Hugging Face's logo

nightmedia
/
Qwen3-Coder-REAP-25B-A3B-qx64-hi-mlx

Text Generation
MLX
Safetensors
English
qwen3_moe
qwen-coder
MOE
pruning
compression
conversational
6-bit
Model card Files
xet
Community
nightmedia commited on
Commit
6819572
·
verified ·
1 Parent(s): 25682c6

Add files using upload-large-folder tool

Browse files
Files changed (17) hide show
  1. .gitattributes +1 -0
  2. README.md +51 -0
  3. added_tokens.json +28 -0
  4. chat_template.jinja +117 -0
  5. config.json +3138 -0
  6. generation_config.json +13 -0
  7. merges.txt +0 -0
  8. model-00001-of-00004.safetensors +3 -0
  9. model-00002-of-00004.safetensors +3 -0
  10. model-00003-of-00004.safetensors +3 -0
  11. model-00004-of-00004.safetensors +3 -0
  12. model.safetensors.index.json +0 -0
  13. qwen3coder_tool_parser.py +689 -0
  14. special_tokens_map.json +31 -0
  15. tokenizer.json +3 -0
  16. tokenizer_config.json +239 -0
  17. vocab.json +0 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ library_name: mlx
5
+ tags:
6
+ - qwen-coder
7
+ - MOE
8
+ - pruning
9
+ - compression
10
+ - mlx
11
+ license: apache-2.0
12
+ name: cerebras/Qwen3-Coder-REAP-25B-A3B
13
+ description: 'This model was obtained by uniformly pruning 20% of experts in Qwen3-Coder-30B-A3B-Instruct
14
+ using the REAP method.
15
+
16
+ '
17
+ readme: 'https://huggingface.co/cerebras/Qwen3-Coder-REAP-25B-A3B/main/README.md
18
+
19
+ '
20
+ license_link: https://huggingface.co/cerebras/Qwen3-Coder-REAP-25B-A3B/blob/main/LICENSE
21
+ pipeline_tag: text-generation
22
+ base_model: cerebras/Qwen3-Coder-REAP-25B-A3B
23
+ ---
24
+
25
+ # Qwen3-Coder-REAP-25B-A3B-qx64-hi-mlx
26
+
27
+ This model [Qwen3-Coder-REAP-25B-A3B-qx64-hi-mlx](https://huggingface.co/Qwen3-Coder-REAP-25B-A3B-qx64-hi-mlx) was
28
+ converted to MLX format from [cerebras/Qwen3-Coder-REAP-25B-A3B](https://huggingface.co/cerebras/Qwen3-Coder-REAP-25B-A3B)
29
+ using mlx-lm version **0.28.3**.
30
+
31
+ ## Use with mlx
32
+
33
+ ```bash
34
+ pip install mlx-lm
35
+ ```
36
+
37
+ ```python
38
+ from mlx_lm import load, generate
39
+
40
+ model, tokenizer = load("Qwen3-Coder-REAP-25B-A3B-qx64-hi-mlx")
41
+
42
+ prompt = "hello"
43
+
44
+ if tokenizer.chat_template is not None:
45
+ messages = [{"role": "user", "content": prompt}]
46
+ prompt = tokenizer.apply_chat_template(
47
+ messages, add_generation_prompt=True
48
+ )
49
+
50
+ response = generate(model, tokenizer, prompt=prompt, verbose=True)
51
+ ```
added_tokens.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</think>": 151668,
3
+ "</tool_call>": 151658,
4
+ "</tool_response>": 151666,
5
+ "<think>": 151667,
6
+ "<tool_call>": 151657,
7
+ "<tool_response>": 151665,
8
+ "<|box_end|>": 151649,
9
+ "<|box_start|>": 151648,
10
+ "<|endoftext|>": 151643,
11
+ "<|file_sep|>": 151664,
12
+ "<|fim_middle|>": 151660,
13
+ "<|fim_pad|>": 151662,
14
+ "<|fim_prefix|>": 151659,
15
+ "<|fim_suffix|>": 151661,
16
+ "<|im_end|>": 151645,
17
+ "<|im_start|>": 151644,
18
+ "<|image_pad|>": 151655,
19
+ "<|object_ref_end|>": 151647,
20
+ "<|object_ref_start|>": 151646,
21
+ "<|quad_end|>": 151651,
22
+ "<|quad_start|>": 151650,
23
+ "<|repo_name|>": 151663,
24
+ "<|video_pad|>": 151656,
25
+ "<|vision_end|>": 151653,
26
+ "<|vision_pad|>": 151654,
27
+ "<|vision_start|>": 151652
28
+ }
chat_template.jinja ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {% macro render_extra_keys(json_dict, handled_keys) %}
2
+ {%- if json_dict is mapping %}
3
+ {%- for json_key in json_dict if json_key not in handled_keys %}
4
+ {%- if json_dict[json_key] is mapping or (json_dict[json_key] is sequence and json_dict[json_key] is not string) %}
5
+ {{- '\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ json_key ~ '>' }}
6
+ {%- else %}
7
+ {{-'\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' }}
8
+ {%- endif %}
9
+ {%- endfor %}
10
+ {%- endif %}
11
+ {% endmacro %}
12
+
13
+ {%- if messages[0]["role"] == "system" %}
14
+ {%- set system_message = messages[0]["content"] %}
15
+ {%- set loop_messages = messages[1:] %}
16
+ {%- else %}
17
+ {%- set loop_messages = messages %}
18
+ {%- endif %}
19
+
20
+ {%- if not tools is defined %}
21
+ {%- set tools = [] %}
22
+ {%- endif %}
23
+
24
+ {%- if system_message is defined %}
25
+ {{- "<|im_start|>system\n" + system_message }}
26
+ {%- else %}
27
+ {%- if tools is iterable and tools | length > 0 %}
28
+ {{- "<|im_start|>system\nYou are Qwen, a helpful AI assistant that can interact with a computer to solve tasks." }}
29
+ {%- endif %}
30
+ {%- endif %}
31
+ {%- if tools is iterable and tools | length > 0 %}
32
+ {{- "\n\n# Tools\n\nYou have access to the following functions:\n\n" }}
33
+ {{- "<tools>" }}
34
+ {%- for tool in tools %}
35
+ {%- if tool.function is defined %}
36
+ {%- set tool = tool.function %}
37
+ {%- endif %}
38
+ {{- "\n<function>\n<name>" ~ tool.name ~ "</name>" }}
39
+ {%- if tool.description is defined %}
40
+ {{- '\n<description>' ~ (tool.description | trim) ~ '</description>' }}
41
+ {%- endif %}
42
+ {{- '\n<parameters>' }}
43
+ {%- if tool.parameters is defined and tool.parameters is mapping and tool.parameters.properties is defined and tool.parameters.properties is mapping %}
44
+ {%- for param_name, param_fields in tool.parameters.properties|items %}
45
+ {{- '\n<parameter>' }}
46
+ {{- '\n<name>' ~ param_name ~ '</name>' }}
47
+ {%- if param_fields.type is defined %}
48
+ {{- '\n<type>' ~ (param_fields.type | string) ~ '</type>' }}
49
+ {%- endif %}
50
+ {%- if param_fields.description is defined %}
51
+ {{- '\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}
52
+ {%- endif %}
53
+ {%- set handled_keys = ['name', 'type', 'description'] %}
54
+ {{- render_extra_keys(param_fields, handled_keys) }}
55
+ {{- '\n</parameter>' }}
56
+ {%- endfor %}
57
+ {%- endif %}
58
+ {% set handled_keys = ['type', 'properties'] %}
59
+ {{- render_extra_keys(tool.parameters, handled_keys) }}
60
+ {{- '\n</parameters>' }}
61
+ {%- set handled_keys = ['type', 'name', 'description', 'parameters'] %}
62
+ {{- render_extra_keys(tool, handled_keys) }}
63
+ {{- '\n</function>' }}
64
+ {%- endfor %}
65
+ {{- "\n</tools>" }}
66
+ {{- '\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>' }}
67
+ {%- endif %}
68
+ {%- if system_message is defined %}
69
+ {{- '<|im_end|>\n' }}
70
+ {%- else %}
71
+ {%- if tools is iterable and tools | length > 0 %}
72
+ {{- '<|im_end|>\n' }}
73
+ {%- endif %}
74
+ {%- endif %}
75
+ {%- for message in loop_messages %}
76
+ {%- if message.role == "assistant" and message.tool_calls is defined and message.tool_calls is iterable and message.tool_calls | length > 0 %}
77
+ {{- '<|im_start|>' + message.role }}
78
+ {%- if message.content is defined and message.content is string and message.content | trim | length > 0 %}
79
+ {{- '\n' + message.content | trim + '\n' }}
80
+ {%- endif %}
81
+ {%- for tool_call in message.tool_calls %}
82
+ {%- if tool_call.function is defined %}
83
+ {%- set tool_call = tool_call.function %}
84
+ {%- endif %}
85
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
86
+ {%- if tool_call.arguments is defined %}
87
+ {%- for args_name, args_value in tool_call.arguments|items %}
88
+ {{- '<parameter=' + args_name + '>\n' }}
89
+ {%- 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 %}
90
+ {{- args_value }}
91
+ {{- '\n</parameter>\n' }}
92
+ {%- endfor %}
93
+ {%- endif %}
94
+ {{- '</function>\n</tool_call>' }}
95
+ {%- endfor %}
96
+ {{- '<|im_end|>\n' }}
97
+ {%- elif message.role == "user" or message.role == "system" or message.role == "assistant" %}
98
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
99
+ {%- elif message.role == "tool" %}
100
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
101
+ {{- '<|im_start|>user\n' }}
102
+ {%- endif %}
103
+ {{- '<tool_response>\n' }}
104
+ {{- message.content }}
105
+ {{- '\n</tool_response>\n' }}
106
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
107
+ {{- '<|im_end|>\n' }}
108
+ {%- elif loop.last %}
109
+ {{- '<|im_end|>\n' }}
110
+ {%- endif %}
111
+ {%- else %}
112
+ {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>\n' }}
113
+ {%- endif %}
114
+ {%- endfor %}
115
+ {%- if add_generation_prompt %}
116
+ {{- '<|im_start|>assistant\n' }}
117
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,3138 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3MoeForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "decoder_sparse_step": 1,
8
+ "eos_token_id": 151645,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 5472,
14
+ "max_position_embeddings": 262144,
15
+ "max_window_layers": 28,
16
+ "mlp_only_layers": [],
17
+ "model_type": "qwen3_moe",
18
+ "moe_intermediate_size": 768,
19
+ "norm_topk_prob": true,
20
+ "num_attention_heads": 32,
21
+ "num_experts": 103,
22
+ "num_experts_per_tok": 8,
23
+ "num_hidden_layers": 48,
24
+ "num_key_value_heads": 4,
25
+ "output_router_logits": false,
26
+ "qkv_bias": false,
27
+ "quantization": {
28
+ "group_size": 32,
29
+ "bits": 6,
30
+ "mode": "affine",
31
+ "model.embed_tokens": {
32
+ "group_size": 32,
33
+ "bits": 4
34
+ },
35
+ "model.layers.0.self_attn.q_proj": {
36
+ "group_size": 32,
37
+ "bits": 6
38
+ },
39
+ "model.layers.0.self_attn.k_proj": {
40
+ "group_size": 32,
41
+ "bits": 6
42
+ },
43
+ "model.layers.0.self_attn.v_proj": {
44
+ "group_size": 32,
45
+ "bits": 6
46
+ },
47
+ "model.layers.0.self_attn.o_proj": {
48
+ "group_size": 32,
49
+ "bits": 6
50
+ },
51
+ "model.layers.0.mlp.gate": {
52
+ "group_size": 32,
53
+ "bits": 6
54
+ },
55
+ "model.layers.0.mlp.switch_mlp.gate_proj": {
56
+ "group_size": 32,
57
+ "bits": 6
58
+ },
59
+ "model.layers.0.mlp.switch_mlp.up_proj": {
60
+ "group_size": 32,
61
+ "bits": 6
62
+ },
63
+ "model.layers.0.mlp.switch_mlp.down_proj": {
64
+ "group_size": 32,
65
+ "bits": 6
66
+ },
67
+ "model.layers.1.self_attn.q_proj": {
68
+ "group_size": 32,
69
+ "bits": 4
70
+ },
71
+ "model.layers.1.self_attn.k_proj": {
72
+ "group_size": 32,
73
+ "bits": 4
74
+ },
75
+ "model.layers.1.self_attn.v_proj": {
76
+ "group_size": 32,
77
+ "bits": 6
78
+ },
79
+ "model.layers.1.self_attn.o_proj": {
80
+ "group_size": 32,
81
+ "bits": 4
82
+ },
83
+ "model.layers.1.mlp.gate": {
84
+ "group_size": 32,
85
+ "bits": 4
86
+ },
87
+ "model.layers.1.mlp.switch_mlp.gate_proj": {
88
+ "group_size": 32,
89
+ "bits": 4
90
+ },
91
+ "model.layers.1.mlp.switch_mlp.up_proj": {
92
+ "group_size": 32,
93
+ "bits": 4
94
+ },
95
+ "model.layers.1.mlp.switch_mlp.down_proj": {
96
+ "group_size": 32,
97
+ "bits": 6
98
+ },
99
+ "model.layers.2.self_attn.q_proj": {
100
+ "group_size": 32,
101
+ "bits": 4
102
+ },
103
+ "model.layers.2.self_attn.k_proj": {
104
+ "group_size": 32,
105
+ "bits": 4
106
+ },
107
+ "model.layers.2.self_attn.v_proj": {
108
+ "group_size": 32,
109
+ "bits": 6
110
+ },
111
+ "model.layers.2.self_attn.o_proj": {
112
+ "group_size": 32,
113
+ "bits": 4
114
+ },
115
+ "model.layers.2.mlp.gate": {
116
+ "group_size": 32,
117
+ "bits": 4
118
+ },
119
+ "model.layers.2.mlp.switch_mlp.gate_proj": {
120
+ "group_size": 32,
121
+ "bits": 4
122
+ },
123
+ "model.layers.2.mlp.switch_mlp.up_proj": {
124
+ "group_size": 32,
125
+ "bits": 4
126
+ },
127
+ "model.layers.2.mlp.switch_mlp.down_proj": {
128
+ "group_size": 32,
129
+ "bits": 6
130
+ },
131
+ "model.layers.3.self_attn.q_proj": {
132
+ "group_size": 32,
133
+ "bits": 4
134
+ },
135
+ "model.layers.3.self_attn.k_proj": {
136
+ "group_size": 32,
137
+ "bits": 4
138
+ },
139
+ "model.layers.3.self_attn.v_proj": {
140
+ "group_size": 32,
141
+ "bits": 6
142
+ },
143
+ "model.layers.3.self_attn.o_proj": {
144
+ "group_size": 32,
145
+ "bits": 4
146
+ },
147
+ "model.layers.3.mlp.gate": {
148
+ "group_size": 32,
149
+ "bits": 4
150
+ },
151
+ "model.layers.3.mlp.switch_mlp.gate_proj": {
152
+ "group_size": 32,
153
+ "bits": 4
154
+ },
155
+ "model.layers.3.mlp.switch_mlp.up_proj": {
156
+ "group_size": 32,
157
+ "bits": 4
158
+ },
159
+ "model.layers.3.mlp.switch_mlp.down_proj": {
160
+ "group_size": 32,
161
+ "bits": 6
162
+ },
163
+ "model.layers.4.self_attn.q_proj": {
164
+ "group_size": 32,
165
+ "bits": 4
166
+ },
167
+ "model.layers.4.self_attn.k_proj": {
168
+ "group_size": 32,
169
+ "bits": 4
170
+ },
171
+ "model.layers.4.self_attn.v_proj": {
172
+ "group_size": 32,
173
+ "bits": 6
174
+ },
175
+ "model.layers.4.self_attn.o_proj": {
176
+ "group_size": 32,
177
+ "bits": 4
178
+ },
179
+ "model.layers.4.mlp.gate": {
180
+ "group_size": 32,
181
+ "bits": 4
182
+ },
183
+ "model.layers.4.mlp.switch_mlp.gate_proj": {
184
+ "group_size": 32,
185
+ "bits": 4
186
+ },
187
+ "model.layers.4.mlp.switch_mlp.up_proj": {
188
+ "group_size": 32,
189
+ "bits": 4
190
+ },
191
+ "model.layers.4.mlp.switch_mlp.down_proj": {
192
+ "group_size": 32,
193
+ "bits": 6
194
+ },
195
+ "model.layers.5.self_attn.q_proj": {
196
+ "group_size": 32,
197
+ "bits": 4
198
+ },
199
+ "model.layers.5.self_attn.k_proj": {
200
+ "group_size": 32,
201
+ "bits": 4
202
+ },
203
+ "model.layers.5.self_attn.v_proj": {
204
+ "group_size": 32,
205
+ "bits": 6
206
+ },
207
+ "model.layers.5.self_attn.o_proj": {
208
+ "group_size": 32,
209
+ "bits": 4
210
+ },
211
+ "model.layers.5.mlp.gate": {
212
+ "group_size": 32,
213
+ "bits": 4
214
+ },
215
