn.in_proj_z": {
1774
+ "bits": 4,
1775
+ "group_size": 64
1776
+ },
1777
+ "language_model.model.layers.24.linear_attn.in_proj_b": {
1778
+ "bits": 4,
1779
+ "group_size": 64
1780
+ },
1781
+ "language_model.model.layers.24.linear_attn.in_proj_a": {
1782
+ "bits": 4,
1783
+ "group_size": 64
1784
+ },
1785
+ "language_model.model.layers.24.linear_attn.out_proj": {
1786
+ "bits": 8,
1787
+ "group_size": 64
1788
+ },
1789
+ "language_model.model.layers.24.mlp.gate_proj": {
1790
+ "bits": 4,
1791
+ "group_size": 64
1792
+ },
1793
+ "language_model.model.layers.24.mlp.down_proj": {
1794
+ "bits": 4,
1795
+ "group_size": 64
1796
+ },
1797
+ "language_model.model.layers.24.mlp.up_proj": {
1798
+ "bits": 4,
1799
+ "group_size": 64
1800
+ },
1801
+ "language_model.model.layers.25.linear_attn.in_proj_qkv": {
1802
+ "bits": 8,
1803
+ "group_size": 64
1804
+ },
1805
+ "language_model.model.layers.25.linear_attn.in_proj_z": {
1806
+ "bits": 4,
1807
+ "group_size": 64
1808
+ },
1809
+ "language_model.model.layers.25.linear_attn.in_proj_b": {
1810
+ "bits": 4,
1811
+ "group_size": 64
1812
+ },
1813
+ "language_model.model.layers.25.linear_attn.in_proj_a": {
1814
+ "bits": 4,
1815
+ "group_size": 64
1816
+ },
1817
+ "language_model.model.layers.25.linear_attn.out_proj": {
1818
+ "bits": 8,
1819
+ "group_size": 64
1820
+ },
1821
+ "language_model.model.layers.25.mlp.gate_proj": {
1822
+ "bits": 4,
1823
+ "group_size": 64
1824
+ },
1825
+ "language_model.model.layers.25.mlp.down_proj": {
1826
+ "bits": 4,
1827
+ "group_size": 64
1828
+ },
1829
+ "language_model.model.layers.25.mlp.up_proj": {
1830
+ "bits": 4,
1831
+ "group_size": 64
1832
+ },
1833
+ "language_model.model.layers.26.linear_attn.in_proj_qkv": {
1834
+ "bits": 8,
1835
+ "group_size": 64
1836
+ },
1837
+ "language_model.model.layers.26.linear_attn.in_proj_z": {
1838
+ "bits": 8,
1839
+ "group_size": 64
1840
+ },
1841
+ "language_model.model.layers.26.linear_attn.in_proj_b": {
1842
+ "bits": 4,
1843
+ "group_size": 64
1844
+ },
1845
+ "language_model.model.layers.26.linear_attn.in_proj_a": {
1846
+ "bits": 4,
1847
+ "group_size": 64
1848
+ },
1849
+ "language_model.model.layers.26.linear_attn.out_proj": {
1850
+ "bits": 4,
1851
+ "group_size": 64
1852
+ },
1853
+ "language_model.model.layers.26.mlp.gate_proj": {
1854
+ "bits": 4,
1855
+ "group_size": 64
1856
+ },
1857
+ "language_model.model.layers.26.mlp.down_proj": {
1858
+ "bits": 8,
1859
+ "group_size": 64
1860
+ },
1861
+ "language_model.model.layers.26.mlp.up_proj": {
1862
+ "bits": 4,
1863
+ "group_size": 64
1864
+ },
1865
+ "language_model.model.layers.27.self_attn.q_proj": {
1866
+ "bits": 4,
1867
+ "group_size": 64
1868
+ },
1869
+ "language_model.model.layers.27.self_attn.k_proj": {
1870
+ "bits": 4,
1871
+ "group_size": 64
1872
+ },
1873
+ "language_model.model.layers.27.self_attn.v_proj": {
1874
+ "bits": 8,
1875
+ "group_size": 64
1876
+ },
1877
+ "language_model.model.layers.27.self_attn.o_proj": {
1878
+ "bits": 8,
1879
+ "group_size": 64
1880
+ },
1881
+ "language_model.model.layers.27.mlp.gate_proj": {
1882
+ "bits": 4,
1883
+ "group_size": 64
1884
+ },
1885
+ "language_model.model.layers.27.mlp.down_proj": {
1886
+ "bits": 4,
1887
+ "group_size": 64
1888
+ },
1889
+ "language_model.model.layers.27.mlp.up_proj": {
1890
+ "bits": 4,
1891
+ "group_size": 64
1892
+ },
1893
+ "language_model.model.layers.28.linear_attn.in_proj_qkv": {
1894
+ "bits": 8,
1895
+ "group_size": 64
1896
+ },
1897
+ "language_model.model.layers.28.linear_attn.in_proj_z": {
1898
+ "bits": 4,
1899
+ "group_size": 64
1900
+ },
1901
+ "language_model.model.layers.28.linear_attn.in_proj_b": {
1902
+ "bits": 4,
1903
+ "group_size": 64
1904
+ },
1905
+ "language_model.model.layers.28.linear_attn.in_proj_a": {
1906
+ "bits": 4,
1907
+ "group_size": 64
1908
+ },
1909
+ "language_model.model.layers.28.linear_attn.out_proj": {
1910
+ "bits": 8,
1911
+ "group_size": 64
1912
+ },
1913
+ "language_model.model.layers.28.mlp.gate_proj": {
1914
+ "bits": 4,
1915
+ "group_size": 64
1916
+ },
1917
+ "language_model.model.layers.28.mlp.down_proj": {
1918
+ "bits": 4,
1919
+ "group_size": 64
1920
+ },
1921
+ "language_model.model.layers.28.mlp.up_proj": {
1922
+ "bits": 4,
1923
+ "group_size": 64
1924
+ },
1925
+ "language_model.model.layers.29.linear_attn.in_proj_qkv": {
1926
+ "bits": 8,
1927
+ "group_size": 64
1928
+ },
1929
+ "language_model.model.layers.29.linear_attn.in_proj_z": {
1930
+ "bits": 8,
1931
+ "group_size": 64
1932
+ },
1933
+ "language_model.model.layers.29.linear_attn.in_proj_b": {
1934
+ "bits": 4,
1935
+ "group_size": 64
1936
+ },
1937
+ "language_model.model.layers.29.linear_attn.in_proj_a": {
1938
+ "bits": 4,
1939
+ "group_size": 64
1940
+ },
1941
+ "language_model.model.layers.29.linear_attn.out_proj": {
1942
+ "bits": 4,
1943
+ "group_size": 64
1944
+ },
1945
+ "language_model.model.layers.29.mlp.gate_proj": {
1946
+ "bits": 4,
1947
+ "group_size": 64
1948
+ },
1949
+ "language_model.model.layers.29.mlp.down_proj": {
1950
+ "bits": 4,
1951
+ "group_size": 64
1952
+ },
1953
+ "language_model.model.layers.29.mlp.up_proj": {
1954
+ "bits": 8,
1955
+ "group_size": 64
1956
+ },
1957
+ "language_model.model.layers.30.linear_attn.in_proj_qkv": {
1958
+ "bits": 8,
1959
+ "group_size": 64
1960
+ },
1961
+ "language_model.model.layers.30.linear_attn.in_proj_z": {
1962
+ "bits": 4,
1963
+ "group_size": 64
1964
+ },
1965
+ "language_model.model.layers.30.linear_attn.in_proj_b": {
1966
+ "bits": 4,
1967
+ "group_size": 64
1968
+ },
1969
+ "language_model.model.layers.30.linear_attn.in_proj_a": {
1970
+ "bits": 4,
1971
+ "group_size": 64
1972
+ },
1973
+ "language_model.model.layers.30.linear_attn.out_proj": {
1974
+ "bits": 8,
1975
+ "group_size": 64
1976
+ },
1977
+ "language_model.model.layers.30.mlp.gate_proj": {
1978
+ "bits": 4,
1979
+ "group_size": 64
1980
+ },
1981
+ "language_model.model.layers.30.mlp.down_proj": {
1982
+ "bits": 4,
1983
+ "group_size": 64
1984
+ },
1985
+ "language_model.model.layers.30.mlp.up_proj": {
1986
+ "bits": 4,
1987
+ "group_size": 64
1988
+ },
1989
+ "language_model.model.layers.31.self_attn.q_proj": {
1990
+ "bits": 8,
1991
+ "group_size": 64
1992
+ },
1993
+ "language_model.model.layers.31.self_attn.k_proj": {
1994
+ "bits": 8,
1995
+ "group_size": 64
1996
+ },
1997
+ "language_model.model.layers.31.self_attn.v_proj": {
1998
+ "bits": 8,
1999
+ "group_size": 64
2000
+ },
2001
+ "language_model.model.layers.31.self_attn.o_proj": {
2002
+ "bits": 8,
2003
+ "group_size": 64
2004
+ },
2005
+ "language_model.model.layers.31.mlp.gate_proj": {
2006
+ "bits": 8,
2007
+ "group_size": 64
2008
+ },
2009
+ "language_model.model.layers.31.mlp.down_proj": {
2010
+ "bits": 8,
2011
+ "group_size": 64
2012
+ },
2013
+ "language_model.model.layers.31.mlp.up_proj": {
2014
+ "bits": 8,
2015
+ "group_size": 64
2016
+ },
2017
+ "language_model.lm_head": {
2018
+ "bits": 8,
2019
+ "group_size": 64
2020
+ }
2021
+ },
2022
+ "text_config": {
2023
+ "attention_bias": false,
2024
+ "attention_dropout": 0.0,
2025
+ "attn_output_gate": true,
2026
+ "bos_token_id": null,
2027
+ "dtype": "bfloat16",
2028
+ "eos_token_id": 248044,
2029
+ "full_attention_interval": 4,
2030
+ "head_dim": 256,
2031
+ "hidden_act": "silu",
2032
+ "hidden_size": 4096,
2033
+ "initializer_range": 0.02,
2034
+ "intermediate_size": 12288,
2035
+ "layer_types": [
2036
+ "linear_attention",
2037
+ "linear_attention",
2038
+ "linear_attention",
2039
+ "full_attention",
2040
+ "linear_attention",
2041
+ "linear_attention",
2042
+ "linear_attention",
2043
+ "full_attention",
2044
+ "linear_attention",
2045
+ "linear_attention",
2046
+ "linear_attention",
2047
+ "full_attention",
2048
+ "linear_attention",
2049
+ "linear_attention",
2050
+ "linear_attention",
2051
+ "full_attention",
2052
+ "linear_attention",
2053
+ "linear_attention",
2054
+ "linear_attention",
2055
+ "full_attention",
2056
+ "linear_attention",
2057
+ "linear_attention",
2058
+ "linear_attention",
2059
+ "full_attention",
2060
+ "linear_attention",
2061
+ "linear_attention",
2062
+ "linear_attention",
2063
+ "full_attention",
2064
+ "linear_attention",
2065
+ "linear_attention",
2066
+ "linear_attention",
2067
+ "full_attention"
2068
+ ],
2069
+ "linear_conv_kernel_dim": 4,
2070
+ "linear_key_head_dim": 128,
2071
+ "linear_num_key_heads": 16,
2072
+ "linear_num_value_heads": 32,
2073
+ "linear_value_head_dim": 128,
2074
+ "mamba_ssm_dtype": "float32",
2075
+ "max_position_embeddings": 262144,
2076
+ "mlp_only_layers": [],
2077
+ "model_type": "qwen3_5_text",
2078
+ "mtp_num_hidden_layers": 1,
2079
+ "mtp_use_dedicated_embeddings": false,
2080
+ "num_attention_heads": 16,
2081
+ "num_hidden_layers": 32,
2082
+ "num_key_value_heads": 4,
2083
+ "pad_token_id": null,
2084
+ "partial_rotary_factor": 0.25,
2085
+ "rms_norm_eps": 1e-06,
2086
+ "rope_parameters": {
2087
+ "mrope_interleaved": true,
2088
+ "mrope_section": [
2089
+ 11,
2090
+ 11,
2091
+ 10
2092
+ ],
2093
+ "partial_rotary_factor": 0.25,
2094
+ "rope_theta": 10000000,
2095
+ "type": "default"
2096
+ },
2097
+ "tie_word_embeddings": false,
2098
+ "use_cache": true,
2099
+ "vocab_size": 248320
2100
+ },
2101
+ "tie_word_embeddings": false,
2102
+ "transformers_version": "5.15.0.dev0",
2103
+ "video_token_id": 248057,
2104
+ "vision_end_token_id": 248054,
2105
+ "vision_start_token_id": 248053,
2106
+ "vision_config": {
2107
+ "deepstack_visual_indexes": [],
2108
+ "depth": 27,
2109
+ "dtype": "bfloat16",
2110
+ "hidden_act": "gelu_pytorch_tanh",
2111
+ "hidden_size": 1152,
2112
+ "in_channels": 3,
2113
+ "initializer_range": 0.02,
2114
+ "intermediate_size": 4304,
2115
+ "model_type": "qwen3_5_vision",
2116
+ "num_heads": 16,
2117
+ "num_position_embeddings": 2304,
2118
+ "out_hidden_size": 4096,