+ "model.layers.5.mlp.switch_mlp.gate_proj": {
216
+ "group_size": 32,
217
+ "bits": 4
218
+ },
219
+ "model.layers.5.mlp.switch_mlp.up_proj": {
220
+ "group_size": 32,
221
+ "bits": 4
222
+ },
223
+ "model.layers.5.mlp.switch_mlp.down_proj": {
224
+ "group_size": 32,
225
+ "bits": 6
226
+ },
227
+ "model.layers.6.self_attn.q_proj": {
228
+ "group_size": 32,
229
+ "bits": 4
230
+ },
231
+ "model.layers.6.self_attn.k_proj": {
232
+ "group_size": 32,
233
+ "bits": 4
234
+ },
235
+ "model.layers.6.self_attn.v_proj": {
236
+ "group_size": 32,
237
+ "bits": 6
238
+ },
239
+ "model.layers.6.self_attn.o_proj": {
240
+ "group_size": 32,
241
+ "bits": 4
242
+ },
243
+ "model.layers.6.mlp.gate": {
244
+ "group_size": 32,
245
+ "bits": 4
246
+ },
247
+ "model.layers.6.mlp.switch_mlp.gate_proj": {
248
+ "group_size": 32,
249
+ "bits": 4
250
+ },
251
+ "model.layers.6.mlp.switch_mlp.up_proj": {
252
+ "group_size": 32,
253
+ "bits": 4
254
+ },
255
+ "model.layers.6.mlp.switch_mlp.down_proj": {
256
+ "group_size": 32,
257
+ "bits": 4
258
+ },
259
+ "model.layers.7.self_attn.q_proj": {
260
+ "group_size": 32,
261
+ "bits": 4
262
+ },
263
+ "model.layers.7.self_attn.k_proj": {
264
+ "group_size": 32,
265
+ "bits": 4
266
+ },
267
+ "model.layers.7.self_attn.v_proj": {
268
+ "group_size": 32,
269
+ "bits": 6
270
+ },
271
+ "model.layers.7.self_attn.o_proj": {
272
+ "group_size": 32,
273
+ "bits": 4
274
+ },
275
+ "model.layers.7.mlp.gate": {
276
+ "group_size": 32,
277
+ "bits": 4
278
+ },
279
+ "model.layers.7.mlp.switch_mlp.gate_proj": {
280
+ "group_size": 32,
281
+ "bits": 4
282
+ },
283
+ "model.layers.7.mlp.switch_mlp.up_proj": {
284
+ "group_size": 32,
285
+ "bits": 4
286
+ },
287
+ "model.layers.7.mlp.switch_mlp.down_proj": {
288
+ "group_size": 32,
289
+ "bits": 4
290
+ },
291
+ "model.layers.8.self_attn.q_proj": {
292
+ "group_size": 32,
293
+ "bits": 4
294
+ },
295
+ "model.layers.8.self_attn.k_proj": {
296
+ "group_size": 32,
297
+ "bits": 4
298
+ },
299
+ "model.layers.8.self_attn.v_proj": {
300
+ "group_size": 32,
301
+ "bits": 6
302
+ },
303
+ "model.layers.8.self_attn.o_proj": {
304
+ "group_size": 32,
305
+ "bits": 4
306
+ },
307
+ "model.layers.8.mlp.gate": {
308
+ "group_size": 32,
309
+ "bits": 4
310
+ },
311
+ "model.layers.8.mlp.switch_mlp.gate_proj": {
312
+ "group_size": 32,
313
+ "bits": 4
314
+ },
315
+ "model.layers.8.mlp.switch_mlp.up_proj": {
316
+ "group_size": 32,
317
+ "bits": 4
318
+ },
319
+ "model.layers.8.mlp.switch_mlp.down_proj": {
320
+ "group_size": 32,
321
+ "bits": 6
322
+ },
323
+ "model.layers.9.self_attn.q_proj": {
324
+ "group_size": 32,
325
+ "bits": 4
326
+ },
327
+ "model.layers.9.self_attn.k_proj": {
328
+ "group_size": 32,
329
+ "bits": 4
330
+ },
331
+ "model.layers.9.self_attn.v_proj": {
332
+ "group_size": 32,
333
+ "bits": 6
334
+ },
335
+ "model.layers.9.self_attn.o_proj": {
336
+ "group_size": 32,
337
+ "bits": 4
338
+ },
339
+ "model.layers.9.mlp.gate": {
340
+ "group_size": 32,
341
+ "bits": 4
342
+ },
343
+ "model.layers.9.mlp.switch_mlp.gate_proj": {
344
+ "group_size": 32,
345
+ "bits": 4
346
+ },
347
+ "model.layers.9.mlp.switch_mlp.up_proj": {
348
+ "group_size": 32,
349
+ "bits": 4
350
+ },
351
+ "model.layers.9.mlp.switch_mlp.down_proj": {
352
+ "group_size": 32,
353
+ "bits": 4
354
+ },
355
+ "model.layers.10.self_attn.q_proj": {
356
+ "group_size": 32,
357
+ "bits": 4
358
+ },
359
+ "model.layers.10.self_attn.k_proj": {
360
+ "group_size": 32,
361
+ "bits": 4
362
+ },
363
+ "model.layers.10.self_attn.v_proj": {
364
+ "group_size": 32,
365
+ "bits": 6
366
+ },
367
+ "model.layers.10.self_attn.o_proj": {
368
+ "group_size": 32,
369
+ "bits": 4
370
+ },
371
+ "model.layers.10.mlp.gate": {
372
+ "group_size": 32,
373
+ "bits": 4
374
+ },
375
+ "model.layers.10.mlp.switch_mlp.gate_proj": {
376
+ "group_size": 32,
377
+ "bits": 4
378
+ },
379
+ "model.layers.10.mlp.switch_mlp.up_proj": {
380
+ "group_size": 32,
381
+ "bits": 4
382
+ },
383
+ "model.layers.10.mlp.switch_mlp.down_proj": {
384
+ "group_size": 32,
385
+ "bits": 4
386
+ },
387
+ "model.layers.11.self_attn.q_proj": {
388
+ "group_size": 32,
389
+ "bits": 4
390
+ },
391
+ "model.layers.11.self_attn.k_proj": {
392
+ "group_size": 32,
393
+ "bits": 4
394
+ },
395
+ "model.layers.11.self_attn.v_proj": {
396
+ "group_size": 32,
397
+ "bits": 6
398
+ },
399
+ "model.layers.11.self_attn.o_proj": {
400
+ "group_size": 32,
401
+ "bits": 4
402
+ },
403
+ "model.layers.11.mlp.gate": {
404
+ "group_size": 32,
405
+ "bits": 4
406
+ },
407
+ "model.layers.11.mlp.switch_mlp.gate_proj": {
408
+ "group_size": 32,
409
+ "bits": 4
410
+ },
411
+ "model.layers.11.mlp.switch_mlp.up_proj": {
412
+ "group_size": 32,
413
+ "bits": 4
414
+ },
415
+ "model.layers.11.mlp.switch_mlp.down_proj": {
416
+ "group_size": 32,
417
+ "bits": 6
418
+ },
419
+ "model.layers.12.self_attn.q_proj": {
420
+ "group_size": 32,
421
+ "bits": 4
422
+ },
423
+ "model.layers.12.self_attn.k_proj": {
424
+ "group_size": 32,
425
+ "bits": 4
426
+ },
427
+ "model.layers.12.self_attn.v_proj": {
428
+ "group_size": 32,
429
+ "bits": 6
430
+ },
431
+ "model.layers.12.self_attn.o_proj": {
432
+ "group_size": 32,
433
+ "bits": 4
434
+ },
435
+ "model.layers.12.mlp.gate": {
436
+ "group_size": 32,
437
+ "bits": 4
438
+ },
439
+ "model.layers.12.mlp.switch_mlp.gate_proj": {
440
+ "group_size": 32,
441
+ "bits": 4
442
+ },
443
+ "model.layers.12.mlp.switch_mlp.up_proj": {
444
+ "group_size": 32,
445
+ "bits": 4
446
+ },
447
+ "model.layers.12.mlp.switch_mlp.down_proj": {
448
+ "group_size": 32,
449
+ "bits": 4
450
+ },
451
+ "model.layers.13.self_attn.q_proj": {
452
+ "group_size": 32,
453
+ "bits": 4
454
+ },
455
+ "model.layers.13.self_attn.k_proj": {
456
+ "group_size": 32,
457
+ "bits": 4
458
+ },
459
+ "model.layers.13.self_attn.v_proj": {
460
+ "group_size": 32,
461
+ "bits": 6
462
+ },
463
+ "model.layers.13.self_attn.o_proj": {
464
+ "group_size": 32,
465
+ "bits": 4
466
+ },
467
+ "model.layers.13.mlp.gate": {
468
+ "group_size": 32,
469
+ "bits": 4
470
+ },
471
+ "model.layers.13.mlp.switch_mlp.gate_proj": {
472
+ "group_size": 32,
473
+ "bits": 4
474
+ },
475
+ "model.layers.13.mlp.switch_mlp.up_proj": {
476
+ "group_size": 32,
477
+ "bits": 4
478
+ },
479
+ "model.layers.13.mlp.switch_mlp.down_proj": {
480
+ "group_size": 32,
481
+ "bits": 4
482
+ },
483
+ "model.layers.14.self_attn.q_proj": {
484
+ "group_size": 32,
485
+ "bits": 4
486
+ },
487
+ "model.layers.14.self_attn.k_proj": {
488
+ "group_size": 32,
489
+ "bits": 4
490
+ },
491
+ "model.layers.14.self_attn.v_proj": {
492
+ "group_size": 32,
493
+ "bits": 6
494
+ },
495
+ "model.layers.14.self_attn.o_proj": {
496
+ "group_size": 32,
497
+ "bits": 4
498
+ },
499
+ "model.layers.14.mlp.gate": {
500
+ "group_size": 32,
501
+ "bits": 4
502
+ },
503
+ "model.layers.14.mlp.switch_mlp.gate_proj": {
504
+ "group_size": 32,
505
+ "bits": 4
506
+ },
507
+ "model.layers.14.mlp.switch_mlp.up_proj": {
508
+ "group_size": 32,
509
+ "bits": 4
510
+ },
511
+ "model.layers.14.mlp.switch_mlp.down_proj": {
512
+ "group_size": 32,
513
+ "bits": 6
514
+ },
515
+ "model.layers.15.self_attn.q_proj": {
516
+ "group_size": 32,
517
+ "bits": 4
518
+ },
519
+ "model.layers.15.self_attn.k_proj": {
520
+ "group_size": 32,
521
+ "bits": 4
522
+ },
523
+ "model.layers.15.self_attn.v_proj": {
524
+ "group_size": 32,
525
+ "bits": 6
526
+ },
527
+ "model.layers.15.self_attn.o_proj": {
528
+ "group_size": 32,
529
+ "bits": 4
530
+ },
531
+ "model.layers.15.mlp.gate": {
532
+ "group_size": 32,
533
+ "bits": 4
534
+ },
535
+ "model.layers.15.mlp.switch_mlp.gate_proj": {
536
+ "group_size": 32,
537
+ "bits": 4
538
+ },
539
+ "model.layers.15.mlp.switch_mlp.up_proj": {
540
+ "group_size": 32,
541
+ "bits": 4
542
+ },
543
+ "model.layers.15.mlp.switch_mlp.down_proj": {
544
+ "group_size": 32,
545
+ "bits": 4
546
+ },
547
+ "model.layers.16.self_attn.q_proj": {
548
+ "group_size": 32,
549
+ "bits": 4
550
+ },
551
+ "model.layers.16.self_attn.k_proj": {
552
+ "group_size": 32,
553
+ "bits": 4
554
+ },
555
+ "model.layers.16.self_attn.v_proj": {
556
+ "group_size": 32,
557
+ "bits": 6
558
+ },
559
+ "model.layers.16.self_attn.o_proj": {
560
+ "group_size": 32,
561
+ "bits": 4
562
+ },
563
+ "model.layers.16.mlp.gate": {
564
+ "group_size": 32,
565
+ "bits": 4
566
+ },
567
+ "model.layers.16.mlp.switch_mlp.gate_proj": {
568
+ "group_size": 32,
569
+ "bits": 4
570
+ },
571
+ "model.layers.16.mlp.switch_mlp.up_proj": {
572
+ "group_size": 32,
573
+ "bits": 4
574
+ },
575
+ "model.layers.16.mlp.switch_mlp.down_proj": {
576
+ "group_size": 32,
577
+ "bits": 4
578
+ },
579
+ "model.layers.17.self_attn.q_proj": {
580
+ "group_size": 32,
581
+ "bits": 4
582
+ },
583
+ "model.layers.17.self_attn.k_proj": {
584
+ "group_size": 32,
585
+ "bits": 4
586
+ },
587
+ "model.layers.17.self_attn.v_proj": {
588
+ "group_size": 32,
589
+ "bits": 6
590
+ },
591
+ "model.layers.17.self_attn.o_proj": {
592
+ "group_size": 32,
593
+ "bits": 4
594
+ },
595
+ "model.layers.17.mlp.gate": {
596
+ "group_size": 32,
597
+ "bits": 4
598
+ },
599
+ "model.layers.17.mlp.switch_mlp.gate_proj": {
600
+ "group_size": 32,
601
+ "bits": 4
602
+ },
603
+ "model.layers.17.mlp.switch_mlp.up_proj": {
604
+ "group_size": 32,
605
+ "bits": 4
606
+ },
607
+ "model.layers.17.mlp.switch_mlp.down_proj": {
608
+ "group_size": 32,
609
+ "bits": 6
610
+ },
611
+ "model.layers.18.self_attn.q_proj": {
612
+ "group_size": 32,
613
+ "bits": 4
614
+ },
615
+ "model.layers.18.self_attn.k_proj": {
616
+ "group_size": 32,
617
+ "bits": 4
618
+ },
619
+ "model.layers.18.self_attn.v_proj": {
620
+ "group_size": 32,
621
+ "bits": 6
622
+ },
623
+ "model.layers.18.self_attn.o_proj": {
624
+ "group_size": 32,
625
+ "bits": 4
626
+ },
627
+ "model.layers.18.mlp.gate": {
628
+ "group_size": 32,
629
+ "bits": 4
630
+ },
631
+ "model.layers.18.mlp.switch_mlp.gate_proj": {
632
+ "group_size": 32,
633
+ "bits": 4
634
+ },
635
+ "model.layers.18.mlp.switch_mlp.up_proj": {
636
+ "group_size": 32,
637
+ "bits": 4
638
+ },
639
+ "model.layers.18.mlp.switch_mlp.down_proj": {
640
+ "group_size": 32,
641
+ "bits": 4
642
+ },
643
+ "model.layers.19.self_attn.q_proj": {
644
+ "group_size": 32,
645
+ "bits": 4
646
+ },
647
+ "model.layers.19.self_attn.k_proj": {
648
+ "group_size": 32,
649
+ "bits": 4
650
+ },
651
+ "model.layers.19.self_attn.v_proj": {
652
+ "group_size": 32,
653
+ "bits": 6
654
+ },
655
+ "model.layers.19.self_attn.o_proj": {
656
+ "group_size": 32,
657
+ "bits": 4
658
+ },
659
+ "model.layers.19.mlp.gate": {
660
+ "group_size": 32,
661
+ "bits": 4
662
+ },
663
+ "model.layers.19.mlp.switch_mlp.gate_proj": {
664
+ "group_size": 32,
665
+ "bits": 4
666
+ },
667
+ "model.layers.19.mlp.switch_mlp.up_proj": {
668
+ "group_size": 32,
669
+ "bits": 4
670
+ },
671
+ "model.layers.19.mlp.switch_mlp.down_proj": {
672
+ "group_size": 32,
673
+ "bits": 4
674
+ },
675
+ "model.layers.20.self_attn.q_proj": {
676
+ "group_size": 32,
677
+ "bits": 4
678
+ },
679
+ "model.layers.20.self_attn.k_proj": {
680
+ "group_size": 32,
681
+ "bits": 4
682
+ },
683
+ "model.layers.20.self_attn.v_proj": {
684
+ "group_size": 32,
685
+ "bits": 6
686
+ },
687
+ "model.layers.20.self_attn.o_proj": {
688
+ "group_size": 32,
689
+ "bits": 4
690
+ },
691
+ "model.layers.20.mlp.gate": {
692
+ "group_size": 32,
693
+ "bits": 4
694
+ },
695
+ "model.layers.20.mlp.switch_mlp.gate_proj": {
696
+ "group_size": 32,
697
+ "bits": 4
698
+ },
699
+ "model.layers.20.mlp.switch_mlp.up_proj": {
700
+ "group_size": 32,
701
+ "bits": 4
702
+ },
703
+ "model.layers.20.mlp.switch_mlp.down_proj": {
704
+ "group_size": 32,
705
+ "bits": 6
706
+ },
707
+ "model.layers.21.self_attn.q_proj": {
708
+ "group_size": 32,
709
+ "bits": 4
710
+ },
711
+ "model.layers.21.self_attn.k_proj": {
712
+ "group_size": 32,
713
+ "bits": 4
714
+ },
715
+ "model.layers.21.self_attn.v_proj": {
716
+ "group_size": 32,
717
+ "bits": 6
718
+ },
719
+ "model.layers.21.self_attn.o_proj": {
720
+ "group_size": 32,
721
+ "bits": 4
722
+ },
723
+ "model.layers.21.mlp.gate": {
724
+ "group_size": 32,
725
+ "bits": 4
726
+ },
727
+ "model.layers.21.mlp.switch_mlp.gate_proj": {
728
+ "group_size": 32,
729
+ "bits": 4
730
+ },
731
+ "model.layers.21.mlp.switch_mlp.up_proj": {
732
+ "group_size": 32,
733
+ "bits": 4
734
+ },
735
+ "model.layers.21.mlp.switch_mlp.down_proj": {
736
+ "group_size": 32,
737
+ "bits": 4
738
+ },
739
+ "model.layers.22.self_attn.q_proj": {
740
+ "group_size": 32,
741
+ "bits": 4
742
+ },
743
+ "model.layers.22.self_attn.k_proj": {
744
+ "group_size": 32,
745
+ "bits": 4
746
+ },
747
+ "model.layers.22.self_attn.v_proj": {
748
+ "group_size": 32,
749
+ "bits": 6
750
+ },
751
+ "model.layers.22.self_attn.o_proj": {
752
+ "group_size": 32,
753
+ "bits": 4
754
+ },
755
+ "model.layers.22.mlp.gate": {
756
+ "group_size": 32,
757
+ "bits": 4
758
+ },
759
+ "model.layers.22.mlp.switch_mlp.gate_proj": {
760
+ "group_size": 32,
761
+ "bits": 4
762
+ },
763
+ "model.layers.22.mlp.switch_mlp.up_proj": {
764
+ "group_size": 32,
765
+ "bits": 4
766
+ },
767
+ "model.layers.22.mlp.switch_mlp.down_proj": {
768
+ "group_size": 32,
769
+ "bits": 4
770
+ },
771
+ "model.layers.23.self_attn.q_proj": {
772
+ "group_size": 32,
773
+ "bits": 4
774
+ },
775
+ "model.layers.23.self_attn.k_proj": {
776
+ "group_size": 32,
777
+ "bits": 4
778
+ },
779
+ "model.layers.23.self_attn.v_proj": {
780
+ "group_size": 32,
781
+ "bits": 6
782
+ },
783
+ "model.layers.23.self_attn.o_proj": {
784
+ "group_size": 32,
785
+ "bits": 4
786
+ },
787
+ "model.layers.23.mlp.gate": {
788
+ "group_size": 32,
789
+ "bits": 4
790
+ },
791
+ "model.layers.23.mlp.switch_mlp.gate_proj": {
792
+ "group_size": 32,
793
+ "bits": 4
794
+ },
795
+ "model.layers.23.mlp.switch_mlp.up_proj": {
796
+ "group_size": 32,
797
+ "bits": 4
798
+ },
799
+ "model.layers.23.mlp.switch_mlp.down_proj": {
800
+ "group_size": 32,
801
+ "bits": 6
802
+ },
803
+ "model.layers.24.self_attn.q_proj": {
804
+ "group_size": 32,
805
+ "bits": 4
806
+ },
807
+ "model.layers.24.self_attn.k_proj": {
808
+ "group_size": 32,
809
+ "bits": 4
810
+ },
811
+ "model.layers.24.self_attn.v_proj": {
812
+ "group_size": 32,
813
+ "bits": 6
814
+ },
815
+ "model.layers.24.self_attn.o_proj": {
816
+ "group_size": 32,
817
+ "bits": 4
818
+ },
819
+ "model.layers.24.mlp.gate": {
820
+ "group_size": 32,
821
+ "bits": 4
822
+ },
823
+ "model.layers.24.mlp.switch_mlp.gate_proj": {
824
+ "group_size": 32,
825
+ "bits": 4
826
+ },
827
+ "model.layers.24.mlp.switch_mlp.up_proj": {
828
+ "group_size": 32,
829
+ "bits": 4
830
+ },
831
+ "model.layers.24.mlp.switch_mlp.down_proj": {
832
+ "group_size": 32,
833
+ "bits": 4
834
+ },
835
+ "model.layers.25.self_attn.q_proj": {
836
+ "group_size": 32,
837
+ "bits": 4
838
+ },
839
+ "model.layers.25.self_attn.k_proj": {
840
+ "group_size": 32,
841
+ "bits": 4
842
+ },
843
+ "model.layers.25.self_attn.v_proj": {
844
+ "group_size": 32,
845
+ "bits": 6
846
+ },
847
+ "model.layers.25.self_attn.o_proj": {
848
+ "group_size": 32,
849
+ "bits": 4
850
+ },
851
+ "model.layers.25.mlp.gate": {
852
+ "group_size": 32,
853
+ "bits": 4
854
+ },
855
+ "model.layers.25.mlp.switch_mlp.gate_proj": {
856
+ "group_size": 32,
857
+ "bits": 4
858
+ },
859
+ "model.layers.25.mlp.switch_mlp.up_proj": {
860
+ "group_size": 32,
861
+ "bits": 4
862
+ },
863
+ "model.layers.25.mlp.switch_mlp.down_proj": {
864
+ "group_size": 32,
865
+ "bits": 4
866
+ },
867
+ "model.layers.26.self_attn.q_proj": {
868
+ "group_size": 32,
869
+ "bits": 4
870
+ },
871
+ "model.layers.26.self_attn.k_proj": {
872
+ "group_size": 32,
873
+ "bits": 4
874
+ },
875
+ "model.layers.26.self_attn.v_proj": {
876
+ "group_size": 32,
877
+ "bits": 6
878
+ },
879
+ "model.layers.26.self_attn.o_proj": {
880
+ "group_size": 32,
881
+ "bits": 4
882
+ },
883
+ "model.layers.26.mlp.gate": {
884
+ "group_size": 32,
885
+ "bits": 4
886
+ },
887
+ "model.layers.26.mlp.switch_mlp.gate_proj": {
888
+ "group_size": 32,
889
+ "bits": 4
890
+ },
891
+ "model.layers.26.mlp.switch_mlp.up_proj": {
892
+ "group_size": 32,
893
+ "bits": 4
894
+ },
895
+ "model.layers.26.mlp.switch_mlp.down_proj": {
896
+ "group_size": 32,
897
+ "bits": 6
898
+ },
899
+ "model.layers.27.self_attn.q_proj": {
900
+ "group_size": 32,
901
+ "bits": 4
902
+ },
903
+ "model.layers.27.self_attn.k_proj": {
904
+ "group_size": 32,
905
+ "bits": 4
906
+ },
907
+ "model.layers.27.self_attn.v_proj": {
908
+ "group_size": 32,
909
+ "bits": 6
910
+ },
911
+ "model.layers.27.self_attn.o_proj": {
912
+ "group_size": 32,
913
+ "bits": 4
914
+ },
915
+ "model.layers.27.mlp.gate": {
916
+ "group_size": 32,
917
+ "bits": 4
918
+ },
919
+ "model.layers.27.mlp.switch_mlp.gate_proj": {
920
+ "group_size": 32,
921
+ "bits": 4
922
+ },
923
+ "model.layers.27.mlp.switch_mlp.up_proj": {
924
+ "group_size": 32,
925
+ "bits": 4
926
+ },
927
+ "model.layers.27.mlp.switch_mlp.down_proj": {
928
+ "group_size": 32,
929
+ "bits": 4
930
+ },
931
+ "model.layers.28.self_attn.q_proj": {
932
+ "group_size": 32,
933
+ "bits": 4
934
+ },
935
+ "model.layers.28.self_attn.k_proj": {
936
+ "group_size": 32,
937
+ "bits": 4
938
+ },
939
+ "model.layers.28.self_attn.v_proj": {
940
+ "group_size": 32,
941
+ "bits": 6
942
+ },
943
+ "model.layers.28.self_attn.o_proj": {
944
+ "group_size": 32,
945
+ "bits": 4
946
+ },
947
+ "model.layers.28.mlp.gate": {
948
+ "group_size": 32,
949
+ "bits": 4
950
+ },
951
+ "model.layers.28.mlp.switch_mlp.gate_proj": {
952
+ "group_size": 32,
953
+ "bits": 4
954
+ },
955
+ "model.layers.28.mlp.switch_mlp.up_proj": {
956
+ "group_size": 32,
957
+ "bits": 4
958
+ },
959
+ "model.layers.28.mlp.switch_mlp.down_proj": {
960
+ "group_size": 32,
961
+ "bits": 4
962
+ },
963
+ "model.layers.29.self_attn.q_proj": {
964
+ "group_size": 32,
965
+ "bits": 4
966
+ },
967
+ "model.layers.29.self_attn.k_proj": {
968
+ "group_size": 32,
969
+ "bits": 4
970
+ },
971
+ "model.layers.29.self_attn.v_proj": {
972
+ "group_size": 32,
973
+ "bits": 6
974
+ },
975
+ "model.layers.29.self_attn.o_proj": {
976
+ "group_size": 32,
977
+ "bits": 4
978
+ },
979
+ "model.layers.29.mlp.gate": {
980
+ "group_size": 32,
981
+ "bits": 4
982
+ },
983
+ "model.layers.29.mlp.switch_mlp.gate_proj": {
984
+ "group_size": 32,
985
+ "bits": 4
986
+ },
987
+ "model.layers.29.mlp.switch_mlp.up_proj": {
988
+ "group_size": 32,
989
+ "bits": 4
990
+ },
991
+ "model.layers.29.mlp.switch_mlp.down_proj": {
992
+ "group_size": 32,
993
+ "bits": 6
994
+ },
995
+ "model.layers.30.self_attn.q_proj": {
996
+ "group_size": 32,
997
+ "bits": 4
998
+ },
999
+ "model.layers.30.self_attn.k_proj": {
1000
+ "group_size": 32,
1001
+ "bits": 4
1002
+ },
1003
+ "model.layers.30.self_attn.v_proj": {
1004
+ "group_size": 32,
1005
+ "bits": 6
1006
+ },
1007
+ "model.layers.30.self_attn.o_proj": {
1008
+ "group_size": 32,
1009
+ "bits": 4
1010
+ },
1011
+ "model.layers.30.mlp.gate": {
1012
+ "group_size": 32,
1013
+ "bits": 4
1014
+ },
1015
+ "model.layers.30.mlp.switch_mlp.gate_proj": {
1016
+ "group_size": 32,
1017
+ "bits": 4
1018
+ },
1019
+ "model.layers.30.mlp.switch_mlp.up_proj": {
1020
+ "group_size": 32,
1021
+ "bits": 4
1022
+ },
1023
+ "model.layers.30.mlp.switch_mlp.down_proj": {
1024
+ "group_size": 32,
1025
+ "bits": 4
1026
+ },
1027
+ "model.layers.31.self_attn.q_proj": {
1028
+ "group_size": 32,
1029
+ "bits": 4
1030
+ },
1031
+ "model.layers.31.self_attn.k_proj": {
1032
+ "group_size": 32,
1033
+ "bits": 4
1034
+ },
1035
+ "model.layers.31.self_attn.v_proj": {
1036
+ "group_size": 32,
1037
+ "bits": 6
1038
+ },
1039
+ "model.layers.31.self_attn.o_proj": {
1040
+ "group_size": 32,
1041
+ "bits": 4
1042
+ },
1043
+ "model.layers.31.mlp.gate": {
1044
+ "group_size": 32,
1045
+ "bits": 4
1046
+ },
1047
+ "model.layers.31.mlp.switch_mlp.gate_proj": {
1048
+ "group_size": 32,
1049
+ "bits": 4
1050
+ },
1051
+ "model.layers.31.mlp.switch_mlp.up_proj": {
1052
+ "group_size": 32,
1053
+ "bits": 4
1054
+ },
1055
+ "model.layers.31.mlp.switch_mlp.down_proj": {
1056
+ "group_size": 32,
1057
+ "bits": 4
1058
+ },
1059
+ "model.layers.32.self_attn.q_proj": {
1060
+ "group_size": 32,
1061
+ "bits": 4
1062
+ },
1063
+ "model.layers.32.self_attn.k_proj": {
1064
+ "group_size": 32,
1065
+ "bits": 4
1066
+ },
1067
+ "model.layers.32.self_attn.v_proj": {
1068
+ "group_size": 32,
1069
+ "bits": 6
1070
+ },
1071
+ "model.layers.32.self_attn.o_proj": {
1072
+ "group_size": 32,
1073
+ "bits": 4
1074
+ },
1075
+ "model.layers.32.mlp.gate": {
1076
+ "group_size": 32,
1077
+ "bits": 4
1078
+ },
1079
+ "model.layers.32.mlp.switch_mlp.gate_proj": {
1080
+ "group_size": 32,
1081
+ "bits": 4
1082
+ },
1083
+ "model.layers.32.mlp.switch_mlp.up_proj": {
1084
+ "group_size": 32,
1085
+ "bits": 4
1086
+ },
1087
+ "model.layers.32.mlp.switch_mlp.down_proj": {
1088
+ "group_size": 32,
1089
+ "bits": 6
1090
+ },
1091
+ "model.layers.33.self_attn.q_proj": {
1092
+ "group_size": 32,
1093
+ "bits": 4
1094
+ },
1095
+ "model.layers.33.self_attn.k_proj": {
1096
+ "group_size": 32,
1097
+ "bits": 4
1098
+ },
1099
+ "model.layers.33.self_attn.v_proj": {
1100
+ "group_size": 32,
1101
+ "bits": 6
1102
+ },
1103
+ "model.layers.33.self_attn.o_proj": {
1104
+ "group_size": 32,
1105
+ "bits": 4
1106
+ },
1107
+ "model.layers.33.mlp.gate": {
1108
+ "group_size": 32,
1109
+ "bits": 4
1110
+ },
1111
+ "model.layers.33.mlp.switch_mlp.gate_proj": {
1112
+ "group_size": 32,
1113
+ "bits": 4
1114
+ },
1115
+ "model.layers.33.mlp.switch_mlp.up_proj": {
1116
+ "group_size": 32,
1117
+ "bits": 4
1118
+ },
1119
+ "model.layers.33.mlp.switch_mlp.down_proj": {
1120
+ "group_size": 32,
1121
+ "bits": 4
1122
+ },
1123
+ "model.layers.34.self_attn.q_proj": {
1124
+ "group_size": 32,
1125
+ "bits": 4
1126
+ },
1127
+ "model.layers.34.self_attn.k_proj": {
1128
+ "group_size": 32,
1129
+ "bits": 4
1130
+ },
1131
+ "model.layers.34.self_attn.v_proj": {
1132
+ "group_size": 32,
1133
+ "bits": 6
1134
+ },
1135
+ "model.layers.34.self_attn.o_proj": {
1136
+ "group_size": 32,
1137
+ "bits": 4
1138
+ },
1139
+ "model.layers.34.mlp.gate": {
1140
+ "group_size": 32,
1141
+ "bits": 4
1142
+ },
1143
+ "model.layers.34.mlp.switch_mlp.gate_proj": {
1144
+ "group_size": 32,
1145
+ "bits": 4
1146
+ },
1147
+ "model.layers.34.mlp.switch_mlp.up_proj": {
1148
+ "group_size": 32,
1149
+ "bits": 4
1150
+ },
1151
+ "model.layers.34.mlp.switch_mlp.down_proj": {
1152
+ "group_size": 32,
1153
+ "bits": 4
1154
+ },
1155
+ "model.layers.35.self_attn.q_proj": {
1156
+ "group_size": 32,
1157
+ "bits": 4
1158
+ },
1159
+ "model.layers.35.self_attn.k_proj": {
1160
+ "group_size": 32,
1161
+ "bits": 4
1162
+ },
1163
+ "model.layers.35.self_attn.v_proj": {
1164
+ "group_size": 32,
1165
+ "bits": 6
1166
+ },
1167
+ "model.layers.35.self_attn.o_proj": {
1168
+ "group_size": 32,
1169
+ "bits": 4
1170
+ },
1171
+ "model.layers.35.mlp.gate": {
1172
+ "group_size": 32,
1173
+ "bits": 4
1174
+ },
1175
+ "model.layers.35.mlp.switch_mlp.gate_proj": {
1176
+ "group_size": 32,
1177
+ "bits": 4
1178
+ },
1179
+ "model.layers.35.mlp.switch_mlp.up_proj": {
1180
+ "group_size": 32,
1181
+ "bits": 4
1182
+ },
1183
+ "model.layers.35.mlp.switch_mlp.down_proj": {
1184
+ "group_size": 32,
1185
+ "bits": 6
1186
+ },
1187
+ "model.layers.36.self_attn.q_proj": {
1188
+ "group_size": 32,
1189
+ "bits": 4
1190
+ },
1191
+ "model.layers.36.self_attn.k_proj": {
1192
+ "group_size": 32,
1193
+ "bits": 4
1194
+ },
1195
+ "model.layers.36.self_attn.v_proj": {
1196
+ "group_size": 32,
1197
+ "bits": 6
1198
+ },
1199
+ "model.layers.36.self_attn.o_proj": {
1200
+ "group_size": 32,
1201
+ "bits": 4
1202
+ },
1203
+ "model.layers.36.mlp.gate": {
1204
+ "group_size": 32,
1205
+ "bits": 4
1206
+ },
1207
+ "model.layers.36.mlp.switch_mlp.gate_proj": {
1208
+ "group_size": 32,
1209
+ "bits": 4
1210
+ },
1211
+ "model.layers.36.mlp.switch_mlp.up_proj": {
1212
+ "group_size": 32,
1213
+ "bits": 4
1214
+ },
1215
+ "model.layers.36.mlp.switch_mlp.down_proj": {
1216
+ "group_size": 32,
1217
+ "bits": 4
1218
+ },
1219
+ "model.layers.37.self_attn.q_proj": {
1220
+ "group_size": 32,
1221
+ "bits": 4
1222
+ },
1223
+ "model.layers.37.self_attn.k_proj": {
1224
+ "group_size": 32,
1225
+ "bits": 4
1226
+ },
1227
+ "model.layers.37.self_attn.v_proj": {
1228
+ "group_size": 32,
1229
+ "bits": 6
1230
+ },
1231
+ "model.layers.37.self_attn.o_proj": {
1232
+ "group_size": 32,
1233
+ "bits": 4
1234
+ },
1235
+ "model.layers.37.mlp.gate": {
1236
+ "group_size": 32,
1237
+ "bits": 4
1238
+ },
1239
+ "model.layers.37.mlp.switch_mlp.gate_proj": {
1240
+ "group_size": 32,
1241
+ "bits": 4
1242
+ },
1243
+ "model.layers.37.mlp.switch_mlp.up_proj": {
1244
+ "group_size": 32,
1245
+ "bits": 4
1246
+ },
1247
+ "model.layers.37.mlp.switch_mlp.down_proj": {
1248
+ "group_size": 32,
1249
+ "bits": 4
1250
+ },
1251
+ "model.layers.38.self_attn.q_proj": {
1252
+ "group_size": 32,
1253
+ "bits": 4
1254
+ },
1255
+ "model.layers.38.self_attn.k_proj": {
1256
+ "group_size": 32,
1257
+ "bits": 4
1258
+ },
1259
+ "model.layers.38.self_attn.v_proj": {
1260
+ "group_size": 32,
1261
+ "bits": 6
1262
+ },
1263
+ "model.layers.38.self_attn.o_proj": {
1264
+ "group_size": 32,
1265
+ "bits": 4
1266
+ },
1267
+ "model.layers.38.mlp.gate": {
1268
+ "group_size": 32,
1269
+ "bits": 4
1270
+ },
1271
+ "model.layers.38.mlp.switch_mlp.gate_proj": {
1272
+ "group_size": 32,
1273
+ "bits": 4
1274
+ },
1275
+ "model.layers.38.mlp.switch_mlp.up_proj": {
1276
+ "group_size": 32,
1277
+ "bits": 4
1278
+ },
1279
+ "model.layers.38.mlp.switch_mlp.down_proj": {
1280
+ "group_size": 32,
1281
+ "bits": 6
1282
+ },
1283
+ "model.layers.39.self_attn.q_proj": {
1284
+ "group_size": 32,
1285
+ "bits": 4
1286
+ },
1287
+ "model.layers.39.self_attn.k_proj": {
1288
+ "group_size": 32,
1289
+ "bits": 4
1290
+ },
1291
+ "model.layers.39.self_attn.v_proj": {
1292
+ "group_size": 32,
1293
+ "bits": 6
1294
+ },
1295
+ "model.layers.39.self_attn.o_proj": {
1296
+ "group_size": 32,
1297
+ "bits": 4
1298
+ },
1299
+ "model.layers.39.mlp.gate": {
1300
+ "group_size": 32,
1301
+ "bits": 4
1302
+ },
1303
+ "model.layers.39.mlp.switch_mlp.gate_proj": {
1304
+ "group_size": 32,
1305
+ "bits": 4
1306
+ },
1307
+ "model.layers.39.mlp.switch_mlp.up_proj": {
1308
+ "group_size": 32,
1309
+ "bits": 4
1310
+ },
1311
+ "model.layers.39.mlp.switch_mlp.down_proj": {
1312
+ "group_size": 32,
1313
+ "bits": 4
1314
+ },
1315
+ "model.layers.40.self_attn.q_proj": {
1316
+ "group_size": 32,
1317
+ "bits": 4
1318
+ },
1319
+ "model.layers.40.self_attn.k_proj": {
1320