2119
+ "patch_size": 16,
2120
+ "spatial_merge_size": 2,
2121
+ "temporal_patch_size": 2
2122
+ },
2123
+ "optiq_vision": {
2124
+ "sidecar": "optiq/optiq_vision.safetensors",
2125
+ "dtype": "bfloat16",
2126
+ "n_tensors": 333,
2127
+ "base_model": "Altworld/Astrea-R8-Chat-9B"
2128
+ }
2129
+ }
generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "temperature": 0.2,
4
+ "top_p": 0.95,
5
+ "eos_token_id": [
6
+ 248044,
7
+ 248046
8
+ ],
9
+ "pad_token_id": 248044,
10
+ "transformers_version": "5.0.0.dev0"
11
+ }
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ea852239f6a5252d2e9886eba3d29336759c2ca51ef1cd6b2bc0537b7b2e2cc
3
+ size 5368685672
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee0eb88ec821760c6fbde2c9a21230e440dda06431fb6b34845c157059d27fd7
3
+ size 1684641870
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
optiq/optiq_vision.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a83be48e5d7355988bccc56eace9373954680f9dd95362c04aeac55f374b45c
3
+ size 912058029
optiq_metadata.json ADDED
@@ -0,0 +1,1008 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "method": "optiq_mixed_precision",
3
+ "base_model": "Altworld/Astrea-R8-Chat-9B",
4
+ "reference": "bf16",
5
+ "target_bpw": 5.0,
6
+ "achieved_bpw": 5.518829281183932,
7
+ "n_high_bits": 109,
8
+ "n_low_bits": 140,
9
+ "threshold": 0.0,
10
+ "per_layer": {
11
+ "language_model.lm_head": {
12
+ "bits": 8,
13
+ "group_size": 64
14
+ },
15
+ "language_model.model.layers.31.mlp.up_proj": {
16
+ "bits": 8,
17
+ "group_size": 64
18
+ },
19
+ "language_model.model.layers.31.mlp.down_proj": {
20
+ "bits": 8,
21
+ "group_size": 64
22
+ },
23
+ "language_model.model.layers.31.mlp.gate_proj": {
24
+ "bits": 8,
25
+ "group_size": 64
26
+ },
27
+ "language_model.model.layers.31.self_attn.o_proj": {
28
+ "bits": 8,
29
+ "group_size": 64
30
+ },
31
+ "language_model.model.layers.31.self_attn.v_proj": {
32
+ "bits": 8,
33
+ "group_size": 64
34
+ },
35
+ "language_model.model.layers.31.self_attn.k_proj": {
36
+ "bits": 8,
37
+ "group_size": 64
38
+ },
39
+ "language_model.model.layers.31.self_attn.q_proj": {
40
+ "bits": 8,
41
+ "group_size": 64
42
+ },
43
+ "language_model.model.layers.30.mlp.up_proj": {
44
+ "bits": 4,
45
+ "group_size": 64
46
+ },
47
+ "language_model.model.layers.30.mlp.down_proj": {
48
+ "bits": 4,
49
+ "group_size": 64
50
+ },
51
+ "language_model.model.layers.30.mlp.gate_proj": {
52
+ "bits": 4,
53
+ "group_size": 64
54
+ },
55
+ "language_model.model.layers.30.linear_attn.out_proj": {
56
+ "bits": 8,
57
+ "group_size": 64
58
+ },
59
+ "language_model.model.layers.30.linear_attn.in_proj_a": {
60
+ "bits": 4,
61
+ "group_size": 64
62
+ },
63
+ "language_model.model.layers.30.linear_attn.in_proj_b": {
64
+ "bits": 4,
65
+ "group_size": 64
66
+ },
67
+ "language_model.model.layers.30.linear_attn.in_proj_z": {
68
+ "bits": 4,
69
+ "group_size": 64
70
+ },
71
+ "language_model.model.layers.30.linear_attn.in_proj_qkv": {
72
+ "bits": 8,
73
+ "group_size": 64
74
+ },
75
+ "language_model.model.layers.29.mlp.up_proj": {
76
+ "bits": 8,
77
+ "group_size": 64
78
+ },
79
+ "language_model.model.layers.29.mlp.down_proj": {
80
+ "bits": 4,
81
+ "group_size": 64
82
+ },
83
+ "language_model.model.layers.29.mlp.gate_proj": {
84
+ "bits": 4,
85
+ "group_size": 64
86
+ },
87
+ "language_model.model.layers.29.linear_attn.out_proj": {
88
+ "bits": 4,
89
+ "group_size": 64
90
+ },
91
+ "language_model.model.layers.29.linear_attn.in_proj_a": {
92
+ "bits": 4,
93
+ "group_size": 64
94
+ },
95
+ "language_model.model.layers.29.linear_attn.in_proj_b": {
96
+ "bits": 4,
97
+ "group_size": 64
98
+ },
99
+ "language_model.model.layers.29.linear_attn.in_proj_z": {
100
+ "bits": 8,
101
+ "group_size": 64
102
+ },
103
+ "language_model.model.layers.29.linear_attn.in_proj_qkv": {
104
+ "bits": 8,
105
+ "group_size": 64
106
+ },
107
+ "language_model.model.layers.28.mlp.up_proj": {
108
+ "bits": 4,
109
+ "group_size": 64
110
+ },
111
+ "language_model.model.layers.28.mlp.down_proj": {
112
+ "bits": 4,
113
+ "group_size": 64
114
+ },
115
+ "language_model.model.layers.28.mlp.gate_proj": {
116
+ "bits": 4,
117
+ "group_size": 64
118
+ },
119
+ "language_model.model.layers.28.linear_attn.out_proj": {
120
+ "bits": 8,
121
+ "group_size": 64
122
+ },
123
+ "language_model.model.layers.28.linear_attn.in_proj_a": {
124
+ "bits": 4,
125
+ "group_size": 64
126
+ },
127
+ "language_model.model.layers.28.linear_attn.in_proj_b": {
128
+ "bits": 4,
129
+ "group_size": 64
130
+ },
131
+ "language_model.model.layers.28.linear_attn.in_proj_z": {
132
+ "bits": 4,
133
+ "group_size": 64
134
+ },
135
+ "language_model.model.layers.28.linear_attn.in_proj_qkv": {
136
+ "bits": 8,
137
+ "group_size": 64
138
+ },
139
+ "language_model.model.layers.27.mlp.up_proj": {
140
+ "bits": 4,
141
+ "group_size": 64
142
+ },
143