+ "group_size": 32,
1321
+ "bits": 4
1322
+ },
1323
+ "model.layers.40.self_attn.v_proj": {
1324
+ "group_size": 32,
1325
+ "bits": 6
1326
+ },
1327
+ "model.layers.40.self_attn.o_proj": {
1328
+ "group_size": 32,
1329
+ "bits": 4
1330
+ },
1331
+ "model.layers.40.mlp.gate": {
1332
+ "group_size": 32,
1333
+ "bits": 4
1334
+ },
1335
+ "model.layers.40.mlp.switch_mlp.gate_proj": {
1336
+ "group_size": 32,
1337
+ "bits": 4
1338
+ },
1339
+ "model.layers.40.mlp.switch_mlp.up_proj": {
1340
+ "group_size": 32,
1341
+ "bits": 4
1342
+ },
1343
+ "model.layers.40.mlp.switch_mlp.down_proj": {
1344
+ "group_size": 32,
1345
+ "bits": 4
1346
+ },
1347
+ "model.layers.41.self_attn.q_proj": {
1348
+ "group_size": 32,
1349
+ "bits": 4
1350
+ },
1351
+ "model.layers.41.self_attn.k_proj": {
1352
+ "group_size": 32,
1353
+ "bits": 4
1354
+ },
1355
+ "model.layers.41.self_attn.v_proj": {
1356
+ "group_size": 32,
1357
+ "bits": 6
1358
+ },
1359
+ "model.layers.41.self_attn.o_proj": {
1360
+ "group_size": 32,
1361
+ "bits": 4
1362
+ },
1363
+ "model.layers.41.mlp.gate": {
1364
+ "group_size": 32,
1365
+ "bits": 4
1366
+ },
1367
+ "model.layers.41.mlp.switch_mlp.gate_proj": {
1368
+ "group_size": 32,
1369
+ "bits": 4
1370
+ },
1371
+ "model.layers.41.mlp.switch_mlp.up_proj": {
1372
+ "group_size": 32,
1373
+ "bits": 4
1374
+ },
1375
+ "model.layers.41.mlp.switch_mlp.down_proj": {
1376
+ "group_size": 32,
1377
+ "bits": 6
1378
+ },
1379
+ "model.layers.42.self_attn.q_proj": {
1380
+ "group_size": 32,
1381
+ "bits": 4
1382
+ },
1383
+ "model.layers.42.self_attn.k_proj": {
1384
+ "group_size": 32,
1385
+ "bits": 4
1386
+ },
1387
+ "model.layers.42.self_attn.v_proj": {
1388
+ "group_size": 32,
1389
+ "bits": 6
1390
+ },
1391
+ "model.layers.42.self_attn.o_proj": {
1392
+ "group_size": 32,
1393
+ "bits": 4
1394
+ },
1395
+ "model.layers.42.mlp.gate": {
1396
+ "group_size": 32,
1397
+ "bits": 4
1398
+ },
1399
+ "model.layers.42.mlp.switch_mlp.gate_proj": {
1400
+ "group_size": 32,
1401
+ "bits": 4
1402
+ },
1403
+ "model.layers.42.mlp.switch_mlp.up_proj": {
1404
+ "group_size": 32,
1405
+ "bits": 4
1406
+ },
1407
+ "model.layers.42.mlp.switch_mlp.down_proj": {
1408
+ "group_size": 32,
1409
+ "bits": 6
1410
+ },
1411
+ "model.layers.43.self_attn.q_proj": {
1412
+ "group_size": 32,
1413
+ "bits": 4
1414
+ },
1415
+ "model.layers.43.self_attn.k_proj": {
1416
+ "group_size": 32,
1417
+ "bits": 4
1418
+ },
1419
+ "model.layers.43.self_attn.v_proj": {
1420
+ "group_size": 32,
1421
+ "bits": 6
1422
+ },
1423
+ "model.layers.43.self_attn.o_proj": {
1424
+ "group_size": 32,
1425
+ "bits": 4
1426
+ },
1427
+ "model.layers.43.mlp.gate": {
1428
+ "group_size": 32,
1429
+ "bits": 4
1430
+ },
1431
+ "model.layers.43.mlp.switch_mlp.gate_proj": {
1432
+ "group_size": 32,
1433
+ "bits": 4
1434
+ },
1435
+ "model.layers.43.mlp.switch_mlp.up_proj": {
1436
+ "group_size": 32,
1437
+ "bits": 4
1438
+ },
1439
+ "model.layers.43.mlp.switch_mlp.down_proj": {
1440
+ "group_size": 32,
1441
+ "bits": 6
1442
+ },
1443
+ "model.layers.44.self_attn.q_proj": {
1444
+ "group_size": 32,
1445
+ "bits": 4
1446
+ },
1447
+ "model.layers.44.self_attn.k_proj": {
1448
+ "group_size": 32,
1449
+ "bits": 4
1450
+ },
1451
+ "model.layers.44.self_attn.v_proj": {
1452
+ "group_size": 32,
1453
+ "bits": 6
1454
+ },
1455
+ "model.layers.44.self_attn.o_proj": {
1456
+ "group_size": 32,
1457
+ "bits": 4
1458
+ },
1459
+ "model.layers.44.mlp.gate": {
1460
+ "group_size": 32,
1461
+ "bits": 4
1462
+ },
1463
+ "model.layers.44.mlp.switch_mlp.gate_proj": {
1464
+ "group_size": 32,
1465
+ "bits": 4
1466
+ },
1467
+ "model.layers.44.mlp.switch_mlp.up_proj": {
1468
+ "group_size": 32,
1469
+ "bits": 4
1470
+ },
1471
+ "model.layers.44.mlp.switch_mlp.down_proj": {
1472
+ "group_size": 32,
1473
+ "bits": 6
1474
+ },
1475
+ "model.layers.45.self_attn.q_proj": {
1476
+ "group_size": 32,
1477
+ "bits": 4
1478
+ },
1479
+ "model.layers.45.self_attn.k_proj": {
1480
+ "group_size": 32,
1481
+ "bits": 4
1482
+ },
1483
+ "model.layers.45.self_attn.v_proj": {
1484
+ "group_size": 32,
1485
+ "bits": 6
1486
+ },
1487
+ "model.layers.45.self_attn.o_proj": {
1488
+ "group_size": 32,
1489
+ "bits": 4
1490
+ },
1491
+ "model.layers.45.mlp.gate": {
1492
+ "group_size": 32,
1493
+ "bits": 4
1494
+ },
1495
+ "model.layers.45.mlp.switch_mlp.gate_proj": {
1496
+ "group_size": 32,
1497
+ "bits": 4
1498
+ },
1499
+ "model.layers.45.mlp.switch_mlp.up_proj": {
1500
+ "group_size": 32,
1501
+ "bits": 4
1502
+ },
1503
+ "model.layers.45.mlp.switch_mlp.down_proj": {
1504
+ "group_size": 32,
1505
+ "bits": 6
1506
+ },
1507
+ "model.layers.46.self_attn.q_proj": {
1508
+ "group_size": 32,
1509
+ "bits": 4
1510
+ },
1511
+ "model.layers.46.self_attn.k_proj": {
1512
+ "group_size": 32,
1513
+ "bits": 4
1514
+ },
1515
+ "model.layers.46.self_attn.v_proj": {
1516
+ "group_size": 32,
1517
+ "bits": 6
1518
+ },
1519
+ "model.layers.46.self_attn.o_proj": {
1520
+ "group_size": 32,
1521
+ "bits": 4
1522
+ },
1523
+ "model.layers.46.mlp.gate": {
1524
+ "group_size": 32,
1525
+ "bits": 4
1526
+ },
1527
+ "model.layers.46.mlp.switch_mlp.gate_proj": {
1528
+ "group_size": 32,
1529
+ "bits": 4
1530
+ },
1531
+ "model.layers.46.mlp.switch_mlp.up_proj": {
1532
+ "group_size": 32,
1533
+ "bits": 4
1534
+ },
1535
+ "model.layers.46.mlp.switch_mlp.down_proj": {
1536
+ "group_size": 32,
1537
+ "bits": 6
1538
+ },
1539
+ "model.layers.47.self_attn.q_proj": {
1540
+ "group_size": 32,
1541
+ "bits": 4
1542
+ },
1543
+ "model.layers.47.self_attn.k_proj": {
1544
+ "group_size": 32,
1545
+ "bits": 4
1546
+ },
1547
+ "model.layers.47.self_attn.v_proj": {
1548
+ "group_size": 32,
1549
+ "bits": 6
1550
+ },
1551
+ "model.layers.47.self_attn.o_proj": {
1552
+ "group_size": 32,
1553
+ "bits": 4
1554
+ },
1555
+ "model.layers.47.mlp.gate": {
1556
+ "group_size": 32,
1557
+ "bits": 4
1558
+ },
1559
+ "model.layers.47.mlp.switch_mlp.gate_proj": {
1560
+ "group_size": 32,
1561
+ "bits": 4
1562
+ },
1563
+ "model.layers.47.mlp.switch_mlp.up_proj": {
1564
+ "group_size": 32,
1565
+ "bits": 4
1566
+ },
1567
+ "model.layers.47.mlp.switch_mlp.down_proj": {
1568
+ "group_size": 32,
1569
+ "bits": 6
1570
+ },
1571
+ "lm_head": {
1572
+ "group_size": 32,
1573
+ "bits": 6
1574
+ }
1575
+ },
1576
+ "quantization_config": {
1577
+ "group_size": 32,
1578
+ "bits": 6,
1579
+ "mode": "affine",
1580
+ "model.embed_tokens": {
1581
+ "group_size": 32,
1582
+ "bits": 4
1583
+ },
1584
+ "model.layers.0.self_attn.q_proj": {
1585
+ "group_size": 32,
1586
+ "bits": 6
1587
+ },
1588
+ "model.layers.0.self_attn.k_proj": {
1589
+ "group_size": 32,
1590
+ "bits": 6
1591
+ },
1592
+ "model.layers.0.self_attn.v_proj": {
1593
+ "group_size": 32,
1594
+ "bits": 6
1595
+ },
1596
+ "model.layers.0.self_attn.o_proj": {
1597
+ "group_size": 32,
1598
+ "bits": 6
1599
+ },
1600
+ "model.layers.0.mlp.gate": {
1601
+ "group_size": 32,
1602
+ "bits": 6
1603
+ },
1604
+ "model.layers.0.mlp.switch_mlp.gate_proj": {
1605
+ "group_size": 32,
1606
+ "bits": 6
1607
+ },
1608
+ "model.layers.0.mlp.switch_mlp.up_proj": {
1609
+ "group_size": 32,
1610
+ "bits": 6
1611
+ },
1612
+ "model.layers.0.mlp.switch_mlp.down_proj": {
1613
+ "group_size": 32,
1614
+ "bits": 6
1615
+ },
1616
+ "model.layers.1.self_attn.q_proj": {
1617
+ "group_size": 32,
1618
+ "bits": 4
1619
+ },
1620
+ "model.layers.1.self_attn.k_proj": {
1621
+ "group_size": 32,
1622
+ "bits": 4
1623
+ },
1624
+ "model.layers.1.self_attn.v_proj": {
1625
+ "group_size": 32,
1626
+ "bits": 6
1627
+ },
1628
+ "model.layers.1.self_attn.o_proj": {
1629
+ "group_size": 32,
1630
+ "bits": 4
1631
+ },
1632
+ "model.layers.1.mlp.gate": {
1633
+ "group_size": 32,
1634
+ "bits": 4
1635
+ },
1636
+ "model.layers.1.mlp.switch_mlp.gate_proj": {
1637
+ "group_size": 32,
1638
+ "bits": 4
1639
+ },
1640
+ "model.layers.1.mlp.switch_mlp.up_proj": {
1641
+ "group_size": 32,
1642
+ "bits": 4
1643
+ },
1644
+ "model.layers.1.mlp.switch_mlp.down_proj": {
1645
+ "group_size": 32,
1646
+ "bits": 6
1647
+ },
1648
+ "model.layers.2.self_attn.q_proj": {
1649
+ "group_size": 32,
1650
+ "bits": 4
1651
+ },
1652
+ "model.layers.2.self_attn.k_proj": {
1653
+ "group_size": 32,
1654
+ "bits": 4
1655
+ },
1656
+ "model.layers.2.self_attn.v_proj": {
1657
+ "group_size": 32,
1658
+ "bits": 6
1659
+ },
1660
+ "model.layers.2.self_attn.o_proj": {
1661
+ "group_size": 32,
1662
+ "bits": 4
1663
+ },
1664
+ "model.layers.2.mlp.gate": {
1665
+ "group_size": 32,
1666
+ "bits": 4
1667
+ },
1668
+ "model.layers.2.mlp.switch_mlp.gate_proj": {
1669
+ "group_size": 32,
1670
+ "bits": 4
1671
+ },
1672
+ "model.layers.2.mlp.switch_mlp.up_proj": {
1673
+ "group_size": 32,
1674
+ "bits": 4
1675
+ },
1676
+ "model.layers.2.mlp.switch_mlp.down_proj": {
1677
+ "group_size": 32,
1678
+ "bits": 6
1679
+ },
1680
+ "model.layers.3.self_attn.q_proj": {
1681
+ "group_size": 32,
1682
+ "bits": 4
1683
+ },
1684
+ "model.layers.3.self_attn.k_proj": {
1685
+ "group_size": 32,
1686
+ "bits": 4
1687
+ },
1688
+ "model.layers.3.self_attn.v_proj": {
1689
+ "group_size": 32,
1690
+ "bits": 6
1691
+ },
1692
+ "model.layers.3.self_attn.o_proj": {
1693
+ "group_size": 32,
1694
+ "bits": 4
1695
+ },
1696
+ "model.layers.3.mlp.gate": {
1697
+ "group_size": 32,
1698
+ "bits": 4
1699
+ },
1700
+ "model.layers.3.mlp.switch_mlp.gate_proj": {
1701
+ "group_size": 32,
1702
+ "bits": 4
1703
+ },
1704
+ "model.layers.3.mlp.switch_mlp.up_proj": {
1705
+ "group_size": 32,
1706
+ "bits": 4
1707
+ },
1708
+ "model.layers.3.mlp.switch_mlp.down_proj": {
1709
+ "group_size": 32,
1710
+ "bits": 6
1711
+ },
1712
+ "model.layers.4.self_attn.q_proj": {
1713
+ "group_size": 32,
1714
+ "bits": 4
1715
+ },
1716
+ "model.layers.4.self_attn.k_proj": {
1717
+ "group_size": 32,
1718
+ "bits": 4
1719
+ },
1720
+ "model.layers.4.self_attn.v_proj": {
1721
+ "group_size": 32,
1722
+ "bits": 6
1723
+ },
1724
+ "model.layers.4.self_attn.o_proj": {
1725
+ "group_size": 32,
1726
+ "bits": 4
1727
+ },
1728
+ "model.layers.4.mlp.gate": {
1729
+ "group_size": 32,
1730
+ "bits": 4
1731
+ },
1732
+ "model.layers.4.mlp.switch_mlp.gate_proj": {
1733
+ "group_size": 32,
1734
+ "bits": 4
1735
+ },
1736
+ "model.layers.4.mlp.switch_mlp.up_proj": {
1737
+ "group_size": 32,
1738
+ "bits": 4
1739
+ },
1740
+ "model.layers.4.mlp.switch_mlp.down_proj": {
1741
+ "group_size": 32,
1742
+ "bits": 6
1743
+ },
1744
+ "model.layers.5.self_attn.q_proj": {
1745
+ "group_size": 32,
1746
+ "bits": 4
1747
+ },
1748
+ "model.layers.5.self_attn.k_proj": {
1749
+ "group_size": 32,
1750
+ "bits": 4
1751
+ },
1752
+ "model.layers.5.self_attn.v_proj": {
1753
+ "group_size": 32,
1754
+ "bits": 6
1755
+ },
1756
+ "model.layers.5.self_attn.o_proj": {
1757
+ "group_size": 32,
1758
+ "bits": 4
1759
+ },
1760
+ "model.layers.5.mlp.gate": {
1761
+ "group_size": 32,
1762
+ "bits": 4
1763
+ },
1764
+ "model.layers.5.mlp.switch_mlp.gate_proj": {
1765
+ "group_size": 32,
1766
+ "bits": 4
1767
+ },
1768
+ "model.layers.5.mlp.switch_mlp.up_proj": {
1769
+ "group_size": 32,
1770
+ "bits": 4
1771
+ },
1772
+ "model.layers.5.mlp.switch_mlp.down_proj": {
1773
+ "group_size": 32,
1774
+ "bits": 6
1775
+ },
1776
+ "model.layers.6.self_attn.q_proj": {
1777
+ "group_size": 32,
1778
+ "bits": 4
1779
+ },
1780
+ "model.layers.6.self_attn.k_proj": {
1781
+ "group_size": 32,
1782
+ "bits": 4
1783
+ },
1784
+ "model.layers.6.self_attn.v_proj": {
1785
+ "group_size": 32,
1786
+ "bits": 6
1787
+ },
1788
+ "model.layers.6.self_attn.o_proj": {
1789
+ "group_size": 32,
1790
+ "bits": 4
1791
+ },
1792
+ "model.layers.6.mlp.gate": {
1793
+ "group_size": 32,
1794
+ "bits": 4
1795
+ },
1796
+ "model.layers.6.mlp.switch_mlp.gate_proj": {
1797
+ "group_size": 32,
1798
+ "bits": 4
1799
+ },
1800
+ "model.layers.6.mlp.switch_mlp.up_proj": {
1801
+ "group_size": 32,
1802
+ "bits": 4
1803
+ },
1804
+ "model.layers.6.mlp.switch_mlp.down_proj": {
1805
+ "group_size": 32,
1806
+ "bits": 4
1807
+ },
1808
+ "model.layers.7.self_attn.q_proj": {
1809
+ "group_size": 32,
1810
+ "bits": 4
1811
+ },
1812
+ "model.layers.7.self_attn.k_proj": {
1813
+ "group_size": 32,
1814
+ "bits": 4
1815
+ },
1816
+ "model.layers.7.self_attn.v_proj": {
1817
+ "group_size": 32,
1818
+ "bits": 6
1819
+ },
1820
+ "model.layers.7.self_attn.o_proj": {
1821
+ "group_size": 32,
1822
+ "bits": 4
1823
+ },
1824
+ "model.layers.7.mlp.gate": {
1825
+ "group_size": 32,
1826
+ "bits": 4
1827
+ },
1828
+ "model.layers.7.mlp.switch_mlp.gate_proj": {
1829
+ "group_size": 32,
1830
+ "bits": 4
1831
+ },
1832
+ "model.layers.7.mlp.switch_mlp.up_proj": {
1833
+ "group_size": 32,
1834
+ "bits": 4
1835
+ },
1836
+ "model.layers.7.mlp.switch_mlp.down_proj": {
1837
+ "group_size": 32,
1838
+ "bits": 4
1839
+ },
1840
+ "model.layers.8.self_attn.q_proj": {
1841
+ "group_size": 32,
1842
+ "bits": 4
1843
+ },
1844
+ "model.layers.8.self_attn.k_proj": {
1845
+ "group_size": 32,
1846
+ "bits": 4
1847
+ },
1848
+ "model.layers.8.self_attn.v_proj": {
1849
+ "group_size": 32,
1850
+ "bits": 6
1851
+ },
1852
+ "model.layers.8.self_attn.o_proj": {
1853
+ "group_size": 32,
1854
+ "bits": 4
1855
+ },
1856
+ "model.layers.8.mlp.gate": {
1857
+ "group_size": 32,
1858
+ "bits": 4
1859
+ },
1860
+ "model.layers.8.mlp.switch_mlp.gate_proj": {
1861
+ "group_size": 32,
1862