+ "language_model.model.layers.27.mlp.down_proj": {
144
+ "bits": 4,
145
+ "group_size": 64
146
+ },
147
+ "language_model.model.layers.27.mlp.gate_proj": {
148
+ "bits": 4,
149
+ "group_size": 64
150
+ },
151
+ "language_model.model.layers.27.self_attn.o_proj": {
152
+ "bits": 8,
153
+ "group_size": 64
154
+ },
155
+ "language_model.model.layers.27.self_attn.v_proj": {
156
+ "bits": 8,
157
+ "group_size": 64
158
+ },
159
+ "language_model.model.layers.27.self_attn.k_proj": {
160
+ "bits": 4,
161
+ "group_size": 64
162
+ },
163
+ "language_model.model.layers.27.self_attn.q_proj": {
164
+ "bits": 4,
165
+ "group_size": 64
166
+ },
167
+ "language_model.model.layers.26.mlp.up_proj": {
168
+ "bits": 4,
169
+ "group_size": 64
170
+ },
171
+ "language_model.model.layers.26.mlp.down_proj": {
172
+ "bits": 8,
173
+ "group_size": 64
174
+ },
175
+ "language_model.model.layers.26.mlp.gate_proj": {
176
+ "bits": 4,
177
+ "group_size": 64
178
+ },
179
+ "language_model.model.layers.26.linear_attn.out_proj": {
180
+ "bits": 4,
181
+ "group_size": 64
182
+ },
183
+ "language_model.model.layers.26.linear_attn.in_proj_a": {
184
+ "bits": 4,
185
+ "group_size": 64
186
+ },
187
+ "language_model.model.layers.26.linear_attn.in_proj_b": {
188
+ "bits": 4,
189
+ "group_size": 64
190
+ },
191
+ "language_model.model.layers.26.linear_attn.in_proj_z": {
192
+ "bits": 8,
193
+ "group_size": 64
194
+ },
195
+ "language_model.model.layers.26.linear_attn.in_proj_qkv": {
196
+ "bits": 8,
197
+ "group_size": 64
198
+ },
199
+ "language_model.model.layers.25.mlp.up_proj": {
200
+ "bits": 4,
201
+ "group_size": 64
202
+ },
203
+ "language_model.model.layers.25.mlp.down_proj": {
204
+ "bits": 4,
205
+ "group_size": 64
206
+ },
207
+ "language_model.model.layers.25.mlp.gate_proj": {
208
+ "bits": 4,
209
+ "group_size": 64
210
+ },
211
+ "language_model.model.layers.25.linear_attn.out_proj": {
212
+ "bits": 8,
213
+ "group_size": 64
214
+ },
215
+ "language_model.model.layers.25.linear_attn.in_proj_a": {
216
+ "bits": 4,
217
+ "group_size": 64
218
+ },
219
+ "language_model.model.layers.25.linear_attn.in_proj_b": {
220
+ "bits": 4,
221
+ "group_size": 64
222
+ },
223
+ "language_model.model.layers.25.linear_attn.in_proj_z": {
224
+ "bits": 4,
225
+ "group_size": 64
226
+ },
227
+ "language_model.model.layers.25.linear_attn.in_proj_qkv": {
228
+ "bits": 8,
229
+ "group_size": 64
230
+ },
231
+ "language_model.model.layers.24.mlp.up_proj": {
232
+ "bits": 4,
233
+ "group_size": 64
234
+ },
235
+ "language_model.model.layers.24.mlp.down_proj": {
236
+ "bits": 4,
237
+ "group_size": 64
238
+ },
239
+ "language_model.model.layers.24.mlp.gate_proj": {
240
+ "bits": 4,
241
+ "group_size": 64
242
+ },
243
+ "language_model.model.layers.24.linear_attn.out_proj": {
244
+ "bits": 8,
245
+ "group_size": 64
246
+ },
247
+ "language_model.model.layers.24.linear_attn.in_proj_a": {
248
+ "bits": 4,
249
+ "group_size": 64
250
+ },
251
+ "language_model.model.layers.24.linear_attn.in_proj_b": {
252
+ "bits": 4,
253
+ "group_size": 64
254
+ },
255
+ "language_model.model.layers.24.linear_attn.in_proj_z": {
256
+ "bits": 4,
257
+ "group_size": 64
258
+ },
259
+ "language_model.model.layers.24.linear_attn.in_proj_qkv": {
260
+ "bits": 8,
261
+ "group_size": 64
262
+ },
263
+ "language_model.model.layers.23.mlp.up_proj": {
264
+ "bits": 4,
265
+ "group_size": 64
266
+ },
267
+ "language_model.model.layers.23.mlp.down_proj": {
268
+ "bits": 8,
269
+ "group_size": 64
270
+ },
271
+ "language_model.model.layers.23.mlp.gate_proj": {
272
+ "bits": 4,
273
+ "group_size": 64
274
+ },
275
+ "language_model.model.layers.23.self_attn.o_proj": {
276
+ "bits": 8,
277
+ "group_size": 64
278
+ },
279
+ "language_model.model.layers.23.self_attn.v_proj": {
280
+ "bits": 8,
281
+ "group_size": 64
282
+ },
283
+ "language_model.model.layers.23.self_attn.k_proj": {
284
+ "bits": 4,
285
+ "group_size": 64
286
+ },
287
+ "language_model.model.layers.23.self_attn.q_proj": {
288
+ "bits": 4,
289
+ "group_size": 64
290
+ },
291
+ "language_model.model.layers.22.mlp.up_proj": {
292
+ "bits": 4,
293
+ "group_size": 64
294
+ },
295
+ "language_model.model.layers.22.mlp.down_proj": {
296
+ "bits": 4,
297
+ "group_size": 64
298
+ },
299
+ "language_model.model.layers.22.mlp.gate_proj": {
300
+ "bits": 4,
301
+ "group_size": 64
302
+ },
303
+ "language_model.model.layers.22.linear_attn.out_proj": {
304
+ "bits": 8,
305
+ "group_size": 64
306
+ },
307
+ "language_model.model.layers.22.linear_attn.in_proj_a": {
308
+ "bits": 4,
309
+ "group_size": 64
310
+ },
311
+ "language_model.model.layers.22.linear_attn.in_proj_b": {
312
+ "bits": 8,
313
+ "group_size": 64
314
+ },
315
+ "language_model.model.layers.22.linear_attn.in_proj_z": {
316
+ "bits": 4,
317
+ "group_size": 64
318
+ },
319
+ "language_model.model.layers.22.linear_attn.in_proj_qkv": {
320
+ "bits": 8,
321
+ "group_size": 64
322