+ "bits": 4
1863
+ },
1864
+ "model.layers.8.mlp.switch_mlp.up_proj": {
1865
+ "group_size": 32,
1866
+ "bits": 4
1867
+ },
1868
+ "model.layers.8.mlp.switch_mlp.down_proj": {
1869
+ "group_size": 32,
1870
+ "bits": 6
1871
+ },
1872
+ "model.layers.9.self_attn.q_proj": {
1873
+ "group_size": 32,
1874
+ "bits": 4
1875
+ },
1876
+ "model.layers.9.self_attn.k_proj": {
1877
+ "group_size": 32,
1878
+ "bits": 4
1879
+ },
1880
+ "model.layers.9.self_attn.v_proj": {
1881
+ "group_size": 32,
1882
+ "bits": 6
1883
+ },
1884
+ "model.layers.9.self_attn.o_proj": {
1885
+ "group_size": 32,
1886
+ "bits": 4
1887
+ },
1888
+ "model.layers.9.mlp.gate": {
1889
+ "group_size": 32,
1890
+ "bits": 4
1891
+ },
1892
+ "model.layers.9.mlp.switch_mlp.gate_proj": {
1893
+ "group_size": 32,
1894
+ "bits": 4
1895
+ },
1896
+ "model.layers.9.mlp.switch_mlp.up_proj": {
1897
+ "group_size": 32,
1898
+ "bits": 4
1899
+ },
1900
+ "model.layers.9.mlp.switch_mlp.down_proj": {
1901
+ "group_size": 32,
1902
+ "bits": 4
1903
+ },
1904
+ "model.layers.10.self_attn.q_proj": {
1905
+ "group_size": 32,
1906
+ "bits": 4
1907
+ },
1908
+ "model.layers.10.self_attn.k_proj": {
1909
+ "group_size": 32,
1910
+ "bits": 4
1911
+ },
1912
+ "model.layers.10.self_attn.v_proj": {
1913
+ "group_size": 32,
1914
+ "bits": 6
1915
+ },
1916
+ "model.layers.10.self_attn.o_proj": {
1917
+ "group_size": 32,
1918
+ "bits": 4
1919
+ },
1920
+ "model.layers.10.mlp.gate": {
1921
+ "group_size": 32,
1922
+ "bits": 4
1923
+ },
1924
+ "model.layers.10.mlp.switch_mlp.gate_proj": {
1925
+ "group_size": 32,
1926
+ "bits": 4
1927
+ },
1928
+ "model.layers.10.mlp.switch_mlp.up_proj": {
1929
+ "group_size": 32,
1930
+ "bits": 4
1931
+ },
1932
+ "model.layers.10.mlp.switch_mlp.down_proj": {
1933
+ "group_size": 32,
1934
+ "bits": 4
1935
+ },
1936
+ "model.layers.11.self_attn.q_proj": {
1937
+ "group_size": 32,
1938
+ "bits": 4
1939
+ },
1940
+ "model.layers.11.self_attn.k_proj": {
1941
+ "group_size": 32,
1942
+ "bits": 4
1943
+ },
1944
+ "model.layers.11.self_attn.v_proj": {
1945
+ "group_size": 32,
1946
+ "bits": 6
1947
+ },
1948
+ "model.layers.11.self_attn.o_proj": {
1949
+ "group_size": 32,
1950
+ "bits": 4
1951
+ },
1952
+ "model.layers.11.mlp.gate": {
1953
+ "group_size": 32,
1954
+ "bits": 4
1955
+ },
1956
+ "model.layers.11.mlp.switch_mlp.gate_proj": {
1957
+ "group_size": 32,
1958
+ "bits": 4
1959
+ },
1960
+ "model.layers.11.mlp.switch_mlp.up_proj": {
1961
+ "group_size": 32,
1962
+ "bits": 4
1963
+ },
1964
+ "model.layers.11.mlp.switch_mlp.down_proj": {
1965
+ "group_size": 32,
1966
+ "bits": 6
1967
+ },
1968
+ "model.layers.12.self_attn.q_proj": {
1969
+ "group_size": 32,
1970
+ "bits": 4
1971
+ },
1972
+ "model.layers.12.self_attn.k_proj": {
1973
+ "group_size": 32,
1974
+ "bits": 4
1975
+ },
1976
+ "model.layers.12.self_attn.v_proj": {
1977
+ "group_size": 32,
1978
+ "bits": 6
1979
+ },
1980
+ "model.layers.12.self_attn.o_proj": {
1981
+ "group_size": 32,
1982
+ "bits": 4
1983
+ },
1984
+ "model.layers.12.mlp.gate": {
1985
+ "group_size": 32,
1986
+ "bits": 4
1987
+ },
1988
+ "model.layers.12.mlp.switch_mlp.gate_proj": {
1989
+ "group_size": 32,
1990
+ "bits": 4
1991
+ },
1992
+ "model.layers.12.mlp.switch_mlp.up_proj": {
1993
+ "group_size": 32,
1994
+ "bits": 4
1995
+ },
1996
+ "model.layers.12.mlp.switch_mlp.down_proj": {
1997
+ "group_size": 32,
1998
+ "bits": 4
1999
+ },
2000
+ "model.layers.13.self_attn.q_proj": {
2001
+ "group_size": 32,
2002
+ "bits": 4
2003
+ },
2004
+ "model.layers.13.self_attn.k_proj": {
2005
+ "group_size": 32,
2006
+ "bits": 4
2007
+ },
2008
+ "model.layers.13.self_attn.v_proj": {
2009
+ "group_size": 32,
2010
+ "bits": 6
2011
+ },
2012
+ "model.layers.13.self_attn.o_proj": {
2013
+ "group_size": 32,
2014
+ "bits": 4
2015
+ },
2016
+ "model.layers.13.mlp.gate": {
2017
+ "group_size": 32,
2018
+ "bits": 4
2019
+ },
2020
+ "model.layers.13.mlp.switch_mlp.gate_proj": {
2021
+ "group_size": 32,
2022
+ "bits": 4
2023
+ },
2024
+ "model.layers.13.mlp.switch_mlp.up_proj": {
2025
+ "group_size": 32,
2026
+ "bits": 4
2027
+ },
2028
+ "model.layers.13.mlp.switch_mlp.down_proj": {
2029
+ "group_size": 32,
2030
+ "bits": 4
2031
+ },
2032
+ "model.layers.14.self_attn.q_proj": {
2033
+ "group_size": 32,
2034
+ "bits": 4
2035
+ },
2036
+ "model.layers.14.self_attn.k_proj": {
2037
+ "group_size": 32,
2038
+ "bits": 4
2039
+ },
2040
+ "model.layers.14.self_attn.v_proj": {
2041
+ "group_size": 32,
2042
+ "bits": 6
2043
+ },
2044
+ "model.layers.14.self_attn.o_proj": {
2045
+ "group_size": 32,
2046
+ "bits": 4
2047
+ },
2048
+ "model.layers.14.mlp.gate": {
2049
+ "group_size": 32,
2050
+ "bits": 4
2051
+ },
2052
+ "model.layers.14.mlp.switch_mlp.gate_proj": {
2053
+ "group_size": 32,
2054
+ "bits": 4
2055
+ },
2056
+ "model.layers.14.mlp.switch_mlp.up_proj": {
2057
+ "group_size": 32,
2058
+ "bits": 4
2059
+ },
2060
+ "model.layers.14.mlp.switch_mlp.down_proj": {
2061
+ "group_size": 32,
2062
+ "bits": 6
2063
+ },
2064
+ "model.layers.15.self_attn.q_proj": {
2065
+ "group_size": 32,
2066
+ "bits": 4
2067
+ },
2068
+ "model.layers.15.self_attn.k_proj": {
2069
+ "group_size": 32,
2070
+ "bits": 4
2071
+ },
2072
+ "model.layers.15.self_attn.v_proj": {
2073
+ "group_size": 32,
2074
+ "bits": 6
2075
+ },
2076
+ "model.layers.15.self_attn.o_proj": {
2077
+ "group_size": 32,
2078
+ "bits": 4
2079
+ },
2080
+ "model.layers.15.mlp.gate": {
2081
+ "group_size": 32,
2082
+ "bits": 4
2083
+ },
2084
+ "model.layers.15.mlp.switch_mlp.gate_proj": {
2085
+ "group_size": 32,
2086
+ "bits": 4
2087
+ },
2088
+ "model.layers.15.mlp.switch_mlp.up_proj": {
2089
+ "group_size": 32,
2090
+ "bits": 4
2091
+ },
2092
+ "model.layers.15.mlp.switch_mlp.down_proj": {
2093
+ "group_size": 32,
2094
+ "bits": 4
2095
+ },
2096
+ "model.layers.16.self_attn.q_proj": {
2097
+ "group_size": 32,
2098
+ "bits": 4
2099
+ },
2100
+ "model.layers.16.self_attn.k_proj": {
2101
+ "group_size": 32,
2102
+ "bits": 4
2103
+ },
2104
+ "model.layers.16.self_attn.v_proj": {
2105
+ "group_size": 32,
2106
+ "bits": 6
2107
+ },
2108
+ "model.layers.16.self_attn.o_proj": {
2109
+ "group_size": 32,
2110
+ "bits": 4
2111
+ },
2112
+ "model.layers.16.mlp.gate": {
2113
+ "group_size": 32,
2114
+ "bits": 4
2115
+ },
2116
+ "model.layers.16.mlp.switch_mlp.gate_proj": {
2117
+ "group_size": 32,
2118
+ "bits": 4
2119
+ },
2120
+ "model.layers.16.mlp.switch_mlp.up_proj": {
2121
+ "group_size": 32,
2122
+ "bits": 4
2123
+ },
2124
+ "model.layers.16.mlp.switch_mlp.down_proj": {
2125
+ "group_size": 32,
2126
+ "bits": 4
2127
+ },
2128
+ "model.layers.17.self_attn.q_proj": {
2129
+ "group_size": 32,
2130
+ "bits": 4
2131
+ },
2132
+ "model.layers.17.self_attn.k_proj": {
2133
+ "group_size": 32,
2134
+ "bits": 4
2135
+ },
2136
+ "model.layers.17.self_attn.v_proj": {
2137
+ "group_size": 32,
2138
+ "bits": 6
2139
+ },
2140
+ "model.layers.17.self_attn.o_proj": {
2141
+ "group_size": 32,
2142
+ "bits": 4
2143
+ },
2144
+ "model.layers.17.mlp.gate": {
2145
+ "group_size": 32,
2146
+ "bits": 4
2147
+ },
2148
+ "model.layers.17.mlp.switch_mlp.gate_proj": {
2149
+ "group_size": 32,
2150
+ "bits": 4
2151
+ },
2152
+ "model.layers.17.mlp.switch_mlp.up_proj": {
2153
+ "group_size": 32,
2154
+ "bits": 4
2155
+ },
2156
+ "model.layers.17.mlp.switch_mlp.down_proj": {
2157
+ "group_size": 32,
2158
+ "bits": 6
2159
+ },
2160
+ "model.layers.18.self_attn.q_proj": {
2161
+ "group_size": 32,
2162
+ "bits": 4
2163
+ },
2164
+ "model.layers.18.self_attn.k_proj": {
2165
+ "group_size": 32,
2166
+ "bits": 4
2167
+ },
2168
+ "model.layers.18.self_attn.v_proj": {
2169
+ "group_size": 32,
2170
+ "bits": 6
2171
+ },
2172
+ "model.layers.18.self_attn.o_proj": {
2173
+ "group_size": 32,
2174
+ "bits": 4
2175
+ },
2176
+ "model.layers.18.mlp.gate": {
2177
+ "group_size": 32,
2178
+ "bits": 4
2179
+ },
2180
+ "model.layers.18.mlp.switch_mlp.gate_proj": {
2181
+ "group_size": 32,
2182
+ "bits": 4
2183
+ },
2184
+ "model.layers.18.mlp.switch_mlp.up_proj": {
2185
+ "group_size": 32,
2186
+ "bits": 4
2187
+ },
2188
+ "model.layers.18.mlp.switch_mlp.down_proj": {
2189
+ "group_size": 32,
2190
+ "bits": 4
2191
+ },
2192
+ "model.layers.19.self_attn.q_proj": {
2193
+ "group_size": 32,
2194
+ "bits": 4
2195
+ },
2196
+ "model.layers.19.self_attn.k_proj": {
2197
+ "group_size": 32,
2198
+ "bits": 4
2199
+ },
2200
+ "model.layers.19.self_attn.v_proj": {
2201
+ "group_size": 32,
2202
+ "bits": 6
2203
+ },
2204
+ "model.layers.19.self_attn.o_proj": {
2205
+ "group_size": 32,
2206
+ "bits": 4
2207
+ },
2208
+ "model.layers.19.mlp.gate": {
2209
+ "group_size": 32,
2210
+ "bits": 4
2211
+ },
2212
+ "model.layers.19.mlp.switch_mlp.gate_proj": {
2213
+ "group_size": 32,
2214
+ "bits": 4
2215
+ },
2216
+ "model.layers.19.mlp.switch_mlp.up_proj": {
2217
+ "group_size": 32,
2218
+ "bits": 4
2219
+ },
2220
+ "model.layers.19.mlp.switch_mlp.down_proj": {
2221
+ "group_size": 32,
2222
+ "bits": 4
2223
+ },
2224
+ "model.layers.20.self_attn.q_proj": {
2225
+ "group_size": 32,
2226
+ "bits": 4
2227
+ },
2228
+ "model.layers.20.self_attn.k_proj": {
2229
+ "group_size": 32,
2230
+ "bits": 4
2231
+ },
2232
+ "model.layers.20.self_attn.v_proj": {
2233
+ "group_size": 32,
2234
+ "bits": 6
2235
+ },
2236
+ "model.layers.20.self_attn.o_proj": {
2237
+ "group_size": 32,
2238
+ "bits": 4
2239
+ },
2240
+ "model.layers.20.mlp.gate": {
2241
+ "group_size": 32,
2242
+ "bits": 4
2243
+ },
2244
+ "model.layers.20.mlp.switch_mlp.gate_proj": {
2245
+ "group_size": 32,
2246
+ "bits": 4
2247
+ },
2248
+ "model.layers.20.mlp.switch_mlp.up_proj": {
2249
+ "group_size": 32,
2250
+ "bits": 4
2251
+ },
2252
+ "model.layers.20.mlp.switch_mlp.down_proj": {
2253
+ "group_size": 32,
2254
+ "bits": 6
2255
+ },
2256
+ "model.layers.21.self_attn.q_proj": {
2257
+ "group_size": 32,
2258
+ "bits": 4
2259
+ },
2260
+ "model.layers.21.self_attn.k_proj": {
2261
+ "group_size": 32,
2262
+ "bits": 4
2263
+ },
2264
+ "model.layers.21.self_attn.v_proj": {
2265
+ "group_size": 32,
2266
+ "bits": 6
2267
+ },
2268
+ "model.layers.21.self_attn.o_proj": {
2269
+ "group_size": 32,
2270
+ "bits": 4
2271
+ },
2272
+ "model.layers.21.mlp.gate": {
2273
+ "group_size": 32,
2274
+ "bits": 4
2275
+ },
2276
+ "model.layers.21.mlp.switch_mlp.gate_proj": {
2277
+ "group_size": 32,
2278
+ "bits": 4
2279
+ },
2280
+ "model.layers.21.mlp.switch_mlp.up_proj": {
2281
+ "group_size": 32,
2282
+ "bits": 4
2283
+ },
2284
+ "model.layers.21.mlp.switch_mlp.down_proj": {
2285
+ "group_size": 32,
2286
+ "bits": 4
2287
+ },
2288
+ "model.layers.22.self_attn.q_proj": {
2289
+ "group_size": 32,
2290
+ "bits": 4
2291
+ },
2292
+ "model.layers.22.self_attn.k_proj": {
2293
+ "group_size": 32,
2294
+ "bits": 4
2295
+ },
2296
+ "model.layers.22.self_attn.v_proj": {
2297
+ "group_size": 32,
2298
+ "bits": 6
2299
+ },
2300
+ "model.layers.22.self_attn.o_proj": {
2301
+ "group_size": 32,
2302
+ "bits": 4
2303
+ },
2304
+ "model.layers.22.mlp.gate": {
2305
+ "group_size": 32,
2306
+ "bits": 4
2307
+ },
2308
+ "model.layers.22.mlp.switch_mlp.gate_proj": {
2309
+ "group_size": 32,
2310
+ "bits": 4
2311
+ },
2312
+ "model.layers.22.mlp.switch_mlp.up_proj": {
2313
+ "group_size": 32,
2314
+ "bits": 4
2315
+ },
2316
+ "model.layers.22.mlp.switch_mlp.down_proj": {
2317
+ "group_size": 32,
2318
+ "bits": 4
2319
+ },
2320
+ "model.layers.23.self_attn.q_proj": {
2321
+ "group_size": 32,
2322
+ "bits": 4
2323
+ },
2324
+ "model.layers.23.self_attn.k_proj": {
2325
+ "group_size": 32,
2326
+ "bits": 4
2327
+ },
2328
+ "model.layers.23.self_attn.v_proj": {
2329
+ "group_size": 32,
2330
+ "bits": 6
2331
+ },
2332
+ "model.layers.23.self_attn.o_proj": {
2333
+ "group_size": 32,
2334
+ "bits": 4
2335
+ },
2336
+ "model.layers.23.mlp.gate": {
2337
+ "group_size": 32,
2338
+ "bits": 4
2339
+ },
2340
+ "model.layers.23.mlp.switch_mlp.gate_proj": {
2341
+ "group_size": 32,
2342
+ "bits": 4
2343
+ },
2344
+ "model.layers.23.mlp.switch_mlp.up_proj": {
2345
+ "group_size": 32,
2346
+ "bits": 4
2347
+ },
2348
+ "model.layers.23.mlp.switch_mlp.down_proj": {
2349
+ "group_size": 32,
2350
+ "bits": 6
2351
+ },
2352
+ "model.layers.24.self_attn.q_proj": {
2353
+ "group_size": 32,
2354
+ "bits": 4
2355
+ },
2356
+ "model.layers.24.self_attn.k_proj": {
2357
+ "group_size": 32,
2358
+ "bits": 4
2359
+ },
2360
+ "model.layers.24.self_attn.v_proj": {
2361
+ "group_size": 32,
2362
+ "bits": 6
2363
+ },
2364
+ "model.layers.24.self_attn.o_proj": {
2365
+ "group_size": 32,
2366
+ "bits": 4
2367
+ },
2368
+ "model.layers.24.mlp.gate": {
2369
+ "group_size": 32,
2370
+ "bits": 4
2371
+ },
2372
+ "model.layers.24.mlp.switch_mlp.gate_proj": {
2373
+ "group_size": 32,
2374
+ "bits": 4
2375
+ },
2376
+ "model.layers.24.mlp.switch_mlp.up_proj": {
2377
+ "group_size": 32,
2378
+ "bits": 4
2379
+ },
2380
+ "model.layers.24.mlp.switch_mlp.down_proj": {
2381
+ "group_size": 32,
2382
+ "bits": 4
2383
+ },
2384
+ "model.layers.25.self_attn.q_proj": {
2385
+ "group_size": 32,
2386
+ "bits": 4
2387
+ },
2388
+ "model.layers.25.self_attn.k_proj": {
2389
+ "group_size": 32,
2390
+ "bits": 4
2391
+ },
2392
+ "model.layers.25.self_attn.v_proj": {
2393
+ "group_size": 32,
2394
+ "bits": 6
2395
+ },
2396
+ "model.layers.25.self_attn.o_proj": {
2397
+ "group_size": 32,
2398
+ "bits": 4
2399
+ },
2400
+ "model.layers.25.mlp.gate": {
2401
+ "group_size": 32,
2402
+ "bits": 4
2403
+ },
2404
+ "model.layers.25.mlp.switch_mlp.gate_proj": {
2405
+ "group_size": 32,
2406
+ "bits": 4
2407
+ },
2408
+ "model.layers.25.mlp.switch_mlp.up_proj": {
2409
+ "group_size": 32,
2410
+ "bits": 4
2411
+ },
2412
+ "model.layers.25.mlp.switch_mlp.down_proj": {
2413
+ "group_size": 32,
2414
+ "bits": 4
2415
+ },
2416
+ "model.layers.26.self_attn.q_proj": {
2417
+ "group_size": 32,
2418
+ "bits": 4
2419
+ },
2420
+ "model.layers.26.self_attn.k_proj": {
2421
+ "group_size": 32,
2422
+ "bits": 4
2423
+ },
2424
+ "model.layers.26.self_attn.v_proj": {
2425
+ "group_size": 32,
2426
+ "bits": 6
2427
+ },
2428
+ "model.layers.26.self_attn.o_proj": {
2429
+ "group_size": 32,
2430
+ "bits": 4
2431
+ },
2432
+ "model.layers.26.mlp.gate": {
2433
+ "group_size": 32,
2434
+ "bits": 4
2435
+ },
2436
+ "model.layers.26.mlp.switch_mlp.gate_proj": {
2437
+ "group_size": 32,
2438
+ "bits": 4
2439
+ },
2440
+ "model.layers.26.mlp.switch_mlp.up_proj": {
2441
+ "group_size": 32,
2442
+ "bits": 4
2443
+ },
2444
+ "model.layers.26.mlp.switch_mlp.down_proj": {
2445
+ "group_size": 32,
2446
+ "bits": 6
2447
+ },
2448
+ "model.layers.27.self_attn.q_proj": {
2449
+ "group_size": 32,
2450
+ "bits": 4
2451
+ },
2452
+ "model.layers.27.self_attn.k_proj": {
2453
+ "group_size": 32,
2454
+ "bits": 4
2455
+ },
2456
+ "model.layers.27.self_attn.v_proj": {
2457
+ "group_size": 32,
2458
+ "bits": 6
2459
+ },
2460
+ "model.layers.27.self_attn.o_proj": {
2461
+ "group_size": 32,
2462
+ "bits": 4
2463
+ },
2464
+ "model.layers.27.mlp.gate": {
2465
+ "group_size": 32,
2466
+ "bits": 4
2467
+ },
2468
+ "model.layers.27.mlp.switch_mlp.gate_proj": {
2469
+ "group_size": 32,
2470
+ "bits": 4
2471
+ },
2472
+ "model.layers.27.mlp.switch_mlp.up_proj": {
2473
+ "group_size": 32,
2474
+ "bits": 4
2475
+ },
2476
+ "model.layers.27.mlp.switch_mlp.down_proj": {
2477
+ "group_size": 32,
2478
+ "bits": 4
2479
+ },
2480
+ "model.layers.28.self_attn.q_proj": {
2481
+ "group_size": 32,
2482
+ "bits": 4
2483
+ },
2484
+ "model.layers.28.self_attn.k_proj": {
2485
+ "group_size": 32,
2486
+ "bits": 4
2487
+ },
2488
+ "model.layers.28.self_attn.v_proj": {
2489
+ "group_size": 32,
2490
+ "bits": 6
2491
+ },
2492
+ "model.layers.28.self_attn.o_proj": {
2493
+ "group_size": 32,
2494
+ "bits": 4
2495
+ },
2496
+ "model.layers.28.mlp.gate": {
2497
+ "group_size": 32,
2498
+ "bits": 4
2499
+ },
2500
+ "model.layers.28.mlp.switch_mlp.gate_proj": {
2501
+ "group_size": 32,
2502
+ "bits": 4
2503
+ },
2504
+ "model.layers.28.mlp.switch_mlp.up_proj": {
2505
+ "group_size": 32,
2506
+ "bits": 4
2507
+ },
2508
+ "model.layers.28.mlp.switch_mlp.down_proj": {
2509
+ "group_size": 32,
2510
+ "bits": 4
2511
+ },
2512
+ "model.layers.29.self_attn.q_proj": {
2513
+ "group_size": 32,
2514
+ "bits": 4
2515
+ },
2516
+ "model.layers.29.self_attn.k_proj": {
2517
+ "group_size": 32,
2518
+ "bits": 4
2519
+ },
2520
+ "model.layers.29.self_attn.v_proj": {
2521
+ "group_size": 32,
2522
+ "bits": 6
2523
+ },
2524
+ "model.layers.29.self_attn.o_proj": {
2525
+ "group_size": 32,
2526
+ "bits": 4
2527
+ },
2528
+ "model.layers.29.mlp.gate": {
2529
+ "group_size": 32,
2530
+ "bits": 4
2531
+ },
2532
+ "model.layers.29.mlp.switch_mlp.gate_proj": {
2533
+ "group_size": 32,
2534
+ "bits": 4
2535
+ },
2536
+ "model.layers.29.mlp.switch_mlp.up_proj": {
2537
+ "group_size": 32,
2538
+ "bits": 4
2539
+ },
2540
+ "model.layers.29.mlp.switch_mlp.down_proj": {
2541
+ "group_size": 32,
2542
+ "bits": 6
2543
+ },
2544
+ "model.layers.30.self_attn.q_proj": {
2545
+ "group_size": 32,
2546
+ "bits": 4
2547
+ },
2548
+ "model.layers.30.self_attn.k_proj": {
2549
+ "group_size": 32,
2550
+ "bits": 4
2551
+ },
2552
+ "model.layers.30.self_attn.v_proj": {
2553
+ "group_size": 32,
2554
+ "bits": 6
2555
+ },
2556
+ "model.layers.30.self_attn.o_proj": {
2557
+ "group_size": 32,
2558
+ "bits": 4
2559
+ },
2560
+ "model.layers.30.mlp.gate": {
2561
+ "group_size": 32,
2562
+ "bits": 4
2563
+ },
2564
+ "model.layers.30.mlp.switch_mlp.gate_proj": {
2565
+ "group_size": 32,
2566
+ "bits": 4
2567
+ },
2568
+ "model.layers.30.mlp.switch_mlp.up_proj": {
2569
+ "group_size": 32,
2570
+ "bits": 4
2571
+ },
2572
+ "model.layers.30.mlp.switch_mlp.down_proj": {
2573
+ "group_size": 32,
2574
+ "bits": 4
2575
+ },
2576
+ "model.layers.31.self_attn.q_proj": {
2577
+ "group_size": 32,
2578
+ "bits": 4
2579
+ },
2580
+ "model.layers.31.self_attn.k_proj": {
2581
+ "group_size": 32,
2582
+ "bits": 4
2583
+ },
2584
+ "model.layers.31.self_attn.v_proj": {
2585
+ "group_size": 32,
2586
+ "bits": 6
2587
+ },
2588
+ "model.layers.31.self_attn.o_proj": {
2589
+ "group_size": 32,
2590
+ "bits": 4
2591
+ },
2592
+ "model.layers.31.mlp.gate": {
2593
+ "group_size": 32,
2594
+ "bits": 4
2595
+ },
2596
+ "model.layers.31.mlp.switch_mlp.gate_proj": {
2597
+ "group_size": 32,
2598
+ "bits": 4
2599
+ },
2600
+ "model.layers.31.mlp.switch_mlp.up_proj": {
2601
+ "group_size": 32,
2602
+ "bits": 4
2603
+ },
2604
+ "model.layers.31.mlp.switch_mlp.down_proj": {
2605
+ "group_size": 32,
2606
+ "bits": 4
2607
+ },
2608
+ "model.layers.32.self_attn.q_proj": {
2609
+ "group_size": 32,
2610
+ "bits": 4
2611
+ },
2612
+ "model.layers.32.self_attn.k_proj": {
2613
+ "group_size": 32,
2614
+ "bits": 4
2615
+ },
2616
+ "model.layers.32.self_attn.v_proj": {
2617
+ "group_size": 32,
2618
+ "bits": 6
2619
+ },
2620
+ "model.layers.32.self_attn.o_proj": {
2621
+ "group_size": 32,
2622
+ "bits": 4
2623
+ },
2624
+ "model.layers.32.mlp.gate": {
2625
+ "group_size": 32,
2626
+ "bits": 4
2627
+ },
2628
+ "model.layers.32.mlp.switch_mlp.gate_proj": {
2629
+ "group_size": 32,
2630
+ "bits": 4
2631
+ },
2632
+ "model.layers.32.mlp.switch_mlp.up_proj": {
2633
+ "group_size": 32,
2634
+ "bits": 4
2635
+ },
2636
+ "model.layers.32.mlp.switch_mlp.down_proj": {
2637
+ "group_size": 32,
2638
+ "bits": 6
2639
+ },
2640
+ "model.layers.33.self_attn.q_proj": {
2641
+ "group_size": 32,
2642
+ "bits": 4
2643
+ },
2644
+ "model.layers.33.self_attn.k_proj": {
2645
+ "group_size": 32,
2646
+ "bits": 4
2647
+ },
2648
+ "model.layers.33.self_attn.v_proj": {
2649
+ "group_size": 32,
2650
+ "bits": 6
2651
+ },
2652
+ "model.layers.33.self_attn.o_proj": {
2653
+ "group_size": 32,
2654
+ "bits": 4
2655
+ },
2656
+ "model.layers.33.mlp.gate": {
2657
+ "group_size": 32,
2658
+ "bits": 4
2659
+ },
2660
+ "model.layers.33.mlp.switch_mlp.gate_proj": {
2661
+ "group_size": 32,
2662
+ "bits": 4
2663
+ },
2664
+ "model.layers.33.mlp.switch_mlp.up_proj": {
2665
+ "group_size": 32,
2666
+ "bits": 4
2667
+ },
2668
+ "model.layers.33.mlp.switch_mlp.down_proj": {
2669
+ "group_size": 32,
2670
+ "bits": 4
2671
+ },
2672
+ "model.layers.34.self_attn.q_proj": {
2673
+ "group_size": 32,
2674
+ "bits": 4
2675
+ },
2676
+ "model.layers.34.self_attn.k_proj": {
2677
+ "group_size": 32,
2678
+ "bits": 4
2679
+ },
2680
+ "model.layers.34.self_attn.v_proj": {
2681
+ "group_size": 32,
2682
+ "bits": 6
2683
+ },
2684
+ "model.layers.34.self_attn.o_proj": {
2685
+ "group_size": 32,
2686
+ "bits": 4
2687
+ },
2688
+ "model.layers.34.mlp.gate": {
2689
+ "group_size": 32,
2690
+ "bits": 4
2691
+ },
2692
+ "model.layers.34.mlp.switch_mlp.gate_proj": {
2693
+ "group_size": 32,
2694
+ "bits": 4
2695
+ },
2696
+ "model.layers.34.mlp.switch_mlp.up_proj": {
2697
+ "group_size": 32,
2698
+ "bits": 4
2699
+ },
2700
+ "model.layers.34.mlp.switch_mlp.down_proj": {
2701
+ "group_size": 32,
2702
+ "bits": 4
2703
+ },
2704
+ "model.layers.35.self_attn.q_proj": {
2705
+ "group_size": 32,
2706
+ "bits": 4
2707
+ },
2708
+ "model.layers.35.self_attn.k_proj": {
2709
+ "group_size": 32,
2710
+ "bits": 4
2711
+ },
2712
+ "model.layers.35.self_attn.v_proj": {
2713
+ "group_size": 32,
2714
+ "bits": 6
2715
+ },
2716
+ "model.layers.35.self_attn.o_proj": {
2717
+ "group_size": 32,
2718
+ "bits": 4
2719
+ },
2720
+ "model.layers.35.mlp.gate": {
2721
+ "group_size": 32,
2722
+ "bits": 4
2723
+ },
2724
+ "model.layers.35.mlp.switch_mlp.gate_proj": {
2725
+ "group_size": 32,
2726
+ "bits": 4
2727
+ },
2728
+ "model.layers.35.mlp.switch_mlp.up_proj": {
2729
+ "group_size": 32,
2730
+ "bits": 4
2731
+ },
2732
+ "model.layers.35.mlp.switch_mlp.down_proj": {
2733
+ "group_size": 32,
2734
+ "bits": 6
2735
+ },
2736
+ "model.layers.36.self_attn.q_proj": {
2737
+ "group_size": 32,
2738
+ "bits": 4
2739
+ },
2740
+ "model.layers.36.self_attn.k_proj": {
2741
+ "group_size": 32,
2742
+ "bits": 4
2743
+ },
2744
+ "model.layers.36.self_attn.v_proj": {
2745
+ "group_size": 32,
2746
+ "bits": 6
2747
+ },
2748
+ "model.layers.36.self_attn.o_proj": {
2749
+ "group_size": 32,
2750
+ "bits": 4
2751
+ },
2752
+ "model.layers.36.mlp.gate": {
2753
+ "group_size": 32,
2754
+ "bits": 4
2755
+ },
2756
+ "model.layers.36.mlp.switch_mlp.gate_proj": {
2757
+ "group_size": 32,
2758
+ "bits": 4
2759
+ },
2760
+ "model.layers.36.mlp.switch_mlp.up_proj": {
2761
+ "group_size": 32,
2762
+ "bits": 4
2763
+ },
2764
+ "model.layers.36.mlp.switch_mlp.down_proj": {
2765
+ "group_size": 32,
2766
+ "bits": 4
2767
+ },
2768
+ "model.layers.37.self_attn.q_proj": {
2769
+ "group_size": 32,
2770
+ "bits": 4
2771
+ },
2772
+ "model.layers.37.self_attn.k_proj": {
2773
+ "group_size": 32,
2774
+ "bits": 4
2775
+ },
2776
+ "model.layers.37.self_attn.v_proj": {
2777
+ "group_size": 32,
2778
+ "bits": 6
2779
+ },
2780
+ "model.layers.37.self_attn.o_proj": {
2781
+ "group_size": 32,
2782
+ "bits": 4
2783
+ },
2784
+ "model.layers.37.mlp.gate": {
2785
+ "group_size": 32,
2786
+ "bits": 4
2787
+ },
2788
+ "model.layers.37.mlp.switch_mlp.gate_proj": {
2789
+ "group_size": 32,
2790
+ "bits": 4
2791
+ },
2792
+ "model.layers.37.mlp.switch_mlp.up_proj": {
2793
+ "group_size": 32,
2794
+ "bits": 4
2795
+ },
2796
+ "model.layers.37.mlp.switch_mlp.down_proj": {
2797
+ "group_size": 32,
2798
+ "bits": 4
2799
+ },
2800
+ "model.layers.38.self_attn.q_proj": {
2801
+ "group_size": 32,
2802
+ "bits": 4
2803
+ },
2804
+ "model.layers.38.self_attn.k_proj": {
2805
+ "group_size": 32,
2806
+ "bits": 4
2807
+ },
2808
+ "model.layers.38.self_attn.v_proj": {
2809
+ "group_size": 32,
2810
+ "bits": 6
2811
+ },
2812
+ "model.layers.38.self_attn.o_proj": {
2813
+ "group_size": 32,
2814
+ "bits": 4
2815
+ },
2816
+ "model.layers.38.mlp.gate": {
2817
+ "group_size": 32,
2818
+ "bits": 4
2819
+ },
2820
+ "model.layers.38.mlp.switch_mlp.gate_proj": {
2821
+ "group_size": 32,
2822
+ "bits": 4
2823
+ },
2824
+ "model.layers.38.mlp.switch_mlp.up_proj": {
2825
+ "group_size": 32,
2826
+ "bits": 4
2827
+ },
2828
+ "model.layers.38.mlp.switch_mlp.down_proj": {
2829
+ "group_size": 32,
2830
+ "bits": 6
2831
+ },
2832
+ "model.layers.39.self_attn.q_proj": {
2833
+ "group_size": 32,
2834
+ "bits": 4
2835
+ },
2836
+ "model.layers.39.self_attn.k_proj": {
2837
+ "group_size": 32,
2838
+ "bits": 4
2839
+ },
2840
+ "model.layers.39.self_attn.v_proj": {
2841
+ "group_size": 32,
2842
+ "bits": 6
2843
+ },
2844
+ "model.layers.39.self_attn.o_proj": {
2845
+ "group_size": 32,
2846
+ "bits": 4
2847
+ },
2848
+ "model.layers.39.mlp.gate": {
2849
+ "group_size": 32,
2850
+ "bits": 4
2851
+ },
2852
+ "model.layers.39.mlp.switch_mlp.gate_proj": {
2853
+ "group_size": 32,
2854
+ "bits": 4
2855
+ },
2856
+ "model.layers.39.mlp.switch_mlp.up_proj": {
2857
+ "group_size": 32,
2858
+ "bits": 4
2859
+ },
2860
+ "model.layers.39.mlp.switch_mlp.down_proj": {
2861
+ "group_size": 32,
2862
+ "bits": 4
2863
+ },
2864
+ "model.layers.40.self_attn.q_proj": {
2865
+ "group_size": 32,
2866
+ "bits": 4
2867
+ },
2868
+ "model.layers.40.self_attn.k_proj": {
2869
+ "group_size": 32,
2870
+ "bits": 4
2871
+ },
2872
+ "model.layers.40.self_attn.v_proj": {
2873
+ "group_size": 32,
2874
+ "bits": 6
2875
+ },
2876
+ "model.layers.40.self_attn.o_proj": {
2877
+ "group_size": 32,
2878
+ "bits": 4
2879
+ },
2880
+ "model.layers.40.mlp.gate": {
2881
+ "group_size": 32,
2882
+ "bits": 4
2883
+ },
2884
+ "model.layers.40.mlp.switch_mlp.gate_proj": {
2885
+ "group_size": 32,
2886
+ "bits": 4
2887
+ },
2888
+ "model.layers.40.mlp.switch_mlp.up_proj": {
2889
+ "group_size": 32,
2890
+ "bits": 4
2891
+ },
2892
+ "model.layers.40.mlp.switch_mlp.down_proj": {
2893
+ "group_size": 32,
2894
+ "bits": 4
2895
+ },
2896
+ "model.layers.41.self_attn.q_proj": {
2897
+ "group_size": 32,
2898
+ "bits": 4
2899
+ },
2900
+ "model.layers.41.self_attn.k_proj": {
2901
+ "group_size": 32,
2902
+ "bits": 4
2903
+ },
2904
+ "model.layers.41.self_attn.v_proj": {
2905
+ "group_size": 32,
2906
+ "bits": 6
2907
+ },
2908
+ "model.layers.41.self_attn.o_proj": {
2909
+ "group_size": 32,
2910
+ "bits": 4
2911
+ },
2912
+ "model.layers.41.mlp.gate": {
2913
+ "group_size": 32,
2914
+ "bits": 4
2915
+ },
2916
+ "model.layers.41.mlp.switch_mlp.gate_proj": {
2917
+ "group_size": 32,
2918
+ "bits": 4
2919
+ },
2920
+ "model.layers.41.mlp.switch_mlp.up_proj": {
2921
+ "group_size": 32,
2922
+ "bits": 4
2923
+ },
2924
+ "model.layers.41.mlp.switch_mlp.down_proj": {
2925
+ "group_size": 32,
2926
+ "bits": 6
2927
+ },
2928
+ "model.layers.42.self_attn.q_proj": {
2929
+ "group_size": 32,
2930
+ "bits": 4
2931
+ },
2932
+ "model.layers.42.self_attn.k_proj": {
2933
+ "group_size": 32,
2934
+ "bits": 4
2935