+ },
323
+ "language_model.model.layers.21.mlp.up_proj": {
324
+ "bits": 4,
325
+ "group_size": 64
326
+ },
327
+ "language_model.model.layers.21.mlp.down_proj": {
328
+ "bits": 4,
329
+ "group_size": 64
330
+ },
331
+ "language_model.model.layers.21.mlp.gate_proj": {
332
+ "bits": 4,
333
+ "group_size": 64
334
+ },
335
+ "language_model.model.layers.21.linear_attn.out_proj": {
336
+ "bits": 8,
337
+ "group_size": 64
338
+ },
339
+ "language_model.model.layers.21.linear_attn.in_proj_a": {
340
+ "bits": 4,
341
+ "group_size": 64
342
+ },
343
+ "language_model.model.layers.21.linear_attn.in_proj_b": {
344
+ "bits": 4,
345
+ "group_size": 64
346
+ },
347
+ "language_model.model.layers.21.linear_attn.in_proj_z": {
348
+ "bits": 4,
349
+ "group_size": 64
350
+ },
351
+ "language_model.model.layers.21.linear_attn.in_proj_qkv": {
352
+ "bits": 8,
353
+ "group_size": 64
354
+ },
355
+ "language_model.model.layers.20.mlp.up_proj": {
356
+ "bits": 4,
357
+ "group_size": 64
358
+ },
359
+ "language_model.model.layers.20.mlp.down_proj": {
360
+ "bits": 8,
361
+ "group_size": 64
362
+ },
363
+ "language_model.model.layers.20.mlp.gate_proj": {
364
+ "bits": 4,
365
+ "group_size": 64
366
+ },
367
+ "language_model.model.layers.20.linear_attn.out_proj": {
368
+ "bits": 8,
369
+ "group_size": 64
370
+ },
371
+ "language_model.model.layers.20.linear_attn.in_proj_a": {
372
+ "bits": 4,
373
+ "group_size": 64
374
+ },
375
+ "language_model.model.layers.20.linear_attn.in_proj_b": {
376
+ "bits": 4,
377
+ "group_size": 64
378
+ },
379
+ "language_model.model.layers.20.linear_attn.in_proj_z": {
380
+ "bits": 4,
381
+ "group_size": 64
382
+ },
383
+ "language_model.model.layers.20.linear_attn.in_proj_qkv": {
384
+ "bits": 8,
385
+ "group_size": 64
386
+ },
387
+ "language_model.model.layers.19.mlp.up_proj": {
388
+ "bits": 4,
389
+ "group_size": 64
390
+ },
391
+ "language_model.model.layers.19.mlp.down_proj": {
392
+ "bits": 4,
393
+ "group_size": 64
394
+ },
395
+ "language_model.model.layers.19.mlp.gate_proj": {
396
+ "bits": 4,
397
+ "group_size": 64
398
+ },
399
+ "language_model.model.layers.19.self_attn.o_proj": {
400
+ "bits": 8,
401
+ "group_size": 64
402
+ },
403
+ "language_model.model.layers.19.self_attn.v_proj": {
404
+ "bits": 8,
405
+ "group_size": 64
406
+ },
407
+ "language_model.model.layers.19.self_attn.k_proj": {
408
+ "bits": 4,
409
+ "group_size": 64
410
+ },
411
+ "language_model.model.layers.19.self_attn.q_proj": {
412
+ "bits": 4,
413
+ "group_size": 64
414
+ },
415
+ "language_model.model.layers.18.mlp.up_proj": {
416
+ "bits": 4,
417
+ "group_size": 64
418
+ },
419
+ "language_model.model.layers.18.mlp.down_proj": {
420
+ "bits": 4,
421
+ "group_size": 64
422
+ },
423
+ "language_model.model.layers.18.mlp.gate_proj": {
424
+ "bits": 4,
425
+ "group_size": 64
426
+ },
427
+ "language_model.model.layers.18.linear_attn.out_proj": {
428
+ "bits": 8,
429
+ "group_size": 64
430
+ },
431
+ "language_model.model.layers.18.linear_attn.in_proj_a": {
432
+ "bits": 8,
433
+ "group_size": 64
434
+ },
435
+ "language_model.model.layers.18.linear_attn.in_proj_b": {
436
+ "bits": 4,
437
+ "group_size": 64
438
+ },
439
+ "language_model.model.layers.18.linear_attn.in_proj_z": {
440
+ "bits": 4,
441
+ "group_size": 64
442
+ },
443
+ "language_model.model.layers.18.linear_attn.in_proj_qkv": {
444
+ "bits": 8,
445
+ "group_size": 64
446
+ },
447
+ "language_model.model.layers.17.mlp.up_proj": {
448
+ "bits": 8,
449
+ "group_size": 64
450
+ },
451
+ "language_model.model.layers.17.mlp.down_proj": {
452
+ "bits": 4,
453
+ "group_size": 64
454
+ },
455
+ "language_model.model.layers.17.mlp.gate_proj": {
456
+ "bits": 4,
457
+ "group_size": 64
458
+ },
459
+ "language_model.model.layers.17.linear_attn.out_proj": {
460
+ "bits": 4,
461
+ "group_size": 64
462
+ },
463
+ "language_model.model.layers.17.linear_attn.in_proj_a": {
464
+ "bits": 8,
465
+ "group_size": 64
466
+ },
467
+ "language_model.model.layers.17.linear_attn.in_proj_b": {
468
+ "bits": 4,
469
+ "group_size": 64
470
+ },
471
+ "language_model.model.layers.17.linear_attn.in_proj_z": {
472
+ "bits": 8,
473
+ "group_size": 64
474
+ },
475
+ "language_model.model.layers.17.linear_attn.in_proj_qkv": {
476
+ "bits": 8,
477
+ "group_size": 64
478
+ },
479
+ "language_model.model.layers.16.mlp.up_proj": {
480
+ "bits": 4,
481
+ "group_size": 64
482
+ },
483
+ "language_model.model.layers.16.mlp.down_proj": {
484
+ "bits": 4,
485
+ "group_size": 64
486
+ },
487
+ "language_model.model.layers.16.mlp.gate_proj": {
488
+ "bits": 4,
489
+ "group_size": 64
490
+ },
491
+ "language_model.model.layers.16.linear_attn.out_proj": {
492
+ "bits": 4,
493
+ "group_size": 64
494
+ },
495
+ "language_model.model.layers.16.linear_attn.in_proj_a": {
496
+ "bits": 8,
497
+ "group_size": 64
498
+ },
499
+ "language_model.model.layers.16.linear_attn.in_proj_b": {
500
+ "bits": 8,
501
+ "group_size": 64
502
+ },
503
+ "language_model.model.layers.16.linear_attn.in_proj_z": {
504
+ "bits": 8,
505
+ "group_size": 64
506
+ },
507
+ "language_model.model.layers.16.linear_attn.in_proj_qkv": {
508
+ "bits": 8,
509
+ "group_size": 64
510
+ },
511
+ "language_model.model.layers.15.mlp.up_proj": {
512
+ "bits": 4,
513
+ "group_size": 64
514
+ },
515
+ "language_model.model.layers.15.mlp.down_proj": {
516
+ "bits": 4,
517
+ "group_size": 64
518
+ },
519
+ "language_model.model.layers.15.mlp.gate_proj": {
520
+ "bits": 4,
521
+ "group_size": 64
522
+ },
523
+ "language_model.model.layers.15.self_attn.o_proj": {
524
+ "bits": 8,
525
+ "group_size": 64
526
+ },
527
+ "language_model.model.layers.15.self_attn.v_proj": {
528
+ "bits": 8,
529
+ "group_size": 64
530
+ },
531
+ "language_model.model.layers.15.self_attn.k_proj": {
532
+ "bits": 4,
533
+ "group_size": 64
534
+ },
535
+ "language_model.model.layers.15.self_attn.q_proj": {
536
+ "bits": 4,
537
+ "group_size": 64
538
+ },
539
+ "language_model.model.layers.14.mlp.up_proj": {
540
+ "bits": 4,
541
+ "group_size": 64
542
+ },
543
+ "language_model.model.layers.14.mlp.down_proj": {
544
+ "bits": 4,
545
+ "group_size": 64
546
+ },
547
+ "language_model.model.layers.14.mlp.gate_proj": {
548
+ "bits": 4,
549
+ "group_size": 64
550
+ },
551
+ "language_model.model.layers.14.linear_attn.out_proj": {
552
+ "bits": 4,
553
+ "group_size": 64
554
+ },
555
+ "language_model.model.layers.14.linear_attn.in_proj_a": {
556
+ "bits": 8,
557
+ "group_size": 64
558
+ },
559
+ "language_model.model.layers.14.linear_attn.in_proj_b": {
560
+ "bits": 8,
561
+ "group_size": 64
562
+ },
563
+ "language_model.model.layers.14.linear_attn.in_proj_z": {
564
+ "bits": 8,
565
+ "group_size": 64
566
+ },
567
+ "language_model.model.layers.14.linear_attn.in_proj_qkv": {
568
+ "bits": 8,
569
+ "group_size": 64
570
+ },
571
+ "language_model.model.layers.13.mlp.up_proj": {
572
+ "bits": 4,
573
+ "group_size": 64
574
+ },
575
+ "language_model.model.layers.13.mlp.down_proj": {
576
+ "bits": 4,
577
+ "group_size": 64
578
+ },
579
+ "language_model.model.layers.13.mlp.gate_proj": {
580
+ "bits": 4,
581
+ "group_size": 64
582
+ },
583
+ "language_model.model.layers.13.linear_attn.out_proj": {
584
+ "bits": 4,
585
+ "group_size": 64
586
+ },
587
+ "language_model.model.layers.13.linear_attn.in_proj_a": {
588
+ "bits": 8,
589
+ "group_size": 64
590
+ },
591
+ "language_model.model.layers.13.linear_attn.in_proj_b": {
592
+ "bits": 8,
593
+ "group_size": 64
594
+ },
595
+ "language_model.model.layers.13.linear_attn.in_proj_z": {
596
+ "bits": 8,
597
+ "group_size": 64
598
+ },
599
+ "language_model.model.layers.13.linear_attn.in_proj_qkv": {
600
+ "bits": 8,
601
+ "group_size": 64
602
+ },
603
+ "language_model.model.layers.12.mlp.up_proj": {
604
+ "bits": 4,
605
+ "group_size": 64
606
+ },
607
+ "language_model.model.layers.12.mlp.down_proj": {
608
+ "bits": 4,
609
+ "group_size": 64
610
+ },
611
+ "language_model.model.layers.12.mlp.gate_proj": {
612
+ "bits": 4,
613
+ "group_size": 64
614
+ },
615
+ "language_model.model.layers.12.linear_attn.out_proj": {
616
+ "bits": 4,
617
+ "group_size": 64
618
+ },
619
+ "language_model.model.layers.12.linear_attn.in_proj_a": {
620
+ "bits": 8,
621
+ "group_size": 64
622
+ },
623
+ "language_model.model.layers.12.linear_attn.in_proj_b": {
624
+ "bits": 8,
625
+ "group_size": 64
626
+ },
627
+ "language_model.model.layers.12.linear_attn.in_proj_z": {
628
+ "bits": 8,
629
+ "group_size": 64
630
+ },
631
+ "language_model.model.layers.12.linear_attn.in_proj_qkv": {
632
+ "bits": 8,
633
+ "group_size": 64
634
+ },
635
+ "language_model.model.layers.11.mlp.up_proj": {
636
+ "bits": 4,
637
+ "group_size": 64
638
+ },
639
+ "language_model.model.layers.11.mlp.down_proj": {
640
+ "bits": 4,
641
+ "group_size": 64
642
+ },
643
+ "language_model.model.layers.11.mlp.gate_proj": {
644
+ "bits": 4,
645
+ "group_size": 64
646
+ },
647
+ "language_model.model.layers.11.self_attn.o_proj": {
648
+ "bits": 8,
649
+ "group_size": 64
650
+ },
651
+ "language_model.model.layers.11.self_attn.v_proj": {
652
+ "bits": 8,
653
+ "group_size": 64
654
+ },
655
+ "language_model.model.layers.11.self_attn.k_proj": {
656
+ "bits": 4,
657
+ "group_size": 64
658
+ },
659
+ "language_model.model.layers.11.self_attn.q_proj": {
660
+ "bits": 4,
661
+ "group_size": 64
662
+ },
663
+ "language_model.model.layers.10.mlp.up_proj": {
664
+ "bits": 4,
665
+ "group_size": 64
666
+ },
667
+ "language_model.model.layers.10.mlp.down_proj": {
668
+ "bits": 4,
669
+ "group_size": 64
670
+ },
671
+ "language_model.model.layers.10.mlp.gate_proj": {
672
+ "bits": 4,
673
+ "group_size": 64
674
+ },
675
+ "language_model.model.layers.10.linear_attn.out_proj": {
676
+ "bits": 4,
677
+ "group_size": 64
678
+ },
679
+ "language_model.model.layers.10.linear_attn.in_proj_a": {
680
+ "bits": 8,
681
+ "group_size": 64
682
+ },
683
+ "language_model.model.layers.10.linear_attn.in_proj_b": {
684
+ "bits": 8,
685
+ "group_size": 64
686
+ },
687
+ "language_model.model.layers.10.linear_attn.in_proj_z": {
688
+ "bits": 8,
689
+ "group_size": 64
690
+ },
691
+ "language_model.model.layers.10.linear_attn.in_proj_qkv": {
692
+ "bits": 8,
693
+ "group_size": 64
694
+ },
695
+ "language_model.model.layers.9.mlp.up_proj": {
696
+ "bits": 4,
697
+ "group_size": 64
698
+ },
699
+ "language_model.model.layers.9.mlp.down_proj": {
700
+ "bits": 4,
701
+ "group_size": 64
702
+ },
703
+ "language_model.model.layers.9.mlp.gate_proj": {
704
+ "bits": 4,
705
+ "group_size": 64
706
+ },
707
+ "language_model.model.layers.9.linear_attn.out_proj": {
708
+ "bits": 8,
709
+ "group_size": 64
710
+ },
711
+ "language_model.model.layers.9.linear_attn.in_proj_a": {
712
+ "bits": 8,
713
+ "group_size": 64
714
+ },
715
+ "language_model.model.layers.9.linear_attn.in_proj_b": {
716
+ "bits": 8,
717
+ "group_size": 64
718
+ },
719
+ "language_model.model.layers.9.linear_attn.in_proj_z": {
720
+ "bits": 4,
721
+ "group_size": 64
722
+ },
723
+ "language_model.model.layers.9.linear_attn.in_proj_qkv": {
724
+ "bits": 8,
725
+ "group_size": 64
726
+ },
727
+ "language_model.model.layers.8.mlp.up_proj": {
728
+ "bits": 4,
729
+ "group_size": 64
730
+ },
731
+ "language_model.model.layers.8.mlp.down_proj": {
732
+ "bits": 4,
733
+ "group_size": 64
734
+ },
735
+ "language_model.model.layers.8.mlp.gate_proj": {
736
+ "bits": 4,
737
+ "group_size": 64
738
+ },
739
+ "language_model.model.layers.8.linear_attn.out_proj": {
740
+ "bits": 4,
741
+ "group_size": 64
742
+ },
743
+ "language_model.model.layers.8.linear_attn.in_proj_a": {
744
+ "bits": 8,
745
+ "group_size": 64
746
+ },
747
+ "language_model.model.layers.8.linear_attn.in_proj_b": {
748
+ "bits": 8,
749
+ "group_size": 64
750
+ },
751
+ "language_model.model.layers.8.linear_attn.in_proj_z": {
752
+ "bits": 8,
753
+ "group_size": 64
754
+ },
755
+ "language_model.model.layers.8.linear_attn.in_proj_qkv": {
756
+ "bits": 8,
757
+ "group_size": 64
758
+ },
759
+ "language_model.model.layers.7.mlp.up_proj": {
760
+ "bits": 4,
761
+ "group_size": 64
762
+ },
763
+ "language_model.model.layers.7.mlp.down_proj": {
764
+ "bits": 4,
765
+ "group_size": 64
766
+ },
767
+ "language_model.model.layers.7.mlp.gate_proj": {
768
+ "bits": 4,
769
+ "group_size": 64
770
+ },
771
+ "language_model.model.layers.7.self_attn.o_proj": {
772
+ "bits": 4,
773
+ "group_size": 64
774
+ },
775
+ "language_model.model.layers.7.self_attn.v_proj": {
776
+ "bits": 8,
777
+ "group_size": 64
778
+ },
779
+ "language_model.model.layers.7.self_attn.k_proj": {
780
+ "bits": 8,
781
+ "group_size": 64
782
+ },
783
+ "language_model.model.layers.7.self_attn.q_proj": {
784
+ "bits": 8,
785
+ "group_size": 64
786
+ },
787
+ "language_model.model.layers.6.mlp.up_proj": {
788
+ "bits": 4,
789
+ "group_size": 64
790
+ },
791
+ "language_model.model.layers.6.mlp.down_proj": {
792
+ "bits": 4,
793
+ "group_size": 64
794
+ },
795
+ "language_model.model.layers.6.mlp.gate_proj": {
796
+ "bits": 4,
797
+ "group_size": 64
798
+ },
799
+ "language_model.model.layers.6.linear_attn.out_proj": {
800
+ "bits": 4,
801
+ "group_size": 64
802
+ },
803
+ "language_model.model.layers.6.linear_attn.in_proj_a": {
804
+ "bits": 8,
805
+ "group_size": 64
806
+ },
807
+ "language_model.model.layers.6.linear_attn.in_proj_b": {
808
+ "bits": 8,
809
+ "group_size": 64
810
+ },
811
+ "language_model.model.layers.6.linear_attn.in_proj_z": {
812
+ "bits": 8,
813
+ "group_size": 64
814
+ },
815
+ "language_model.model.layers.6.linear_attn.in_proj_qkv": {
816
+ "bits": 8,
817
+ "group_size": 64
818
+ },
819
+ "language_model.model.layers.5.mlp.up_proj": {
820
+ "bits": 4,
821
+ "group_size": 64
822
+ },
823
+ "language_model.model.layers.5.mlp.down_proj": {
824
+ "bits": 4,
825
+ "group_size": 64
826
+ },
827
+ "language_model.model.layers.5.mlp.gate_proj": {
828
+ "bits": 4,
829
+ "group_size": 64
830
+ },
831
+ "language_model.model.layers.5.linear_attn.out_proj": {
832
+ "bits": 8,
833
+ "group_size": 64
834
+ },
835
+ "language_model.model.layers.5.linear_attn.in_proj_a": {
836
+ "bits": 8,
837
+ "group_size": 64
838
+ },
839
+ "language_model.model.layers.5.linear_attn.in_proj_b": {
840
+ "bits": 8,
841
+ "group_size": 64
842
+ },
843
+ "language_model.model.layers.5.linear_attn.in_proj_z": {
844
+ "bits": 4,
845
+ "group_size": 64
846
+ },
847
+ "language_model.model.layers.5.linear_attn.in_proj_qkv": {
848
+ "bits": 8,
849
+ "group_size": 64
850
+ },
851
+ "language_model.model.layers.4.mlp.up_proj": {
852
+ "bits": 4,
853
+ "group_size": 64
854
+ },
855
+ "language_model.model.layers.4.mlp.down_proj": {
856
+ "bits": 4,
857
+ "group_size": 64
858
+ },
859
+ "language_model.model.layers.4.mlp.gate_proj": {
860
+ "bits": 4,
861
+ "group_size": 64
862
+ },
863
+ "language_model.model.layers.4.linear_attn.out_proj": {
864
+ "bits": 4,
865
+ "group_size": 64
866
+ },
867
+ "language_model.model.layers.4.linear_attn.in_proj_a": {
868
+ "bits": 8,
869
+ "group_size": 64
870
+ },
871
+ "language_model.model.layers.4.linear_attn.in_proj_b": {
872
+ "bits": 8,
873
+ "group_size": 64
874
+ },
875
+ "language_model.model.layers.4.linear_attn.in_proj_z": {
876
+ "bits": 8,
877
+ "group_size": 64
878
+ },
879
+ "language_model.model.layers.4.linear_attn.in_proj_qkv": {
880
+ "bits": 8,
881
+ "group_size": 64
882
+ },
883
+ "language_model.model.layers.3.mlp.up_proj": {
884
+ "bits": 4,
885
+ "group_size": 64
886
+ },
887
+ "language_model.model.layers.3.mlp.down_proj": {
888
+ "bits": 4,
889
+ "group_size": 64
890
+ },
891
+ "language_model.model.layers.3.mlp.gate_proj": {
892
+ "bits": 4,
893
+ "group_size": 64
894
+ },
895
+ "language_model.model.layers.3.self_attn.o_proj": {
896
+ "bits": 8,
897
+ "group_size": 64
898
+ },
899
+ "language_model.model.layers.3.self_attn.v_proj": {
900
+ "bits": 8,
901
+ "group_size": 64
902
+ },
903
+ "language_model.model.layers.3.self_attn.k_proj": {
904
+ "bits": 8,
905
+ "group_size": 64
906
+ },
907
+ "language_model.model.layers.3.self_attn.q_proj": {
908
+ "bits": 4,
909
+ "group_size": 64
910
+ },
911
+ "language_model.model.layers.2.mlp.up_proj": {
912
+ "bits": 4,
913
+ "group_size": 64
914
+ },
915
+ "language_model.model.layers.2.mlp.down_proj": {
916
+ "bits": 4,
917
+ "group_size": 64
918
+ },
919
+ "language_model.model.layers.2.mlp.gate_proj": {
920
+ "bits": 4,
921
+ "group_size": 64
922
+ },
923
+ "language_model.model.layers.2.linear_attn.out_proj": {
924
+ "bits": 4,
925
+ "group_size": 64
926
+ },
927
+ "language_model.model.layers.2.linear_attn.in_proj_a": {
928
+ "bits": 8,
929
+ "group_size": 64
930
+ },
931
+ "language_model.model.layers.2.linear_attn.in_proj_b": {
932
+ "bits": 8,
933
+ "group_size": 64
934
+ },
935
+ "language_model.model.layers.2.linear_attn.in_proj_z": {
936
+ "bits": 4,
937
+ "group_size": 64
938
+ },
939
+ "language_model.model.layers.2.linear_attn.in_proj_qkv": {
940
+ "bits": 8,
941
+ "group_size": 64
942
+ },
943
+ "language_model.model.layers.1.mlp.up_proj": {
944
+ "bits": 4,
945
+ "group_size": 64
946
+ },
947
+ "language_model.model.layers.1.mlp.down_proj": {
948
+ "bits": 4,
949
+ "group_size": 64
950
+ },
951
+ "language_model.model.layers.1.mlp.gate_proj": {
952
+ "bits": 4,
953
+ "group_size": 64
954
+ },
955
+ "language_model.model.layers.1.linear_attn.out_proj": {
956
+ "bits": 4,
957
+ "group_size": 64
958
+ },
959
+ "language_model.model.layers.1.linear_attn.in_proj_a": {
960
+ "bits": 8,
961
+ "group_size": 64
962
+ },
963
+ "language_model.model.layers.1.linear_attn.in_proj_b": {
964
+ "bits": 8,
965
+ "group_size": 64
966
+ },
967
+ "language_model.model.layers.1.linear_attn.in_proj_z": {
968
+ "bits": 8,
969
+ "group_size": 64
970
+ },
971
+ "language_model.model.layers.1.linear_attn.in_proj_qkv": {
972
+ "bits": 8,
973
+ "group_size": 64
974
+ },
975
+ "language_model.model.layers.0.mlp.up_proj": {
976
+ "bits": 8,
977
+ "group_size": 64
978
+ },
979
+ "language_model.model.layers.0.mlp.down_proj": {
980
+ "bits": 8,
981
+ "group_size": 64
982
+ },
983
+ "language_model.model.layers.0.mlp.gate_proj": {
984
+ "bits": 8,
985
+ "group_size": 64
986
+ },
987
+ "language_model.model.layers.0.linear_attn.out_proj": {
988
+ "bits": 8,
989
+ "group_size": 64
990
+ },
991
+ "language_model.model.layers.0.linear_attn.in_proj_a": {
992
+ "bits": 8,
993
+ "group_size": 64
994
+ },
995
+ "language_model.model.layers.0.linear_attn.in_proj_b": {
996
+ "bits": 8,
997
+ "group_size": 64
998
+ },
999
+ "language_model.model.layers.0.linear_attn.in_proj_z": {
1000
+ "bits": 8,
1001
+ "group_size": 64
1002
+ },
1003
+ "language_model.model.layers.0.linear_attn.in_proj_qkv": {
1004
+ "bits": 8,
1005
+ "group_size": 64
1006
+ }
1007
+ }
1008
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
3
+ size 19989325
tokenizer_config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "local_files_only": false,
14
+ "model_max_length": 262144,
15
+ "model_specific_special_tokens": {
16
+ "audio_bos_token": "<|audio_start|>",
17
+ "audio_eos_token": "<|audio_end|>",
18
+ "audio_token": "<|audio_pad|>",
19
+ "image_token": "<|image_pad|>",
20
+ "video_token": "<|video_pad|>",
21
+ "vision_bos_token": "<|vision_start|>",
22
+ "vision_eos_token": "<|vision_end|>"
23
+ },
24
+ "pad_token": "<|endoftext|>",
25
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "split_special_tokens": false,
27
+ "tokenizer_class": "Qwen2Tokenizer",
28
+ "tool_parser_type": "qwen3_coder",
29
+ "unk_token": null,
30
+ "video_token": "<|video_pad|>",
31
+ "vision_bos_token": "<|vision_start|>",
32
+ "vision_eos_token": "<|vision_end|>"
33
+ }