+ },
2936
+ "model.layers.42.self_attn.v_proj": {
2937
+ "group_size": 32,
2938
+ "bits": 6
2939
+ },
2940
+ "model.layers.42.self_attn.o_proj": {
2941
+ "group_size": 32,
2942
+ "bits": 4
2943
+ },
2944
+ "model.layers.42.mlp.gate": {
2945
+ "group_size": 32,
2946
+ "bits": 4
2947
+ },
2948
+ "model.layers.42.mlp.switch_mlp.gate_proj": {
2949
+ "group_size": 32,
2950
+ "bits": 4
2951
+ },
2952
+ "model.layers.42.mlp.switch_mlp.up_proj": {
2953
+ "group_size": 32,
2954
+ "bits": 4
2955
+ },
2956
+ "model.layers.42.mlp.switch_mlp.down_proj": {
2957
+ "group_size": 32,
2958
+ "bits": 6
2959
+ },
2960
+ "model.layers.43.self_attn.q_proj": {
2961
+ "group_size": 32,
2962
+ "bits": 4
2963
+ },
2964
+ "model.layers.43.self_attn.k_proj": {
2965
+ "group_size": 32,
2966
+ "bits": 4
2967
+ },
2968
+ "model.layers.43.self_attn.v_proj": {
2969
+ "group_size": 32,
2970
+ "bits": 6
2971
+ },
2972
+ "model.layers.43.self_attn.o_proj": {
2973
+ "group_size": 32,
2974
+ "bits": 4
2975
+ },
2976
+ "model.layers.43.mlp.gate": {
2977
+ "group_size": 32,
2978
+ "bits": 4
2979
+ },
2980
+ "model.layers.43.mlp.switch_mlp.gate_proj": {
2981
+ "group_size": 32,
2982
+ "bits": 4
2983
+ },
2984
+ "model.layers.43.mlp.switch_mlp.up_proj": {
2985
+ "group_size": 32,
2986
+ "bits": 4
2987
+ },
2988
+ "model.layers.43.mlp.switch_mlp.down_proj": {
2989
+ "group_size": 32,
2990
+ "bits": 6
2991
+ },
2992
+ "model.layers.44.self_attn.q_proj": {
2993
+ "group_size": 32,
2994
+ "bits": 4
2995
+ },
2996
+ "model.layers.44.self_attn.k_proj": {
2997
+ "group_size": 32,
2998
+ "bits": 4
2999
+ },
3000
+ "model.layers.44.self_attn.v_proj": {
3001
+ "group_size": 32,
3002
+ "bits": 6
3003
+ },
3004
+ "model.layers.44.self_attn.o_proj": {
3005
+ "group_size": 32,
3006
+ "bits": 4
3007
+ },
3008
+ "model.layers.44.mlp.gate": {
3009
+ "group_size": 32,
3010
+ "bits": 4
3011
+ },
3012
+ "model.layers.44.mlp.switch_mlp.gate_proj": {
3013
+ "group_size": 32,
3014
+ "bits": 4
3015
+ },
3016
+ "model.layers.44.mlp.switch_mlp.up_proj": {
3017
+ "group_size": 32,
3018
+ "bits": 4
3019
+ },
3020
+ "model.layers.44.mlp.switch_mlp.down_proj": {
3021
+ "group_size": 32,
3022
+ "bits": 6
3023
+ },
3024
+ "model.layers.45.self_attn.q_proj": {
3025
+ "group_size": 32,
3026
+ "bits": 4
3027
+ },
3028
+ "model.layers.45.self_attn.k_proj": {
3029
+ "group_size": 32,
3030
+ "bits": 4
3031
+ },
3032
+ "model.layers.45.self_attn.v_proj": {
3033
+ "group_size": 32,
3034
+ "bits": 6
3035
+ },
3036
+ "model.layers.45.self_attn.o_proj": {
3037
+ "group_size": 32,
3038
+ "bits": 4
3039
+ },
3040
+ "model.layers.45.mlp.gate": {
3041
+ "group_size": 32,
3042
+ "bits": 4
3043
+ },
3044
+ "model.layers.45.mlp.switch_mlp.gate_proj": {
3045
+ "group_size": 32,
3046
+ "bits": 4
3047
+ },
3048
+ "model.layers.45.mlp.switch_mlp.up_proj": {
3049
+ "group_size": 32,
3050
+ "bits": 4
3051
+ },
3052
+ "model.layers.45.mlp.switch_mlp.down_proj": {
3053
+ "group_size": 32,
3054
+ "bits": 6
3055
+ },
3056
+ "model.layers.46.self_attn.q_proj": {
3057
+ "group_size": 32,
3058
+ "bits": 4
3059
+ },
3060
+ "model.layers.46.self_attn.k_proj": {
3061
+ "group_size": 32,
3062
+ "bits": 4
3063
+ },
3064
+ "model.layers.46.self_attn.v_proj": {
3065
+ "group_size": 32,
3066
+ "bits": 6
3067
+ },
3068
+ "model.layers.46.self_attn.o_proj": {
3069
+ "group_size": 32,
3070
+ "bits": 4
3071
+ },
3072
+ "model.layers.46.mlp.gate": {
3073
+ "group_size": 32,
3074
+ "bits": 4
3075
+ },
3076
+ "model.layers.46.mlp.switch_mlp.gate_proj": {
3077
+ "group_size": 32,
3078
+ "bits": 4
3079
+ },
3080
+ "model.layers.46.mlp.switch_mlp.up_proj": {
3081
+ "group_size": 32,
3082
+ "bits": 4
3083
+ },
3084
+ "model.layers.46.mlp.switch_mlp.down_proj": {
3085
+ "group_size": 32,
3086
+ "bits": 6
3087
+ },
3088
+ "model.layers.47.self_attn.q_proj": {
3089
+ "group_size": 32,
3090
+ "bits": 4
3091
+ },
3092
+ "model.layers.47.self_attn.k_proj": {
3093
+ "group_size": 32,
3094
+ "bits": 4
3095
+ },
3096
+ "model.layers.47.self_attn.v_proj": {
3097
+ "group_size": 32,
3098
+ "bits": 6
3099
+ },
3100
+ "model.layers.47.self_attn.o_proj": {
3101
+ "group_size": 32,
3102
+ "bits": 4
3103
+ },
3104
+ "model.layers.47.mlp.gate": {
3105
+ "group_size": 32,
3106
+ "bits": 4
3107
+ },
3108
+ "model.layers.47.mlp.switch_mlp.gate_proj": {
3109
+ "group_size": 32,
3110
+ "bits": 4
3111
+ },
3112
+ "model.layers.47.mlp.switch_mlp.up_proj": {
3113
+ "group_size": 32,
3114
+ "bits": 4
3115
+ },
3116
+ "model.layers.47.mlp.switch_mlp.down_proj": {
3117
+ "group_size": 32,
3118
+ "bits": 6
3119
+ },
3120
+ "lm_head": {
3121
+ "group_size": 32,
3122
+ "bits": 6
3123
+ }
3124
+ },
3125
+ "rms_norm_eps": 1e-06,
3126
+ "rope_scaling": null,
3127
+ "rope_theta": 10000000,
3128
+ "router_aux_loss_coef": 0.0,
3129
+ "shared_expert_intermediate_size": 0,
3130
+ "sliding_window": null,
3131
+ "tie_word_embeddings": false,
3132
+ "torch_dtype": "bfloat16",
3133
+ "transformers_version": "4.55.0",
3134
+ "use_cache": true,
3135
+ "use_qk_norm": true,
3136
+ "use_sliding_window": false,
3137
+ "vocab_size": 151936
3138
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "repetition_penalty": 1.05,
9
+ "temperature": 0.7,
10
+ "top_k": 20,
11
+ "top_p": 0.8,
12
+ "transformers_version": "4.55.0"
13
+ }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model-00001-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d9f91e7448a5d2eceb056fa0b229864d74e188303d231e200cc9564c6073991
3
+ size 5363677484
model-00002-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b2c0216736906a7a1c92a7320944301309174e3d996b39aa59c4cbf43cefbbfe
3
+ size 5290347597
model-00003-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12c5f194ca9fcc12ce778b7f0afcd444b0cc43fc664255c5ba2f84f88833995d
3
+ size 5265944369
model-00004-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348826aeb9b7b4e995755b5418a6d045a85712509966f313502a1b8b19b08e0a
3
+ size 770491573
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3coder_tool_parser.py ADDED
@@ -0,0 +1,689 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # SPDX-License-Identifier: Apache-2.0
2
+ # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+ import ast
4
+ import json
5
+ import uuid
6
+ from collections.abc import Sequence
7
+ from typing import Any, List, Optional, Union
8
+
9
+ import regex as re
10
+
11
+ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
12
+ ChatCompletionToolsParam,
13
+ DeltaFunctionCall, DeltaMessage,
14
+ DeltaToolCall,
15
+ ExtractedToolCallInformation,
16
+ FunctionCall, ToolCall)
17
+ from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
18
+ ToolParser, ToolParserManager)
19
+ from vllm.logger import init_logger
20
+ from vllm.transformers_utils.tokenizer import AnyTokenizer
21
+
22
+ logger = init_logger(__name__)
23
+
24
+
25
+ @ToolParserManager.register_module("qwen3_coder")
26
+ class Qwen3CoderToolParser(ToolParser):
27
+
28
+ def __init__(self, tokenizer: AnyTokenizer):
29
+ super().__init__(tokenizer)
30
+
31
+ self.current_tool_name_sent: bool = False
32
+ self.prev_tool_call_arr: list[dict] = []
33
+ self.current_tool_id: int = -1
34
+ self.streamed_args_for_tool: list[str] = []
35
+
36
+ # Sentinel tokens for streaming mode
37
+ self.tool_call_start_token: str = "<tool_call>"
38
+ self.tool_call_end_token: str = "</tool_call>"
39
+ self.tool_call_prefix: str = "<function="
40
+ self.function_end_token: str = "</function>"
41
+ self.parameter_prefix: str = "<parameter="
42
+ self.parameter_end_token: str = "</parameter>"
43
+ self.is_tool_call_started: bool = False
44
+ self.failed_count: int = 0
45
+
46
+ # Enhanced streaming state - reset for each new message
47
+ self._reset_streaming_state()
48
+
49
+ # Regex patterns
50
+ self.tool_call_complete_regex = re.compile(
51
+ r"<tool_call>(.*?)</tool_call>", re.DOTALL)
52
+ self.tool_call_regex = re.compile(
53
+ r"<tool_call>(.*?)</tool_call>|<tool_call>(.*?)$", re.DOTALL)
54
+ self.tool_call_function_regex = re.compile(
55
+ r"<function=(.*?)</function>|<function=(.*)$", re.DOTALL)
56
+ self.tool_call_parameter_regex = re.compile(
57
+ r"<parameter=(.*?)(?:</parameter>|(?=<parameter=)|(?=</function>)|$)",
58
+ re.DOTALL)
59
+
60
+ if not self.model_tokenizer:
61
+ raise ValueError(
62
+ "The model tokenizer must be passed to the ToolParser "
63
+ "constructor during construction.")
64
+
65
+ self.tool_call_start_token_id = self.vocab.get(
66
+ self.tool_call_start_token)
67
+ self.tool_call_end_token_id = self.vocab.get(self.tool_call_end_token)
68
+
69
+ if self.tool_call_start_token_id is None or self.tool_call_end_token_id is None:
70
+ raise RuntimeError(
71
+ "Qwen3 XML Tool parser could not locate tool call start/end "
72
+ "tokens in the tokenizer!")
73
+
74
+ logger.info(
75
+ f"vLLM Successfully import tool parser {self.__class__.__name__} !"
76
+ )
77
+
78
+ def _generate_tool_call_id(self) -> str:
79
+ """Generate a unique tool call ID."""
80
+ return f"call_{uuid.uuid4().hex[:24]}"
81
+
82
+ def _reset_streaming_state(self):
83
+ """Reset all streaming state."""
84
+ self.current_tool_index = 0
85
+ self.is_tool_call_started = False
86
+ self.header_sent = False
87
+ self.current_tool_id = None
88
+ self.current_function_name = None
89
+ self.current_param_name = None
90
+ self.current_param_value = ""
91
+ self.param_count = 0
92
+ self.in_param = False
93
+ self.in_function = False
94
+ self.accumulated_text = ""
95
+ self.json_started = False
96
+ self.json_closed = False
97
+ # Store accumulated parameters for type conversion
98
+ self.accumulated_params = {}
99
+ self.streaming_request = None
100
+
101
+ def _get_arguments_config(
102
+ self, func_name: str,
103
+ tools: Optional[list[ChatCompletionToolsParam]]) -> dict:
104
+ """Extract argument configuration for a function."""
105
+ if tools is None:
106
+ return {}
107
+ for config in tools:
108
+ if not hasattr(config, "type") or not (hasattr(
109
+ config, "function") and hasattr(config.function, "name")):
110
+ continue
111
+ if config.type == "function" and config.function.name == func_name:
112
+ if not hasattr(config.function, "parameters"):
113
+ return {}
114
+ params = config.function.parameters
115
+ if isinstance(params, dict) and "properties" in params:
116
+ return params["properties"]
117
+ elif isinstance(params, dict):
118
+ return params
119
+ else:
120
+ return {}
121
+ logger.warning(f"Tool '{func_name}' is not defined in the tools list.")
122
+ return {}
123
+
124
+ def _convert_param_value(self, param_value: str, param_name: str,
125
+ param_config: dict, func_name: str) -> Any:
126
+ """Convert parameter value based on its type in the schema."""
127
+ # Handle null value for any type
128
+ if param_value.lower() == "null":
129
+ return None
130
+
131
+ if param_name not in param_config:
132
+ if param_config != {}:
133
+ logger.warning(
134
+ f"Parsed parameter '{param_name}' is not defined in the tool "
135
+ f"parameters for tool '{func_name}', directly returning the string value."
136
+ )
137
+ return param_value
138
+
139
+ if isinstance(param_config[param_name],
140
+ dict) and "type" in param_config[param_name]:
141
+ param_type = str(param_config[param_name]["type"]).strip().lower()
142
+ else:
143
+ param_type = "string"
144
+ if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
145
+ return param_value
146
+ elif param_type.startswith("int") or param_type.startswith(
147
+ "uint") or param_type.startswith(
148
+ "long") or param_type.startswith(
149
+ "short") or param_type.startswith("unsigned"):
150
+ try:
151
+ param_value = int(param_value)
152
+ except:
153
+ logger.warning(
154
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not an integer in tool "
155
+ f"'{func_name}', degenerating to string.")
156
+ return param_value
157
+ elif param_type.startswith("num") or param_type.startswith("float"):
158
+ try:
159
+ float_param_value = float(param_value)
160
+ param_value = float_param_value if float_param_value - int(
161
+ float_param_value) != 0 else int(float_param_value)
162
+ except:
163
+ logger.warning(
164
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not a float in tool "
165
+ f"'{func_name}', degenerating to string.")
166
+ return param_value
167
+ elif param_type in ["boolean", "bool", "binary"]:
168
+ param_value = param_value.lower()
169
+ if param_value not in ["true", "false"]:
170
+ logger.warning(
171
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not a boolean (`true` of `false`) in tool '{func_name}', degenerating to false."
172
+ )
173
+ return param_value == "true"
174
+ else:
175
+ if param_type in ["object", "array", "arr"
176
+ ] or param_type.startswith(
177
+ "dict") or param_type.startswith("list"):
178
+ try:
179
+ param_value = json.loads(param_value)
180
+ return param_value
181
+ except:
182
+ logger.warning(
183
+ f"Parsed value '{param_value}' of parameter '{param_name}' cannot be parsed with json.loads in tool "
184
+ f"'{func_name}', will try other methods to parse it.")
185
+ try:
186
+ param_value = ast.literal_eval(param_value) # safer
187
+ except:
188
+ logger.warning(
189
+ f"Parsed value '{param_value}' of parameter '{param_name}' cannot be converted via Python `ast.literal_eval()` in tool '{func_name}', degenerating to string."
190
+ )
191
+ return param_value
192
+
193
+ def _parse_xml_function_call(
194
+ self, function_call_str: str,
195
+ tools: Optional[list[ChatCompletionToolsParam]]
196
+ ) -> Optional[ToolCall]:
197
+
198
+ # Extract function name
199
+ end_index = function_call_str.index(">")
200
+ function_name = function_call_str[:end_index]
201
+ param_config = self._get_arguments_config(function_name, tools)
202
+ parameters = function_call_str[end_index + 1:]
203
+ param_dict = {}
204
+ for match_text in self.tool_call_parameter_regex.findall(parameters):
205
+ idx = match_text.index(">")
206
+ param_name = match_text[:idx]
207
+ param_value = str(match_text[idx + 1:])
208
+ # Remove prefix and trailing \n
209
+ if param_value.startswith("\n"):
210
+ param_value = param_value[1:]
211
+ if param_value.endswith("\n"):
212
+ param_value = param_value[:-1]
213
+
214
+ param_dict[param_name] = self._convert_param_value(
215
+ param_value, param_name, param_config, function_name)
216
+ return ToolCall(
217
+ type="function",
218
+ function=FunctionCall(name=function_name,
219
+ arguments=json.dumps(param_dict,
220
+ ensure_ascii=False)),
221
+ )
222
+
223
+ def _get_function_calls(self, model_output: str) -> List[str]:
224
+ # Find all tool calls
225
+ matched_ranges = self.tool_call_regex.findall(model_output)
226
+ raw_tool_calls = [
227
+ match[0] if match[0] else match[1] for match in matched_ranges
228
+ ]
229
+
230
+ # Back-off strategy if no tool_call tags found
231
+ if len(raw_tool_calls) == 0:
232
+ raw_tool_calls = [model_output]
233
+
234
+ raw_function_calls = []
235
+ for tool_call in raw_tool_calls:
236
+ raw_function_calls.extend(
237
+ self.tool_call_function_regex.findall(tool_call))
238
+
239
+ function_calls = [
240
+ match[0] if match[0] else match[1] for match in raw_function_calls
241
+ ]
242
+ return function_calls
243
+
244
+ def extract_tool_calls(
245
+ self,
246
+ model_output: str,
247
+ request: ChatCompletionRequest,
248
+ ) -> ExtractedToolCallInformation:
249
+ # Quick check to avoid unnecessary processing
250
+ if self.tool_call_prefix not in model_output:
251
+ return ExtractedToolCallInformation(tools_called=False,
252
+ tool_calls=[],
253
+ content=model_output)
254
+
255
+ try:
256
+ function_calls = self._get_function_calls(model_output)
257
+ if len(function_calls) == 0:
258
+ return ExtractedToolCallInformation(tools_called=False,
259
+ tool_calls=[],
260
+ content=model_output)
261
+
262
+ tool_calls = [
263
+ self._parse_xml_function_call(function_call_str, request.tools)
264
+ for function_call_str in function_calls
265
+ ]
266
+
267
+ # Populate prev_tool_call_arr for serving layer to set finish_reason
268
+ self.prev_tool_call_arr.clear() # Clear previous calls
269
+ for tool_call in tool_calls:
270
+ if tool_call:
271
+ self.prev_tool_call_arr.append({
272
+ "name":
273
+ tool_call.function.name,
274
+ "arguments":
275
+ tool_call.function.arguments,
276
+ })
277
+
278
+ # Extract content before tool calls
279
+ content_index = model_output.find(self.tool_call_start_token)
280
+ content_index = content_index if content_index >= 0 else model_output.find(
281
+ self.tool_call_prefix)
282
+ content = model_output[:content_index] # .rstrip()
283
+
284
+ return ExtractedToolCallInformation(
285
+ tools_called=(len(tool_calls) > 0),
286
+ tool_calls=tool_calls,
287
+ content=content if content else None,
288
+ )
289
+
290
+ except Exception:
291
+ logger.exception("Error in extracting tool call from response.")
292
+ return ExtractedToolCallInformation(tools_called=False,
293
+ tool_calls=[],
294
+ content=model_output)
295
+
296
+ def extract_tool_calls_streaming(
297
+ self,
298
+ previous_text: str,
299
+ current_text: str,
300
+ delta_text: str,
301
+ previous_token_ids: Sequence[int],
302
+ current_token_ids: Sequence[int],
303
+ delta_token_ids: Sequence[int],
304
+ request: ChatCompletionRequest,
305
+ ) -> Union[DeltaMessage, None]:
306
+ # Store request for type conversion
307
+ if not previous_text:
308
+ self._reset_streaming_state()
309
+ self.streaming_request = request
310
+
311
+ # If no delta text, return None unless it's an EOS token after tool calls
312
+ if not delta_text:
313
+ # Check if this is an EOS token after all tool calls are complete
314
+ # We check for tool calls in the text even if is_tool_call_started is False
315
+ # because it might have been reset after processing all tools
316
+ if delta_token_ids and self.tool_call_end_token_id not in delta_token_ids:
317
+ # Count complete tool calls
318
+ complete_calls = len(
319
+ self.tool_call_complete_regex.findall(current_text))
320
+
321
+ # If we have completed tool calls and populated prev_tool_call_arr
322
+ if complete_calls > 0 and len(self.prev_tool_call_arr) > 0:
323
+ # Check if all tool calls are closed
324
+ open_calls = current_text.count(
325
+ self.tool_call_start_token) - current_text.count(
326
+ self.tool_call_end_token)
327
+ if open_calls == 0:
328
+ # Return empty delta message to allow finish_reason processing
329
+ return DeltaMessage(content="")
330
+ elif not self.is_tool_call_started and current_text:
331
+ # This is a regular content response that's now complete
332
+ return DeltaMessage(content="")
333
+ return None
334
+
335
+ # Update accumulated text
336
+ self.accumulated_text = current_text
337
+
338
+ # Check if we need to advance to next tool
339
+ if self.json_closed and not self.in_function:
340
+ # Check if this tool call has ended
341
+ tool_ends = current_text.count(self.tool_call_end_token)
342
+ if tool_ends > self.current_tool_index:
343
+ # This tool has ended, advance to next
344
+ self.current_tool_index += 1
345
+ self.header_sent = False
346
+ self.param_count = 0
347
+ self.json_started = False
348
+ self.json_closed = False
349
+ self.accumulated_params = {}
350
+
351
+ # Check if there are more tool calls
352
+ tool_starts = current_text.count(self.tool_call_start_token)
353
+ if self.current_tool_index >= tool_starts:
354
+ # No more tool calls
355
+ self.is_tool_call_started = False
356
+ # Continue processing next tool
357
+ return None
358
+
359
+ # Handle normal content before tool calls
360
+ if not self.is_tool_call_started:
361
+ # Check if tool call is starting
362
+ if self.tool_call_start_token_id in delta_token_ids or self.tool_call_start_token in delta_text:
363
+ self.is_tool_call_started = True
364
+ # Return any content before the tool call
365
+ if self.tool_call_start_token in delta_text:
366
+ content_before = delta_text[:delta_text.index(
367
+ self.tool_call_start_token)]
368
+ if content_before:
369
+ return DeltaMessage(content=content_before)
370
+ return None
371
+ else:
372
+ # Check if we're between tool calls - skip whitespace
373
+ if current_text.rstrip().endswith(self.tool_call_end_token):
374
+ # We just ended a tool call, skip whitespace
375
+ if delta_text.strip() == "":
376
+ return None
377
+ # Normal content, no tool call
378
+ return DeltaMessage(content=delta_text)
379
+
380
+ # Check if we're between tool calls (waiting for next one)
381
+ # Count tool calls we've seen vs processed
382
+ tool_starts_count = current_text.count(self.tool_call_start_token)
383
+ if self.current_tool_index >= tool_starts_count:
384
+ # We're past all tool calls, shouldn't be here
385
+ return None
386
+
387
+ # We're in a tool call, find the current tool call portion
388
+ # Need to find the correct tool call based on current_tool_index
389
+ tool_starts = []
390
+ idx = 0
391
+ while True:
392
+ idx = current_text.find(self.tool_call_start_token, idx)
393
+ if idx == -1:
394
+ break
395
+ tool_starts.append(idx)
396
+ idx += len(self.tool_call_start_token)
397
+
398
+ if self.current_tool_index >= len(tool_starts):
399
+ # No more tool calls to process yet
400
+ return None
401
+
402
+ tool_start_idx = tool_starts[self.current_tool_index]
403
+ # Find where this tool call ends (or current position if not ended yet)
404
+ tool_end_idx = current_text.find(self.tool_call_end_token,
405
+ tool_start_idx)
406
+ if tool_end_idx == -1:
407
+ tool_text = current_text[tool_start_idx:]
408
+ else:
409
+ tool_text = current_text[tool_start_idx:tool_end_idx +
410
+ len(self.tool_call_end_token)]
411
+
412
+ # Looking for function header
413
+ if not self.header_sent:
414
+ if self.tool_call_prefix in tool_text:
415
+ func_start = tool_text.find(self.tool_call_prefix) + len(
416
+ self.tool_call_prefix)
417
+ func_end = tool_text.find(">", func_start)
418
+
419
+ if func_end != -1:
420
+ # Found complete function name
421
+ self.current_function_name = tool_text[func_start:func_end]
422
+ self.current_tool_id = self._generate_tool_call_id()
423
+ self.header_sent = True
424
+ self.in_function = True
425
+
426
+ # IMPORTANT: Add to prev_tool_call_arr immediately when we detect a tool call
427
+ # This ensures finish_reason="tool_calls" even if parsing isn't complete
428
+ already_added = any(
429
+ tool.get("name") == self.current_function_name
430
+ for tool in self.prev_tool_call_arr)
431
+ if not already_added:
432
+ self.prev_tool_call_arr.append({
433
+ "name": self.current_function_name,
434
+ "arguments":
435
+ "{}", # Placeholder, will be updated later
436
+ })
437
+
438
+ # Send header with function info
439
+ return DeltaMessage(tool_calls=[
440
+ DeltaToolCall(
441
+ index=self.current_tool_index,
442
+ id=self.current_tool_id,
443
+ function=DeltaFunctionCall(
444
+ name=self.current_function_name, arguments=""),
445
+ type="function",
446
+ )
447
+ ])
448
+ return None
449
+
450
+ # We've sent header, now handle function body
451
+ if self.in_function:
452
+ # Send opening brace if not sent yet
453
+ if not self.json_started and self.parameter_prefix not in delta_text:
454
+ self.json_started = True
455
+ return DeltaMessage(tool_calls=[
456
+ DeltaToolCall(
457
+ index=self.current_tool_index,
458
+ function=DeltaFunctionCall(arguments="{"),
459
+ )
460
+ ])
461
+
462
+ # Make sure json_started is set if we're processing parameters
463
+ if not self.json_started:
464
+ self.json_started = True
465
+
466
+ # Check for function end in accumulated text
467
+ if not self.json_closed and self.function_end_token in tool_text:
468
+ # Close JSON
469
+ self.json_closed = True
470
+
471
+ # Extract the complete tool call to update prev_tool_call_arr with final arguments
472
+ # Find the function content
473
+ func_start = tool_text.find(self.tool_call_prefix) + len(
474
+ self.tool_call_prefix)
475
+ func_content_end = tool_text.find(self.function_end_token,
476
+ func_start)
477
+ if func_content_end != -1:
478
+ func_content = tool_text[func_start:func_content_end]
479
+ # Parse to get the complete arguments
480
+ try:
481
+ parsed_tool = self._parse_xml_function_call(
482
+ func_content, self.streaming_request.tools
483
+ if self.streaming_request else None)
484
+ if parsed_tool:
485
+ # Update existing entry in prev_tool_call_arr with complete arguments
486
+ for i, tool in enumerate(self.prev_tool_call_arr):
487
+ if tool.get(
488
+ "name") == parsed_tool.function.name:
489
+ self.prev_tool_call_arr[i][
490
+ "arguments"] = parsed_tool.function.arguments
491
+ break
492
+ except Exception:
493
+ pass # Ignore parsing errors during streaming
494
+
495
+ result = DeltaMessage(tool_calls=[
496
+ DeltaToolCall(
497
+ index=self.current_tool_index,
498
+ function=DeltaFunctionCall(arguments="}"),
499
+ )
500
+ ])
501
+
502
+ # Reset state for next tool
503
+ self.in_function = False
504
+ self.json_closed = True
505
+ self.accumulated_params = {}
506
+
507
+ return result
508
+
509
+ # Look for parameters
510
+ # Find all parameter starts
511
+ param_starts = []
512
+ idx = 0
513
+ while True:
514
+ idx = tool_text.find(self.parameter_prefix, idx)
515
+ if idx == -1:
516
+ break
517
+ param_starts.append(idx)
518
+ idx += len(self.parameter_prefix)
519
+
520
+ # Check if we should start a new parameter
521
+ if not self.in_param and self.param_count < len(param_starts):
522
+
523
+ if len(param_starts) > self.param_count:
524
+ # Process the next parameter
525
+ param_idx = param_starts[self.param_count]
526
+ param_start = param_idx + len(self.parameter_prefix)
527
+ remaining = tool_text[param_start:]
528
+
529
+ if ">" in remaining:
530
+ # We have the complete parameter name
531
+ name_end = remaining.find(">")
532
+ self.current_param_name = remaining[:name_end]
533
+
534
+ # Find the parameter value
535
+ value_start = param_start + name_end + 1
536
+ value_text = tool_text[value_start:]
537
+ if value_text.startswith("\n"):
538
+ value_text = value_text[1:]
539
+
540
+ # Find where this parameter ends
541
+ param_end_idx = value_text.find(
542
+ self.parameter_end_token)
543
+ if param_end_idx == -1:
544
+ # No closing tag, look for next parameter or function end
545
+ next_param_idx = value_text.find(
546
+ self.parameter_prefix)
547
+ func_end_idx = value_text.find(
548
+ self.function_end_token)
549
+
550
+ if next_param_idx != -1 and (func_end_idx == -1
551
+ or next_param_idx
552
+ < func_end_idx):
553
+ param_end_idx = next_param_idx
554
+ elif func_end_idx != -1:
555
+ param_end_idx = func_end_idx
556
+ else:
557
+ # Neither found, check if tool call is complete
558
+ if self.tool_call_end_token in tool_text:
559
+ # Tool call is complete, so parameter must be complete too
560
+ # Use all remaining text before function end as value
561
+ param_end_idx = len(value_text)
562
+ else:
563
+ # Still streaming, wait for more content
564
+ return None
565
+
566
+ if param_end_idx != -1:
567
+ # Complete parameter found
568
+ param_value = value_text[:param_end_idx]
569
+ if param_value.endswith("\n"):
570
+ param_value = param_value[:-1]
571
+
572
+ # Store raw value for later processing
573
+ self.accumulated_params[
574
+ self.current_param_name] = param_value
575
+
576
+ # Get parameter configuration for type conversion
577
+ param_config = self._get_arguments_config(
578
+ self.current_function_name,
579
+ self.streaming_request.tools
580
+ if self.streaming_request else None)
581
+
582
+ # Convert the parameter value to the appropriate type
583
+ converted_value = self._convert_param_value(
584
+ param_value, self.current_param_name,
585
+ param_config, self.current_function_name)
586
+
587
+ # Build JSON fragment based on the converted type
588
+ # Use json.dumps to properly serialize the value
589
+ serialized_value = json.dumps(converted_value,
590
+ ensure_ascii=False)
591
+
592
+ if self.param_count == 0:
593
+ json_fragment = f'"{self.current_param_name}": {serialized_value}'
594
+ else:
595
+ json_fragment = f', "{self.current_param_name}": {serialized_value}'
596
+
597
+ self.param_count += 1
598
+
599
+ return DeltaMessage(tool_calls=[
600
+ DeltaToolCall(
601
+ index=self.current_tool_index,
602
+ function=DeltaFunctionCall(
603
+ arguments=json_fragment),
604
+ )
605
+ ])
606
+
607
+ # Continue parameter value - Not used in the current implementation
608
+ # since we process complete parameters above
609
+ if self.in_param:
610
+ if self.parameter_end_token in delta_text:
611
+ # End of parameter
612
+ end_idx = delta_text.find(self.parameter_end_token)
613
+ value_chunk = delta_text[:end_idx]
614
+
615
+ # Skip past > if at start
616
+ if not self.current_param_value and ">" in value_chunk:
617
+ gt_idx = value_chunk.find(">")
618
+ value_chunk = value_chunk[gt_idx + 1:]
619
+
620
+ if not self.current_param_value and value_chunk.startswith(
621
+ "\n"):
622
+ value_chunk = value_chunk[1:]
623
+
624
+ # Store complete value
625
+ full_value = self.current_param_value + value_chunk
626
+ self.accumulated_params[
627
+ self.current_param_name] = full_value
628
+
629
+ # Get parameter configuration for type conversion
630
+ param_config = self._get_arguments_config(
631
+ self.current_function_name,
632
+ self.streaming_request.tools
633
+ if self.streaming_request else None)
634
+
635
+ # Convert the parameter value to the appropriate type
636
+ converted_value = self._convert_param_value(
637
+ full_value, self.current_param_name, param_config,
638
+ self.current_function_name)
639
+
640
+ # Serialize the converted value
641
+ serialized_value = json.dumps(converted_value,
642
+ ensure_ascii=False)
643
+
644
+ # Since we've been streaming the quoted version, we need to close it properly
645
+ # This is complex - for now just complete the value
646
+ self.in_param = False
647
+ self.current_param_value = ""
648
+
649
+ # Just close the current parameter string
650
+ return DeltaMessage(tool_calls=[
651
+ DeltaToolCall(
652
+ index=self.current_tool_index,
653
+ function=DeltaFunctionCall(
654
+ arguments='"'), # Close the string quote
655
+ )
656
+ ])
657
+ else:
658
+ # Continue accumulating value
659
+ value_chunk = delta_text
660
+
661
+ # Handle first chunk after param name
662
+ if not self.current_param_value and ">" in value_chunk:
663
+ gt_idx = value_chunk.find(">")
664
+ value_chunk = value_chunk[gt_idx + 1:]
665
+
666
+ if not self.current_param_value and value_chunk.startswith(
667
+ "\n"):
668
+ value_chunk = value_chunk[1:]
669
+
670
+ if value_chunk:
671
+ # Stream the escaped delta
672
+ prev_escaped = json.dumps(
673
+ self.current_param_value, ensure_ascii=False
674
+ )[1:-1] if self.current_param_value else ""
675
+ self.current_param_value += value_chunk
676
+ full_escaped = json.dumps(self.current_param_value,
677
+ ensure_ascii=False)[1:-1]
678
+ delta_escaped = full_escaped[len(prev_escaped):]
679
+
680
+ if delta_escaped:
681
+ return DeltaMessage(tool_calls=[
682
+ DeltaToolCall(
683
+ index=self.current_tool_index,
684
+ function=DeltaFunctionCall(
685
+ arguments=delta_escaped),
686
+ )
687
+ ])
688
+
689
+ return None
special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
3
+ size 11422654
tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "extra_special_tokens": {},
234
+ "model_max_length": 1048576,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff