Hugging Face's logo

zec707
/
Qwen3-Coder-Next-oQ4

MLX
Safetensors
qwen3_next
oq
quantized
oQ4
80b
agent
qwen3
4-bit precision
Model card Files
xet
 Community
zec707 commited on
Commit
4b67494
·
verified ·
1 Parent(s): 5ecb05a

Upload Qwen3-Coder-Next-oQ4 via oMLX

Browse files
Files changed (19) hide show
  1. README.md +18 -3
  2. chat_template.jinja +117 -0
  3. config.json +4024 -0
  4. generation_config.json +13 -0
  5. model-00001-of-00009.safetensors +3 -0
  6. model-00002-of-00009.safetensors +3 -0
  7. model-00003-of-00009.safetensors +3 -0
  8. model-00004-of-00009.safetensors +3 -0
  9. model-00005-of-00009.safetensors +3 -0
  10. model-00006-of-00009.safetensors +3 -0
  11. model-00007-of-00009.safetensors +3 -0
  12. model-00008-of-00009.safetensors +3 -0
  13. model-00009-of-00009.safetensors +3 -0
  14. model.safetensors.index.json +0 -0
  15. qwen3_coder_detector_sgl.py +474 -0
  16. qwen3coder_tool_parser_vllm.py +690 -0
  17. tokenizer.json +0 -0
  18. tokenizer_config.json +239 -0
  19. vocab.json +0 -0
README.md CHANGED
@@ -1,3 +1,18 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: mlx
3
+ tags:
4
+ - mlx
5
+ - oq
6
+ - quantized
7
+ ---
8
+
9
+ # Qwen3-Coder-Next-oQ4
10
+
11
+ This model was quantized using [oQ](https://github.com/jundot/omlx) mixed-precision quantization.
12
+
13
+ ## Quantization details
14
+
15
+ - **Model type**: qwen3_next
16
+ - **Bits**: 4
17
+ - **Group size**: 64
18
+ - **Format**: MLX safetensors
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 string %}
5
+ {{-'\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' }}
6
+ {%- else %}
7
+ {{- '\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ 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 if args_value is string else args_value | tojson | safe %}
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' }}
102
+ {%- endif %}
103
+ {{- '\n<tool_response>\n' }}
104
+ {{- message.content }}
105
+ {{- '\n</tool_response>' }}
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 +1,4024 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3NextForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0,
7
+ "bos_token_id": 151643,
8
+ "decoder_sparse_step": 1,
9
+ "eos_token_id": 151645,
10
+ "full_attention_interval": 4,
11
+ "head_dim": 256,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 2048,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 5120,
16
+ "linear_conv_kernel_dim": 4,
17
+ "linear_key_head_dim": 128,
18
+ "linear_num_key_heads": 16,
19
+ "linear_num_value_heads": 32,
20
+ "linear_value_head_dim": 128,
21
+ "max_position_embeddings": 262144,
22
+ "mlp_only_layers": [],
23
+ "model_type": "qwen3_next",
24
+ "moe_intermediate_size": 512,
25
+ "norm_topk_prob": true,
26
+ "num_attention_heads": 16,
27
+ "num_experts": 512,
28
+ "num_experts_per_tok": 10,
29
+ "num_hidden_layers": 48,
30
+ "num_key_value_heads": 2,
31
+ "output_router_logits": false,
32
+ "partial_rotary_factor": 0.25,
33
+ "rms_norm_eps": 1e-06,
34
+ "rope_scaling": null,
35
+ "rope_theta": 5000000,
36
+ "router_aux_loss_coef": 0.001,
37
+ "shared_expert_intermediate_size": 512,
38
+ "tie_word_embeddings": false,
39
+ "torch_dtype": "bfloat16",
40
+ "transformers_version": "4.57.0.dev0",
41
+ "use_cache": true,
42
+ "use_sliding_window": false,
43
+ "vocab_size": 151936,
44
+ "quantization": {
45
+ "group_size": 64,
46
+ "bits": 4,
47
+ "mode": "affine",
48
+ "model.layers.1.mlp.shared_expert_gate": {
49
+ "bits": 8,
50
+ "group_size": 64,
51
+ "mode": "affine"
52
+ },
53
+ "model.layers.1.mlp.shared_expert.up_proj": {
54
+ "bits": 6,
55
+ "group_size": 64,
56
+ "mode": "affine"
57
+ },
58
+ "model.layers.1.mlp.shared_expert.gate_proj": {
59
+ "bits": 6,
60
+ "group_size": 64,
61
+ "mode": "affine"
62
+ },
63
+ "model.layers.1.linear_attn.in_proj_qkvz": {
64
+ "bits": 6,
65
+ "group_size": 64,
66
+ "mode": "affine"
67
+ },
68
+ "model.layers.0.mlp.shared_expert.up_proj": {
69
+ "bits": 6,
70
+ "group_size": 64,
71
+ "mode": "affine"
72
+ },
73
+ "model.layers.0.mlp.shared_expert.gate_proj": {
74
+ "bits": 6,
75
+ "group_size": 64,
76
+ "mode": "affine"
77
+ },
78
+ "model.layers.0.mlp.gate": {
79
+ "bits": 8,
80
+ "group_size": 64,
81
+ "mode": "affine"
82
+ },
83
+ "model.layers.1.mlp.shared_expert.down_proj": {
84
+ "bits": 6,
85
+ "group_size": 64,
86
+ "mode": "affine"
87
+ },
88
+ "model.layers.1.mlp.gate": {
89
+ "bits": 8,
90
+ "group_size": 64,
91
+ "mode": "affine"
92
+ },
93
+ "model.layers.1.linear_attn.in_proj_ba": {
94
+ "bits": 6,
95
+ "group_size": 64,
96
+ "mode": "affine"
97
+ },
98
+ "model.layers.0.mlp.shared_expert_gate": {
99
+ "bits": 8,
100
+ "group_size": 64,
101
+ "mode": "affine"
102
+ },
103
+ "model.layers.0.linear_attn.in_proj_ba": {
104
+ "bits": 6,
105
+ "group_size": 64,
106
+ "mode": "affine"
107
+ },
108
+ "model.layers.0.linear_attn.out_proj": {
109
+ "bits": 6,
110
+ "group_size": 64,
111
+ "mode": "affine"
112
+ },
113
+ "model.layers.1.linear_attn.out_proj": {
114
+ "bits": 6,
115
+ "group_size": 64,
116
+ "mode": "affine"
117
+ },
118
+ "model.layers.0.mlp.shared_expert.down_proj": {
119
+ "bits": 6,
120
+ "group_size": 64,
121
+ "mode": "affine"
122
+ },
123
+ "model.layers.0.linear_attn.in_proj_qkvz": {
124
+ "bits": 6,
125
+ "group_size": 64,
126
+ "mode": "affine"
127
+ },
128
+ "model.layers.2.mlp.shared_expert_gate": {
129
+ "bits": 8,
130
+ "group_size": 64,
131
+ "mode": "affine"
132
+ },
133
+ "model.layers.2.mlp.gate": {
134
+ "bits": 8,
135
+ "group_size": 64,
136
+ "mode": "affine"
137
+ },
138
+ "model.layers.2.mlp.shared_expert.up_proj": {
139
+ "bits": 6,
140
+ "group_size": 64,
141
+ "mode": "affine"
142
+ },
143
+ "model.layers.2.linear_attn.out_proj": {
144
+ "bits": 6,
145
+ "group_size": 64,
146
+ "mode": "affine"
147
+ },
148
+ "model.layers.2.linear_attn.in_proj_ba": {
149
+ "bits": 6,
150
+ "group_size": 64,
151
+ "mode": "affine"
152
+ },
153
+ "model.layers.2.linear_attn.in_proj_qkvz": {
154
+ "bits": 6,
155
+ "group_size": 64,
156
+ "mode": "affine"
157
+ },
158
+ "model.layers.2.mlp.shared_expert.gate_proj": {
159
+ "bits": 6,
160
+ "group_size": 64,
161
+ "mode": "affine"
162
+ },
163
+ "model.layers.2.mlp.shared_expert.down_proj": {
164
+ "bits": 6,
165
+ "group_size": 64,
166
+ "mode": "affine"
167
+ },
168
+ "model.layers.3.self_attn.v_proj": {
169
+ "bits": 8,
170
+ "group_size": 32,
171
+ "mode": "mxfp8"
172
+ },
173
+ "model.layers.3.self_attn.q_proj": {
174
+ "bits": 8,
175
+ "group_size": 32,
176
+ "mode": "mxfp8"
177
+ },
178
+ "model.layers.3.self_attn.o_proj": {
179
+ "bits": 8,
180
+ "group_size": 32,
181
+ "mode": "mxfp8"
182
+ },
183
+ "model.layers.3.mlp.gate": {
184
+ "bits": 8,
185
+ "group_size": 64,
186
+ "mode": "affine"
187
+ },
188
+ "model.layers.3.mlp.shared_expert.gate_proj": {
189
+ "bits": 8,
190
+ "group_size": 32,
191
+ "mode": "mxfp8"
192
+ },
193
+ "model.layers.3.mlp.shared_expert.up_proj": {
194
+ "bits": 8,
195
+ "group_size": 32,
196
+ "mode": "mxfp8"
197
+ },
198
+ "model.layers.3.mlp.shared_expert_gate": {
199
+ "bits": 8,
200
+ "group_size": 64,
201
+ "mode": "affine"
202
+ },
203
+ "model.layers.3.mlp.shared_expert.down_proj": {
204
+ "bits": 8,
205
+ "group_size": 32,
206
+ "mode": "mxfp8"
207
+ },
208
+ "model.layers.3.self_attn.k_proj": {
209
+ "bits": 8,
210
+ "group_size": 32,
211
+ "mode": "mxfp8"
212
+ },
213
+ "model.layers.4.mlp.shared_expert.up_proj": {
214
+ "bits": 6,
215
+ "group_size": 64,
216
+ "mode": "affine"
217
+ },
218
+ "model.layers.4.mlp.shared_expert.gate_proj": {
219
+ "bits": 6,
220
+ "group_size": 64,
221
+ "mode": "affine"
222
+ },
223
+ "model.layers.4.mlp.shared_expert.down_proj": {
224
+ "bits": 6,
225
+ "group_size": 64,
226
+ "mode": "affine"
227
+ },
228
+ "model.layers.4.mlp.gate": {
229
+ "bits": 8,
230
+ "group_size": 64,
231
+ "mode": "affine"
232
+ },
233
+ "model.layers.4.linear_attn.in_proj_qkvz": {
234
+ "bits": 6,
235
+ "group_size": 64,
236
+ "mode": "affine"
237
+ },
238
+ "model.layers.4.linear_attn.in_proj_ba": {
239
+ "bits": 6,
240
+ "group_size": 64,
241
+ "mode": "affine"
242
+ },
243
+ "model.layers.4.linear_attn.out_proj": {
244
+ "bits": 6,
245
+ "group_size": 64,
246
+ "mode": "affine"
247
+ },
248
+ "model.layers.4.mlp.shared_expert_gate": {
249
+ "bits": 8,
250
+ "group_size": 64,
251
+ "mode": "affine"
252
+ },
253
+ "model.layers.5.mlp.shared_expert.up_proj": {
254
+ "bits": 5,
255
+ "group_size": 64,
256
+ "mode": "affine"
257
+ },
258
+ "model.layers.5.mlp.shared_expert.down_proj": {
259
+ "bits": 5,
260
+ "group_size": 64,
261
+ "mode": "affine"
262
+ },
263
+ "model.layers.5.mlp.shared_expert.gate_proj": {
264
+ "bits": 5,
265
+ "group_size": 64,
266
+ "mode": "affine"
267
+ },
268
+ "model.layers.5.mlp.shared_expert_gate": {
269
+ "bits": 8,
270
+ "group_size": 64,
271
+ "mode": "affine"
272
+ },
273
+ "model.layers.5.linear_attn.out_proj": {
274
+ "bits": 5,
275
+ "group_size": 64,
276
+ "mode": "affine"
277
+ },
278
+ "model.layers.5.linear_attn.in_proj_qkvz": {
279
+ "bits": 5,
280
+ "group_size": 64,
281
+ "mode": "affine"
282
+ },
283
+ "model.layers.5.linear_attn.in_proj_ba": {
284
+ "bits": 5,
285
+ "group_size": 64,
286
+ "mode": "affine"
287
+ },
288
+ "model.layers.5.mlp.gate": {
289
+ "bits": 8,
290
+ "group_size": 64,
291
+ "mode": "affine"
292
+ },
293
+ "model.layers.7.self_attn.v_proj": {
294
+ "bits": 6,
295
+ "group_size": 64,
296
+ "mode": "affine"
297
+ },
298
+ "model.layers.7.self_attn.q_proj": {
299
+ "bits": 6,
300
+ "group_size": 64,
301
+ "mode": "affine"
302
+ },
303
+ "model.layers.7.self_attn.o_proj": {
304
+ "bits": 6,
305
+ "group_size": 64,
306
+ "mode": "affine"
307
+ },
308
+ "model.layers.7.self_attn.k_proj": {
309
+ "bits": 6,
310
+ "group_size": 64,
311
+ "mode": "affine"
312
+ },
313
+ "model.layers.7.mlp.shared_expert.up_proj": {
314
+ "bits": 6,
315
+ "group_size": 64,
316
+ "mode": "affine"
317
+ },
318
+ "model.layers.7.mlp.shared_expert.down_proj": {
319
+ "bits": 6,
320
+ "group_size": 64,
321
+ "mode": "affine"
322
+ },
323
+ "model.layers.6.mlp.gate": {
324
+ "bits": 8,
325
+ "group_size": 64,
326
+ "mode": "affine"
327
+ },
328
+ "model.layers.7.mlp.gate": {
329
+ "bits": 8,
330
+ "group_size": 64,
331
+ "mode": "affine"
332
+ },
333
+ "model.layers.6.mlp.shared_expert.gate_proj": {
334
+ "bits": 6,
335
+ "group_size": 64,
336
+ "mode": "affine"
337
+ },
338
+ "model.layers.7.mlp.shared_expert_gate": {
339
+ "bits": 8,
340
+ "group_size": 64,
341
+ "mode": "affine"
342
+ },
343
+ "model.layers.6.mlp.shared_expert_gate": {
344
+ "bits": 8,
345
+ "group_size": 64,
346
+ "mode": "affine"
347
+ },
348
+ "model.layers.6.linear_attn.in_proj_ba": {
349
+ "bits": 6,
350
+ "group_size": 64,
351
+ "mode": "affine"
352
+ },
353
+ "model.layers.6.linear_attn.out_proj": {
354
+ "bits": 6,
355
+ "group_size": 64,
356
+ "mode": "affine"
357
+ },
358
+ "model.layers.6.mlp.shared_expert.up_proj": {
359
+ "bits": 6,
360
+ "group_size": 64,
361
+ "mode": "affine"
362
+ },
363
+ "model.layers.7.mlp.shared_expert.gate_proj": {
364
+ "bits": 6,
365
+ "group_size": 64,
366
+ "mode": "affine"
367
+ },
368
+ "model.layers.6.mlp.shared_expert.down_proj": {
369
+ "bits": 6,
370
+ "group_size": 64,
371
+ "mode": "affine"
372
+ },
373
+ "model.layers.6.linear_attn.in_proj_qkvz": {
374
+ "bits": 6,
375
+ "group_size": 64,
376
+ "mode": "affine"
377
+ },
378
+ "model.layers.8.mlp.shared_expert_gate": {
379
+ "bits": 8,
380
+ "group_size": 64,
381
+ "mode": "affine"
382
+ },
383
+ "model.layers.8.mlp.shared_expert.gate_proj": {
384
+ "bits": 6,
385
+ "group_size": 64,
386
+ "mode": "affine"
387
+ },
388
+ "model.layers.8.mlp.gate": {
389
+ "bits": 8,
390
+ "group_size": 64,
391
+ "mode": "affine"
392
+ },
393
+ "model.layers.8.linear_attn.in_proj_ba": {
394
+ "bits": 6,
395
+ "group_size": 64,
396
+ "mode": "affine"
397
+ },
398
+ "model.layers.8.mlp.shared_expert.up_proj": {
399
+ "bits": 6,
400
+ "group_size": 64,
401
+ "mode": "affine"
402
+ },
403
+ "model.layers.8.linear_attn.in_proj_qkvz": {
404
+ "bits": 6,
405
+ "group_size": 64,
406
+ "mode": "affine"
407
+ },
408
+ "model.layers.8.linear_attn.out_proj": {
409
+ "bits": 6,
410
+ "group_size": 64,
411
+ "mode": "affine"
412
+ },
413
+ "model.layers.8.mlp.shared_expert.down_proj": {
414
+ "bits": 6,
415
+ "group_size": 64,
416
+ "mode": "affine"
417
+ },
418
+ "model.layers.9.mlp.shared_expert.up_proj": {
419
+ "bits": 6,
420
+ "group_size": 64,
421
+ "mode": "affine"
422
+ },
423
+ "model.layers.9.mlp.shared_expert.gate_proj": {
424
+ "bits": 6,
425
+ "group_size": 64,
426
+ "mode": "affine"
427
+ },
428
+ "model.layers.9.mlp.shared_expert.down_proj": {
429
+ "bits": 6,
430
+ "group_size": 64,
431
+ "mode": "affine"
432
+ },
433
+ "model.layers.9.mlp.gate": {
434
+ "bits": 8,
435
+ "group_size": 64,
436
+ "mode": "affine"
437
+ },
438
+ "model.layers.9.linear_attn.in_proj_ba": {
439
+ "bits": 6,
440
+ "group_size": 64,
441
+ "mode": "affine"
442
+ },
443
+ "model.layers.9.linear_attn.in_proj_qkvz": {
444
+ "bits": 6,
445
+ "group_size": 64,
446
+ "mode": "affine"
447
+ },
448
+ "model.layers.9.mlp.shared_expert_gate": {
449
+ "bits": 8,
450
+ "group_size": 64,
451
+ "mode": "affine"
452
+ },
453
+ "model.layers.9.linear_attn.out_proj": {
454
+ "bits": 6,
455
+ "group_size": 64,
456
+ "mode": "affine"
457
+ },
458
+ "model.layers.10.mlp.shared_expert_gate": {
459
+ "bits": 8,
460
+ "group_size": 64,
461
+ "mode": "affine"
462
+ },
463
+ "model.layers.10.mlp.shared_expert.gate_proj": {
464
+ "bits": 6,
465
+ "group_size": 64,
466
+ "mode": "affine"
467
+ },
468
+ "model.layers.10.mlp.shared_expert.down_proj": {
469
+ "bits": 6,
470
+ "group_size": 64,
471
+ "mode": "affine"
472
+ },
473
+ "model.layers.10.mlp.gate": {
474
+ "bits": 8,
475
+ "group_size": 64,
476
+ "mode": "affine"
477
+ },
478
+ "model.layers.10.linear_attn.in_proj_ba": {
479
+ "bits": 6,
480
+ "group_size": 64,
481
+ "mode": "affine"
482
+ },
483
+ "model.layers.10.mlp.shared_expert.up_proj": {
484
+ "bits": 6,
485
+ "group_size": 64,
486
+ "mode": "affine"
487
+ },
488
+ "model.layers.10.linear_attn.in_proj_qkvz": {
489
+ "bits": 6,
490
+ "group_size": 64,
491
+ "mode": "affine"
492
+ },
493
+ "model.layers.10.linear_attn.out_proj": {
494
+ "bits": 6,
495
+ "group_size": 64,
496
+ "mode": "affine"
497
+ },
498
+ "model.layers.11.self_attn.v_proj": {
499
+ "bits": 6,
500
+ "group_size": 64,
501
+ "mode": "affine"
502
+ },
503
+ "model.layers.11.self_attn.o_proj": {
504
+ "bits": 6,
505
+ "group_size": 64,
506
+ "mode": "affine"
507
+ },
508
+ "model.layers.11.mlp.shared_expert.gate_proj": {
509
+ "bits": 6,
510
+ "group_size": 64,
511
+ "mode": "affine"
512
+ },
513
+ "model.layers.11.mlp.shared_expert.down_proj": {
514
+ "bits": 6,
515
+ "group_size": 64,
516
+ "mode": "affine"
517
+ },
518
+ "model.layers.11.mlp.shared_expert.up_proj": {
519
+ "bits": 6,
520
+ "group_size": 64,
521
+ "mode": "affine"
522
+ },
523
+ "model.layers.11.mlp.shared_expert_gate": {
524
+ "bits": 8,
525
+ "group_size": 64,
526
+ "mode": "affine"
527
+ },
528
+ "model.layers.11.self_attn.k_proj": {
529
+ "bits": 6,
530
+ "group_size": 64,
531
+ "mode": "affine"
532
+ },
533
+ "model.layers.11.mlp.gate": {
534
+ "bits": 8,
535
+ "group_size": 64,
536
+ "mode": "affine"
537
+ },
538
+ "model.layers.11.self_attn.q_proj": {
539
+ "bits": 6,
540
+ "group_size": 64,
541
+ "mode": "affine"
542
+ },
543
+ "model.layers.13.mlp.shared_expert.up_proj": {
544
+ "bits": 5,
545
+ "group_size": 64,
546
+ "mode": "affine"
547
+ },
548
+ "model.layers.13.mlp.shared_expert.down_proj": {
549
+ "bits": 5,
550
+ "group_size": 64,
551
+ "mode": "affine"
552
+ },
553
+ "model.layers.13.linear_attn.out_proj": {
554
+ "bits": 5,
555
+ "group_size": 64,
556
+ "mode": "affine"
557
+ },
558
+ "model.layers.13.linear_attn.in_proj_qkvz": {
559
+ "bits": 5,
560
+ "group_size": 64,
561
+ "mode": "affine"
562
+ },
563
+ "model.layers.13.linear_attn.in_proj_ba": {
564
+ "bits": 5,
565
+ "group_size": 64,
566
+ "mode": "affine"
567
+ },
568
+ "model.layers.12.mlp.shared_expert_gate": {
569
+ "bits": 8,
570
+ "group_size": 64,
571
+ "mode": "affine"
572
+ },
573
+ "model.layers.12.mlp.shared_expert.up_proj": {
574
+ "bits": 6,
575
+ "group_size": 64,
576
+ "mode": "affine"
577
+ },
578
+ "model.layers.12.mlp.shared_expert.gate_proj": {
579
+ "bits": 6,
580
+ "group_size": 64,
581
+ "mode": "affine"
582
+ },
583
+ "model.layers.12.mlp.shared_expert.down_proj": {
584
+ "bits": 6,
585
+ "group_size": 64,
586
+ "mode": "affine"
587
+ },
588
+ "model.layers.12.mlp.gate": {
589
+ "bits": 8,
590
+ "group_size": 64,
591
+ "mode": "affine"
592
+ },
593
+ "model.layers.13.mlp.shared_expert_gate": {
594
+ "bits": 8,
595
+ "group_size": 64,
596
+ "mode": "affine"
597
+ },
598
+ "model.layers.12.linear_attn.in_proj_ba": {
599
+ "bits": 6,
600
+ "group_size": 64,
601
+ "mode": "affine"
602
+ },
603
+ "model.layers.13.mlp.gate": {
604
+ "bits": 8,
605
+ "group_size": 64,
606
+ "mode": "affine"
607
+ },
608
+ "model.layers.13.mlp.shared_expert.gate_proj": {
609
+ "bits": 5,
610
+ "group_size": 64,
611
+ "mode": "affine"
612
+ },
613
+ "model.layers.12.linear_attn.in_proj_qkvz": {
614
+ "bits": 6,
615
+ "group_size": 64,
616
+ "mode": "affine"
617
+ },
618
+ "model.layers.12.linear_attn.out_proj": {
619
+ "bits": 6,
620
+ "group_size": 64,
621
+ "mode": "affine"
622
+ },
623
+ "model.layers.14.mlp.shared_expert.up_proj": {
624
+ "bits": 5,
625
+ "group_size": 64,
626
+ "mode": "affine"
627
+ },
628
+ "model.layers.14.mlp.shared_expert.gate_proj": {
629
+ "bits": 5,
630
+ "group_size": 64,
631
+ "mode": "affine"
632
+ },
633
+ "model.layers.14.linear_attn.in_proj_qkvz": {
634
+ "bits": 5,
635
+ "group_size": 64,
636
+ "mode": "affine"
637
+ },
638
+ "model.layers.14.linear_attn.out_proj": {
639
+ "bits": 5,
640
+ "group_size": 64,
641
+ "mode": "affine"
642
+ },
643
+ "model.layers.14.mlp.gate": {
644
+ "bits": 8,
645
+ "group_size": 64,
646
+ "mode": "affine"
647
+ },
648
+ "model.layers.14.linear_attn.in_proj_ba": {
649
+ "bits": 5,
650
+ "group_size": 64,
651
+ "mode": "affine"
652
+ },
653
+ "model.layers.14.mlp.shared_expert_gate": {
654
+ "bits": 8,
655
+ "group_size": 64,
656
+ "mode": "affine"
657
+ },
658
+ "model.layers.14.mlp.shared_expert.down_proj": {
659
+ "bits": 5,
660
+ "group_size": 64,
661
+ "mode": "affine"
662
+ },
663
+ "model.layers.15.self_attn.o_proj": {
664
+ "bits": 6,
665
+ "group_size": 64,
666
+ "mode": "affine"
667
+ },
668
+ "model.layers.15.self_attn.k_proj": {
669
+ "bits": 6,
670
+ "group_size": 64,
671
+ "mode": "affine"
672
+ },
673
+ "model.layers.15.mlp.gate": {
674
+ "bits": 8,
675
+ "group_size": 64,
676
+ "mode": "affine"
677
+ },
678
+ "model.layers.15.mlp.shared_expert_gate": {
679
+ "bits": 8,
680
+ "group_size": 64,
681
+ "mode": "affine"
682
+ },
683
+ "model.layers.15.mlp.shared_expert.down_proj": {
684
+ "bits": 6,
685
+ "group_size": 64,
686
+ "mode": "affine"
687
+ },
688
+ "model.layers.15.self_attn.v_proj": {
689
+ "bits": 6,
690
+ "group_size": 64,
691
+ "mode": "affine"
692
+ },
693
+ "model.layers.15.mlp.shared_expert.up_proj": {
694
+ "bits": 6,
695
+ "group_size": 64,
696
+ "mode": "affine"
697
+ },
698
+ "model.layers.15.self_attn.q_proj": {
699
+ "bits": 6,
700
+ "group_size": 64,
701
+ "mode": "affine"
702
+ },
703
+ "model.layers.15.mlp.shared_expert.gate_proj": {
704
+ "bits": 6,
705
+ "group_size": 64,
706
+ "mode": "affine"
707
+ },
708
+ "model.layers.16.mlp.shared_expert.up_proj": {
709
+ "bits": 5,
710
+ "group_size": 64,
711
+ "mode": "affine"
712
+ },
713
+ "model.layers.16.mlp.shared_expert.down_proj": {
714
+ "bits": 5,
715
+ "group_size": 64,
716
+ "mode": "affine"
717
+ },
718
+ "model.layers.16.mlp.gate": {
719
+ "bits": 8,
720
+ "group_size": 64,
721
+ "mode": "affine"
722
+ },
723
+ "model.layers.16.mlp.shared_expert_gate": {
724
+ "bits": 8,
725
+ "group_size": 64,
726
+ "mode": "affine"
727
+ },
728
+ "model.layers.16.linear_attn.in_proj_ba": {
729
+ "bits": 5,
730
+ "group_size": 64,
731
+ "mode": "affine"
732
+ },
733
+ "model.layers.16.linear_attn.in_proj_qkvz": {
734
+ "bits": 5,
735
+ "group_size": 64,
736
+ "mode": "affine"
737
+ },
738
+ "model.layers.16.mlp.shared_expert.gate_proj": {
739
+ "bits": 5,
740
+ "group_size": 64,
741
+ "mode": "affine"
742
+ },
743
+ "model.layers.16.linear_attn.out_proj": {
744
+ "bits": 5,
745
+ "group_size": 64,
746
+ "mode": "affine"
747
+ },
748
+ "model.layers.18.mlp.shared_expert_gate": {
749
+ "bits": 8,
750
+ "group_size": 64,
751
+ "mode": "affine"
752
+ },
753
+ "model.layers.18.mlp.shared_expert.up_proj": {
754
+ "bits": 5,
755
+ "group_size": 64,
756
+ "mode": "affine"
757
+ },
758
+ "model.layers.18.linear_attn.out_proj": {
759
+ "bits": 5,
760
+ "group_size": 64,
761
+ "mode": "affine"
762
+ },
763
+ "model.layers.18.linear_attn.in_proj_qkvz": {
764
+ "bits": 5,
765
+ "group_size": 64,
766
+ "mode": "affine"
767
+ },
768
+ "model.layers.18.linear_attn.in_proj_ba": {
769
+ "bits": 5,
770
+ "group_size": 64,
771
+ "mode": "affine"
772
+ },
773
+ "model.layers.17.mlp.shared_expert.gate_proj": {
774
+ "bits": 5,
775
+ "group_size": 64,
776
+ "mode": "affine"
777
+ },
778
+ "model.layers.17.mlp.shared_expert.down_proj": {
779
+ "bits": 5,
780
+ "group_size": 64,
781
+ "mode": "affine"
782
+ },
783
+ "model.layers.17.mlp.gate": {
784
+ "bits": 8,
785
+ "group_size": 64,
786
+ "mode": "affine"
787
+ },
788
+ "model.layers.18.mlp.shared_expert.gate_proj": {
789
+ "bits": 5,
790
+ "group_size": 64,
791
+ "mode": "affine"
792
+ },
793
+ "model.layers.17.mlp.shared_expert.up_proj": {
794
+ "bits": 5,
795
+ "group_size": 64,
796
+ "mode": "affine"
797
+ },
798
+ "model.layers.17.linear_attn.out_proj": {
799
+ "bits": 5,
800
+ "group_size": 64,
801
+ "mode": "affine"
802
+ },
803
+ "model.layers.17.mlp.shared_expert_gate": {
804
+ "bits": 8,
805
+ "group_size": 64,
806
+ "mode": "affine"
807
+ },
808
+ "model.layers.18.mlp.shared_expert.down_proj": {
809
+ "bits": 5,
810
+ "group_size": 64,
811
+ "mode": "affine"
812
+ },
813
+ "model.layers.17.linear_attn.in_proj_ba": {
814
+ "bits": 5,
815
+ "group_size": 64,
816
+ "mode": "affine"
817
+ },
818
+ "model.layers.17.linear_attn.in_proj_qkvz": {
819
+ "bits": 5,
820
+ "group_size": 64,
821
+ "mode": "affine"
822
+ },
823
+ "model.layers.18.mlp.gate": {
824
+ "bits": 8,
825
+ "group_size": 64,
826
+ "mode": "affine"
827
+ },
828
+ "model.layers.19.self_attn.q_proj": {
829
+ "bits": 5,
830
+ "group_size": 64,
831
+ "mode": "affine"
832
+ },
833
+ "model.layers.19.self_attn.o_proj": {
834
+ "bits": 5,
835
+ "group_size": 64,
836
+ "mode": "affine"
837
+ },
838
+ "model.layers.19.mlp.shared_expert.down_proj": {
839
+ "bits": 5,
840
+ "group_size": 64,
841
+ "mode": "affine"
842
+ },
843
+ "model.layers.19.mlp.gate": {
844
+ "bits": 8,
845
+ "group_size": 64,
846
+ "mode": "affine"
847
+ },
848
+ "model.layers.19.self_attn.v_proj": {
849
+ "bits": 5,
850
+ "group_size": 64,
851
+ "mode": "affine"
852
+ },
853
+ "model.layers.19.mlp.shared_expert.gate_proj": {
854
+ "bits": 5,
855
+ "group_size": 64,
856
+ "mode": "affine"
857
+ },
858
+ "model.layers.19.mlp.shared_expert.up_proj": {
859
+ "bits": 5,
860
+ "group_size": 64,
861
+ "mode": "affine"
862
+ },
863
+ "model.layers.19.self_attn.k_proj": {
864
+ "bits": 5,
865
+ "group_size": 64,
866
+ "mode": "affine"
867
+ },
868
+ "model.layers.19.mlp.shared_expert_gate": {
869
+ "bits": 8,
870
+ "group_size": 64,
871
+ "mode": "affine"
872
+ },
873
+ "model.layers.20.mlp.shared_expert.up_proj": {
874
+ "bits": 6,
875
+ "group_size": 64,
876
+ "mode": "affine"
877
+ },
878
+ "model.layers.20.mlp.gate": {
879
+ "bits": 8,
880
+ "group_size": 64,
881
+ "mode": "affine"
882
+ },
883
+ "model.layers.20.mlp.shared_expert.gate_proj": {
884
+ "bits": 6,
885
+ "group_size": 64,
886
+ "mode": "affine"
887
+ },
888
+ "model.layers.20.mlp.shared_expert.down_proj": {
889
+ "bits": 6,
890
+ "group_size": 64,
891
+ "mode": "affine"
892
+ },
893
+ "model.layers.20.linear_attn.in_proj_ba": {
894
+ "bits": 6,
895
+ "group_size": 64,
896
+ "mode": "affine"
897
+ },
898
+ "model.layers.20.linear_attn.in_proj_qkvz": {
899
+ "bits": 6,
900
+ "group_size": 64,
901
+ "mode": "affine"
902
+ },
903
+ "model.layers.20.linear_attn.out_proj": {
904
+ "bits": 6,
905
+ "group_size": 64,
906
+ "mode": "affine"
907
+ },
908
+ "model.layers.20.mlp.shared_expert_gate": {
909
+ "bits": 8,
910
+ "group_size": 64,
911
+ "mode": "affine"
912
+ },
913
+ "model.layers.21.mlp.shared_expert_gate": {
914
+ "bits": 8,
915
+ "group_size": 64,
916
+ "mode": "affine"
917
+ },
918
+ "model.layers.21.mlp.shared_expert.up_proj": {
919
+ "bits": 5,
920
+ "group_size": 64,
921
+ "mode": "affine"
922
+ },
923
+ "model.layers.21.mlp.shared_expert.gate_proj": {
924
+ "bits": 5,
925
+ "group_size": 64,
926
+ "mode": "affine"
927
+ },
928
+ "model.layers.21.mlp.shared_expert.down_proj": {
929
+ "bits": 5,
930
+ "group_size": 64,
931
+ "mode": "affine"
932
+ },
933
+ "model.layers.21.mlp.gate": {
934
+ "bits": 8,
935
+ "group_size": 64,
936
+ "mode": "affine"
937
+ },
938
+ "model.layers.21.linear_attn.in_proj_qkvz": {
939
+ "bits": 5,
940
+ "group_size": 64,
941
+ "mode": "affine"
942
+ },
943
+ "model.layers.21.linear_attn.in_proj_ba": {
944
+ "bits": 5,
945
+ "group_size": 64,
946
+ "mode": "affine"
947
+ },
948
+ "model.layers.21.linear_attn.out_proj": {
949
+ "bits": 5,
950
+ "group_size": 64,
951
+ "mode": "affine"
952
+ },
953
+ "model.layers.22.mlp.shared_expert.gate_proj": {
954
+ "bits": 6,
955
+ "group_size": 64,
956
+ "mode": "affine"
957
+ },
958
+ "model.layers.22.mlp.shared_expert.down_proj": {
959
+ "bits": 6,
960
+ "group_size": 64,
961
+ "mode": "affine"
962
+ },
963
+ "model.layers.22.linear_attn.in_proj_ba": {
964
+ "bits": 6,
965
+ "group_size": 64,
966
+ "mode": "affine"
967
+ },
968
+ "model.layers.22.mlp.shared_expert_gate": {
969
+ "bits": 8,
970
+ "group_size": 64,
971
+ "mode": "affine"
972
+ },
973
+ "model.layers.22.linear_attn.out_proj": {
974
+ "bits": 6,
975
+ "group_size": 64,
976
+ "mode": "affine"
977
+ },
978
+ "model.layers.22.linear_attn.in_proj_qkvz": {
979
+ "bits": 6,
980
+ "group_size": 64,
981
+ "mode": "affine"
982
+ },
983
+ "model.layers.22.mlp.shared_expert.up_proj": {
984
+ "bits": 6,
985
+ "group_size": 64,
986
+ "mode": "affine"
987
+ },
988
+ "model.layers.22.mlp.gate": {
989
+ "bits": 8,
990
+ "group_size": 64,
991
+ "mode": "affine"
992
+ },
993
+ "model.layers.24.mlp.shared_expert_gate": {
994
+ "bits": 8,
995
+ "group_size": 64,
996
+ "mode": "affine"
997
+ },
998
+ "model.layers.24.mlp.shared_expert.up_proj": {
999
+ "bits": 6,
1000
+ "group_size": 64,
1001
+ "mode": "affine"
1002
+ },
1003
+ "model.layers.24.mlp.shared_expert.gate_proj": {
1004
+ "bits": 6,
1005
+ "group_size": 64,
1006
+ "mode": "affine"
1007
+ },
1008
+ "model.layers.24.mlp.shared_expert.down_proj": {
1009
+ "bits": 6,
1010
+ "group_size": 64,
1011
+ "mode": "affine"
1012
+ },
1013
+ "model.layers.24.mlp.gate": {
1014
+ "bits": 8,
1015
+ "group_size": 64,
1016
+ "mode": "affine"
1017
+ },
1018
+ "model.layers.24.linear_attn.in_proj_qkvz": {
1019
+ "bits": 6,
1020
+ "group_size": 64,
1021
+ "mode": "affine"
1022
+ },
1023
+ "model.layers.24.linear_attn.in_proj_ba": {
1024
+ "bits": 6,
1025
+ "group_size": 64,
1026
+ "mode": "affine"
1027
+ },
1028
+ "model.layers.23.self_attn.v_proj": {
1029
+ "bits": 6,
1030
+ "group_size": 64,
1031
+ "mode": "affine"
1032
+ },
1033
+ "model.layers.23.self_attn.q_proj": {
1034
+ "bits": 6,
1035
+ "group_size": 64,
1036
+ "mode": "affine"
1037
+ },
1038
+ "model.layers.23.self_attn.o_proj": {
1039
+ "bits": 6,
1040
+ "group_size": 64,
1041
+ "mode": "affine"
1042
+ },
1043
+ "model.layers.23.mlp.shared_expert.up_proj": {
1044
+ "bits": 6,
1045
+ "group_size": 64,
1046
+ "mode": "affine"
1047
+ },
1048
+ "model.layers.23.mlp.shared_expert.gate_proj": {
1049
+ "bits": 6,
1050
+ "group_size": 64,
1051
+ "mode": "affine"
1052
+ },
1053
+ "model.layers.23.mlp.shared_expert.down_proj": {
1054
+ "bits": 6,
1055
+ "group_size": 64,
1056
+ "mode": "affine"
1057
+ },
1058
+ "model.layers.23.mlp.gate": {
1059
+ "bits": 8,
1060
+ "group_size": 64,
1061
+ "mode": "affine"
1062
+ },
1063
+ "model.layers.23.self_attn.k_proj": {
1064
+ "bits": 6,
1065
+ "group_size": 64,
1066
+ "mode": "affine"
1067
+ },
1068
+ "model.layers.23.mlp.shared_expert_gate": {
1069
+ "bits": 8,
1070
+ "group_size": 64,
1071
+ "mode": "affine"
1072
+ },
1073
+ "model.layers.24.linear_attn.out_proj": {
1074
+ "bits": 6,
1075
+ "group_size": 64,
1076
+ "mode": "affine"
1077
+ },
1078
+ "model.layers.25.mlp.shared_expert_gate": {
1079
+ "bits": 8,
1080
+ "group_size": 64,
1081
+ "mode": "affine"
1082
+ },
1083
+ "model.layers.25.mlp.shared_expert.up_proj": {
1084
+ "bits": 6,
1085
+ "group_size": 64,
1086
+ "mode": "affine"
1087
+ },
1088
+ "model.layers.25.mlp.shared_expert.down_proj": {
1089
+ "bits": 6,
1090
+ "group_size": 64,
1091
+ "mode": "affine"
1092
+ },
1093
+ "model.layers.25.mlp.gate": {
1094
+ "bits": 8,
1095
+ "group_size": 64,
1096
+ "mode": "affine"
1097
+ },
1098
+ "model.layers.25.linear_attn.out_proj": {
1099
+ "bits": 6,
1100
+ "group_size": 64,
1101
+ "mode": "affine"
1102
+ },
1103
+ "model.layers.25.linear_attn.in_proj_qkvz": {
1104
+ "bits": 6,
1105
+ "group_size": 64,
1106
+ "mode": "affine"
1107
+ },
1108
+ "model.layers.25.mlp.shared_expert.gate_proj": {
1109
+ "bits": 6,
1110
+ "group_size": 64,
1111
+ "mode": "affine"
1112
+ },
1113
+ "model.layers.25.linear_attn.in_proj_ba": {
1114
+ "bits": 6,
1115
+ "group_size": 64,
1116
+ "mode": "affine"
1117
+ },
1118
+ "model.layers.26.mlp.shared_expert_gate": {
1119
+ "bits": 8,
1120
+ "group_size": 64,
1121
+ "mode": "affine"
1122
+ },
1123
+ "model.layers.26.mlp.shared_expert.up_proj": {
1124
+ "bits": 5,
1125
+ "group_size": 64,
1126
+ "mode": "affine"
1127
+ },
1128
+ "model.layers.26.mlp.shared_expert.gate_proj": {
1129
+ "bits": 5,
1130
+ "group_size": 64,
1131
+ "mode": "affine"
1132
+ },
1133
+ "model.layers.26.mlp.shared_expert.down_proj": {
1134
+ "bits": 5,
1135
+ "group_size": 64,
1136
+ "mode": "affine"
1137
+ },
1138
+ "model.layers.26.linear_attn.out_proj": {
1139
+ "bits": 5,
1140
+ "group_size": 64,
1141
+ "mode": "affine"
1142
+ },
1143
+ "model.layers.26.linear_attn.in_proj_qkvz": {
1144
+ "bits": 5,
1145
+ "group_size": 64,
1146
+ "mode": "affine"
1147
+ },
1148
+ "model.layers.26.linear_attn.in_proj_ba": {
1149
+ "bits": 5,
1150
+ "group_size": 64,
1151
+ "mode": "affine"
1152
+ },
1153
+ "model.layers.26.mlp.gate": {
1154
+ "bits": 8,
1155
+ "group_size": 64,
1156
+ "mode": "affine"
1157
+ },
1158
+ "model.layers.27.self_attn.v_proj": {
1159
+ "bits": 6,
1160
+ "group_size": 64,
1161
+ "mode": "affine"
1162
+ },
1163
+ "model.layers.27.self_attn.q_proj": {
1164
+ "bits": 6,
1165
+ "group_size": 64,
1166
+ "mode": "affine"
1167
+ },
1168
+ "model.layers.27.self_attn.o_proj": {
1169
+ "bits": 6,
1170
+ "group_size": 64,
1171
+ "mode": "affine"
1172
+ },
1173
+ "model.layers.27.self_attn.k_proj": {
1174
+ "bits": 6,
1175
+ "group_size": 64,
1176
+ "mode": "affine"
1177
+ },
1178
+ "model.layers.27.mlp.shared_expert_gate": {
1179
+ "bits": 8,
1180
+ "group_size": 64,
1181
+ "mode": "affine"
1182
+ },
1183
+ "model.layers.27.mlp.shared_expert.up_proj": {
1184
+ "bits": 6,
1185
+ "group_size": 64,
1186
+ "mode": "affine"
1187
+ },
1188
+ "model.layers.27.mlp.shared_expert.gate_proj": {
1189
+ "bits": 6,
1190
+ "group_size": 64,
1191
+ "mode": "affine"
1192
+ },
1193
+ "model.layers.27.mlp.gate": {
1194
+ "bits": 8,
1195
+ "group_size": 64,
1196
+ "mode": "affine"
1197
+ },
1198
+ "model.layers.27.mlp.shared_expert.down_proj": {
1199
+ "bits": 6,
1200
+ "group_size": 64,
1201
+ "mode": "affine"
1202
+ },
1203
+ "model.layers.28.mlp.shared_expert.up_proj": {
1204
+ "bits": 5,
1205
+ "group_size": 64,
1206
+ "mode": "affine"
1207
+ },
1208
+ "model.layers.28.mlp.shared_expert.down_proj": {
1209
+ "bits": 5,
1210
+ "group_size": 64,
1211
+ "mode": "affine"
1212
+ },
1213
+ "model.layers.28.mlp.gate": {
1214
+ "bits": 8,
1215
+ "group_size": 64,
1216
+ "mode": "affine"
1217
+ },
1218
+ "model.layers.28.linear_attn.out_proj": {
1219
+ "bits": 5,
1220
+ "group_size": 64,
1221
+ "mode": "affine"
1222
+ },
1223
+ "model.layers.28.linear_attn.in_proj_qkvz": {
1224
+ "bits": 5,
1225
+ "group_size": 64,
1226
+ "mode": "affine"
1227
+ },
1228
+ "model.layers.28.linear_attn.in_proj_ba": {
1229
+ "bits": 5,
1230
+ "group_size": 64,
1231
+ "mode": "affine"
1232
+ },
1233
+ "model.layers.28.mlp.shared_expert.gate_proj": {
1234
+ "bits": 5,
1235
+ "group_size": 64,
1236
+ "mode": "affine"
1237
+ },
1238
+ "model.layers.28.mlp.shared_expert_gate": {
1239
+ "bits": 8,
1240
+ "group_size": 64,
1241
+ "mode": "affine"
1242
+ },
1243
+ "model.layers.30.linear_attn.in_proj_qkvz": {
1244
+ "bits": 5,
1245
+ "group_size": 64,
1246
+ "mode": "affine"
1247
+ },
1248
+ "model.layers.30.linear_attn.in_proj_ba": {
1249
+ "bits": 5,
1250
+ "group_size": 64,
1251
+ "mode": "affine"
1252
+ },
1253
+ "model.layers.29.mlp.shared_expert_gate": {
1254
+ "bits": 8,
1255
+ "group_size": 64,
1256
+ "mode": "affine"
1257
+ },
1258
+ "model.layers.29.mlp.shared_expert.up_proj": {
1259
+ "bits": 5,
1260
+ "group_size": 64,
1261
+ "mode": "affine"
1262
+ },
1263
+ "model.layers.29.mlp.shared_expert.down_proj": {
1264
+ "bits": 5,
1265
+ "group_size": 64,
1266
+ "mode": "affine"
1267
+ },
1268
+ "model.layers.29.mlp.gate": {
1269
+ "bits": 8,
1270
+ "group_size": 64,
1271
+ "mode": "affine"
1272
+ },
1273
+ "model.layers.30.mlp.gate": {
1274
+ "bits": 8,
1275
+ "group_size": 64,
1276
+ "mode": "affine"
1277
+ },
1278
+ "model.layers.29.mlp.shared_expert.gate_proj": {
1279
+ "bits": 5,
1280
+ "group_size": 64,
1281
+ "mode": "affine"
1282
+ },
1283
+ "model.layers.30.mlp.shared_expert.gate_proj": {
1284
+ "bits": 5,
1285
+ "group_size": 64,
1286
+ "mode": "affine"
1287
+ },
1288
+ "model.layers.30.mlp.shared_expert.up_proj": {
1289
+ "bits": 5,
1290
+ "group_size": 64,
1291
+ "mode": "affine"
1292
+ },
1293
+ "model.layers.29.linear_attn.out_proj": {
1294
+ "bits": 5,
1295
+ "group_size": 64,
1296
+ "mode": "affine"
1297
+ },
1298
+ "model.layers.30.mlp.shared_expert.down_proj": {
1299
+ "bits": 5,
1300
+ "group_size": 64,
1301
+ "mode": "affine"
1302
+ },
1303
+ "model.layers.30.linear_attn.out_proj": {
1304
+ "bits": 5,
1305
+ "group_size": 64,
1306
+ "mode": "affine"
1307
+ },
1308
+ "model.layers.30.mlp.shared_expert_gate": {
1309
+ "bits": 8,
1310
+ "group_size": 64,
1311
+ "mode": "affine"
1312
+ },
1313
+ "model.layers.29.linear_attn.in_proj_qkvz": {
1314
+ "bits": 5,
1315
+ "group_size": 64,
1316
+ "mode": "affine"
1317
+ },
1318
+ "model.layers.29.linear_attn.in_proj_ba": {
1319
+ "bits": 5,
1320
+ "group_size": 64,
1321
+ "mode": "affine"
1322
+ },
1323
+ "model.layers.31.self_attn.v_proj": {
1324
+ "bits": 5,
1325
+ "group_size": 64,
1326
+ "mode": "affine"
1327
+ },
1328
+ "model.layers.31.self_attn.q_proj": {
1329
+ "bits": 5,
1330
+ "group_size": 64,
1331
+ "mode": "affine"
1332
+ },
1333
+ "model.layers.31.self_attn.k_proj": {
1334
+ "bits": 5,
1335
+ "group_size": 64,
1336
+ "mode": "affine"
1337
+ },
1338
+ "model.layers.31.mlp.shared_expert_gate": {
1339
+ "bits": 8,
1340
+ "group_size": 64,
1341
+ "mode": "affine"
1342
+ },
1343
+ "model.layers.31.mlp.shared_expert.up_proj": {
1344
+ "bits": 5,
1345
+ "group_size": 64,
1346
+ "mode": "affine"
1347
+ },
1348
+ "model.layers.31.mlp.shared_expert.gate_proj": {
1349
+ "bits": 5,
1350
+ "group_size": 64,
1351
+ "mode": "affine"
1352
+ },
1353
+ "model.layers.31.mlp.shared_expert.down_proj": {
1354
+ "bits": 5,
1355
+ "group_size": 64,
1356
+ "mode": "affine"
1357
+ },
1358
+ "model.layers.31.mlp.gate": {
1359
+ "bits": 8,
1360
+ "group_size": 64,
1361
+ "mode": "affine"
1362
+ },
1363
+ "model.layers.31.self_attn.o_proj": {
1364
+ "bits": 5,
1365
+ "group_size": 64,
1366
+ "mode": "affine"
1367
+ },
1368
+ "model.layers.32.mlp.shared_expert_gate": {
1369
+ "bits": 8,
1370
+ "group_size": 64,
1371
+ "mode": "affine"
1372
+ },
1373
+ "model.layers.32.mlp.shared_expert.down_proj": {
1374
+ "bits": 5,
1375
+ "group_size": 64,
1376
+ "mode": "affine"
1377
+ },
1378
+ "model.layers.32.linear_attn.out_proj": {
1379
+ "bits": 5,
1380
+ "group_size": 64,
1381
+ "mode": "affine"
1382
+ },
1383
+ "model.layers.32.linear_attn.in_proj_qkvz": {
1384
+ "bits": 5,
1385
+ "group_size": 64,
1386
+ "mode": "affine"
1387
+ },
1388
+ "model.layers.32.linear_attn.in_proj_ba": {
1389
+ "bits": 5,
1390
+ "group_size": 64,
1391
+ "mode": "affine"
1392
+ },
1393
+ "model.layers.32.mlp.shared_expert.up_proj": {
1394
+ "bits": 5,
1395
+ "group_size": 64,
1396
+ "mode": "affine"
1397
+ },
1398
+ "model.layers.32.mlp.gate": {
1399
+ "bits": 8,
1400
+ "group_size": 64,
1401
+ "mode": "affine"
1402
+ },
1403
+ "model.layers.32.mlp.shared_expert.gate_proj": {
1404
+ "bits": 5,
1405
+ "group_size": 64,
1406
+ "mode": "affine"
1407
+ },
1408
+ "model.layers.33.mlp.shared_expert_gate": {
1409
+ "bits": 8,
1410
+ "group_size": 64,
1411
+ "mode": "affine"
1412
+ },
1413
+ "model.layers.33.mlp.shared_expert.up_proj": {
1414
+ "bits": 5,
1415
+ "group_size": 64,
1416
+ "mode": "affine"
1417
+ },
1418
+ "model.layers.33.mlp.shared_expert.gate_proj": {
1419
+ "bits": 5,
1420
+ "group_size": 64,
1421
+ "mode": "affine"
1422
+ },
1423
+ "model.layers.33.mlp.shared_expert.down_proj": {
1424
+ "bits": 5,
1425
+ "group_size": 64,
1426
+ "mode": "affine"
1427
+ },
1428
+ "model.layers.33.mlp.gate": {
1429
+ "bits": 8,
1430
+ "group_size": 64,
1431
+ "mode": "affine"
1432
+ },
1433
+ "model.layers.33.linear_attn.out_proj": {
1434
+ "bits": 5,
1435
+ "group_size": 64,
1436
+ "mode": "affine"
1437
+ },
1438
+ "model.layers.33.linear_attn.in_proj_qkvz": {
1439
+ "bits": 5,
1440
+ "group_size": 64,
1441
+ "mode": "affine"
1442
+ },
1443
+ "model.layers.33.linear_attn.in_proj_ba": {
1444
+ "bits": 5,
1445
+ "group_size": 64,
1446
+ "mode": "affine"
1447
+ },
1448
+ "model.layers.35.self_attn.o_proj": {
1449
+ "bits": 6,
1450
+ "group_size": 64,
1451
+ "mode": "affine"
1452
+ },
1453
+ "model.layers.35.mlp.shared_expert_gate": {
1454
+ "bits": 8,
1455
+ "group_size": 64,
1456
+ "mode": "affine"
1457
+ },
1458
+ "model.layers.35.mlp.gate": {
1459
+ "bits": 8,
1460
+ "group_size": 64,
1461
+ "mode": "affine"
1462
+ },
1463
+ "model.layers.34.mlp.shared_expert_gate": {
1464
+ "bits": 8,
1465
+ "group_size": 64,
1466
+ "mode": "affine"
1467
+ },
1468
+ "model.layers.34.mlp.shared_expert.up_proj": {
1469
+ "bits": 5,
1470
+ "group_size": 64,
1471
+ "mode": "affine"
1472
+ },
1473
+ "model.layers.34.mlp.shared_expert.gate_proj": {
1474
+ "bits": 5,
1475
+ "group_size": 64,
1476
+ "mode": "affine"
1477
+ },
1478
+ "model.layers.35.mlp.shared_expert.gate_proj": {
1479
+ "bits": 6,
1480
+ "group_size": 64,
1481
+ "mode": "affine"
1482
+ },
1483
+ "model.layers.35.mlp.shared_expert.up_proj": {
1484
+ "bits": 6,
1485
+ "group_size": 64,
1486
+ "mode": "affine"
1487
+ },
1488
+ "model.layers.34.linear_attn.in_proj_ba": {
1489
+ "bits": 5,
1490
+ "group_size": 64,
1491
+ "mode": "affine"
1492
+ },
1493
+ "model.layers.35.self_attn.k_proj": {
1494
+ "bits": 6,
1495
+ "group_size": 64,
1496
+ "mode": "affine"
1497
+ },
1498
+ "model.layers.35.self_attn.v_proj": {
1499
+ "bits": 6,
1500
+ "group_size": 64,
1501
+ "mode": "affine"
1502
+ },
1503
+ "model.layers.34.linear_attn.out_proj": {
1504
+ "bits": 5,
1505
+ "group_size": 64,
1506
+ "mode": "affine"
1507
+ },
1508
+ "model.layers.35.mlp.shared_expert.down_proj": {
1509
+ "bits": 6,
1510
+ "group_size": 64,
1511
+ "mode": "affine"
1512
+ },
1513
+ "model.layers.34.linear_attn.in_proj_qkvz": {
1514
+ "bits": 5,
1515
+ "group_size": 64,
1516
+ "mode": "affine"
1517
+ },
1518
+ "model.layers.34.mlp.gate": {
1519
+ "bits": 8,
1520
+ "group_size": 64,
1521
+ "mode": "affine"
1522
+ },
1523
+ "model.layers.34.mlp.shared_expert.down_proj": {
1524
+ "bits": 5,
1525
+ "group_size": 64,
1526
+ "mode": "affine"
1527
+ },
1528
+ "model.layers.35.self_attn.q_proj": {
1529
+ "bits": 6,
1530
+ "group_size": 64,
1531
+ "mode": "affine"
1532
+ },
1533
+ "model.layers.36.mlp.shared_expert_gate": {
1534
+ "bits": 8,
1535
+ "group_size": 64,
1536
+ "mode": "affine"
1537
+ },
1538
+ "model.layers.36.mlp.shared_expert.up_proj": {
1539
+ "bits": 6,
1540
+ "group_size": 64,
1541
+ "mode": "affine"
1542
+ },
1543
+ "model.layers.36.mlp.shared_expert.gate_proj": {
1544
+ "bits": 6,
1545
+ "group_size": 64,
1546
+ "mode": "affine"
1547
+ },
1548
+ "model.layers.36.linear_attn.in_proj_ba": {
1549
+ "bits": 6,
1550
+ "group_size": 64,
1551
+ "mode": "affine"
1552
+ },
1553
+ "model.layers.36.linear_attn.in_proj_qkvz": {
1554
+ "bits": 6,
1555
+ "group_size": 64,
1556
+ "mode": "affine"
1557
+ },
1558
+ "model.layers.36.linear_attn.out_proj": {
1559
+ "bits": 6,
1560
+ "group_size": 64,
1561
+ "mode": "affine"
1562
+ },
1563
+ "model.layers.36.mlp.shared_expert.down_proj": {
1564
+ "bits": 6,
1565
+ "group_size": 64,
1566
+ "mode": "affine"
1567
+ },
1568
+ "model.layers.36.mlp.gate": {
1569
+ "bits": 8,
1570
+ "group_size": 64,
1571
+ "mode": "affine"
1572
+ },
1573
+ "model.layers.37.mlp.shared_expert_gate": {
1574
+ "bits": 8,
1575
+ "group_size": 64,
1576
+ "mode": "affine"
1577
+ },
1578
+ "model.layers.37.mlp.shared_expert.up_proj": {
1579
+ "bits": 5,
1580
+ "group_size": 64,
1581
+ "mode": "affine"
1582
+ },
1583
+ "model.layers.37.mlp.shared_expert.gate_proj": {
1584
+ "bits": 5,
1585
+ "group_size": 64,
1586
+ "mode": "affine"
1587
+ },
1588
+ "model.layers.37.mlp.gate": {
1589
+ "bits": 8,
1590
+ "group_size": 64,
1591
+ "mode": "affine"
1592
+ },
1593
+ "model.layers.37.linear_attn.out_proj": {
1594
+ "bits": 5,
1595
+ "group_size": 64,
1596
+ "mode": "affine"
1597
+ },
1598
+ "model.layers.37.linear_attn.in_proj_qkvz": {
1599
+ "bits": 5,
1600
+ "group_size": 64,
1601
+ "mode": "affine"
1602
+ },
1603
+ "model.layers.37.linear_attn.in_proj_ba": {
1604
+ "bits": 5,
1605
+ "group_size": 64,
1606
+ "mode": "affine"
1607
+ },
1608
+ "model.layers.37.mlp.shared_expert.down_proj": {
1609
+ "bits": 5,
1610
+ "group_size": 64,
1611
+ "mode": "affine"
1612
+ },
1613
+ "model.layers.38.mlp.shared_expert.down_proj": {
1614
+ "bits": 5,
1615
+ "group_size": 64,
1616
+ "mode": "affine"
1617
+ },
1618
+ "model.layers.38.mlp.gate": {
1619
+ "bits": 8,
1620
+ "group_size": 64,
1621
+ "mode": "affine"
1622
+ },
1623
+ "model.layers.38.linear_attn.out_proj": {
1624
+ "bits": 5,
1625
+ "group_size": 64,
1626
+ "mode": "affine"
1627
+ },
1628
+ "model.layers.38.mlp.shared_expert.gate_proj": {
1629
+ "bits": 5,
1630
+ "group_size": 64,
1631
+ "mode": "affine"
1632
+ },
1633
+ "model.layers.38.mlp.shared_expert_gate": {
1634
+ "bits": 8,
1635
+ "group_size": 64,
1636
+ "mode": "affine"
1637
+ },
1638
+ "model.layers.38.linear_attn.in_proj_qkvz": {
1639
+ "bits": 5,
1640
+ "group_size": 64,
1641
+ "mode": "affine"
1642
+ },
1643
+ "model.layers.38.linear_attn.in_proj_ba": {
1644
+ "bits": 5,
1645
+ "group_size": 64,
1646
+ "mode": "affine"
1647
+ },
1648
+ "model.layers.38.mlp.shared_expert.up_proj": {
1649
+ "bits": 5,
1650
+ "group_size": 64,
1651
+ "mode": "affine"
1652
+ },
1653
+ "model.layers.39.self_attn.v_proj": {
1654
+ "bits": 6,
1655
+ "group_size": 64,
1656
+ "mode": "affine"
1657
+ },
1658
+ "model.layers.39.self_attn.o_proj": {
1659
+ "bits": 6,
1660
+ "group_size": 64,
1661
+ "mode": "affine"
1662
+ },
1663
+ "model.layers.39.self_attn.k_proj": {
1664
+ "bits": 6,
1665
+ "group_size": 64,
1666
+ "mode": "affine"
1667
+ },
1668
+ "model.layers.39.mlp.shared_expert_gate": {
1669
+ "bits": 8,
1670
+ "group_size": 64,
1671
+ "mode": "affine"
1672
+ },
1673
+ "model.layers.39.mlp.shared_expert.gate_proj": {
1674
+ "bits": 6,
1675
+ "group_size": 64,
1676
+ "mode": "affine"
1677
+ },
1678
+ "model.layers.39.mlp.shared_expert.down_proj": {
1679
+ "bits": 6,
1680
+ "group_size": 64,
1681
+ "mode": "affine"
1682
+ },
1683
+ "model.layers.39.mlp.gate": {
1684
+ "bits": 8,
1685
+ "group_size": 64,
1686
+ "mode": "affine"
1687
+ },
1688
+ "model.layers.39.self_attn.q_proj": {
1689
+ "bits": 6,
1690
+ "group_size": 64,
1691
+ "mode": "affine"
1692
+ },
1693
+ "model.layers.39.mlp.shared_expert.up_proj": {
1694
+ "bits": 6,
1695
+ "group_size": 64,
1696
+ "mode": "affine"
1697
+ },
1698
+ "model.layers.41.mlp.shared_expert.up_proj": {
1699
+ "bits": 6,
1700
+ "group_size": 64,
1701
+ "mode": "affine"
1702
+ },
1703
+ "model.layers.41.mlp.shared_expert.gate_proj": {
1704
+ "bits": 6,
1705
+ "group_size": 64,
1706
+ "mode": "affine"
1707
+ },
1708
+ "model.layers.41.mlp.gate": {
1709
+ "bits": 8,
1710
+ "group_size": 64,
1711
+ "mode": "affine"
1712
+ },
1713
+ "model.layers.41.linear_attn.out_proj": {
1714
+ "bits": 6,
1715
+ "group_size": 64,
1716
+ "mode": "affine"
1717
+ },
1718
+ "model.layers.40.mlp.shared_expert.down_proj": {
1719
+ "bits": 6,
1720
+ "group_size": 64,
1721
+ "mode": "affine"
1722
+ },
1723
+ "model.layers.40.mlp.gate": {
1724
+ "bits": 8,
1725
+ "group_size": 64,
1726
+ "mode": "affine"
1727
+ },
1728
+ "model.layers.41.linear_attn.in_proj_qkvz": {
1729
+ "bits": 6,
1730
+ "group_size": 64,
1731
+ "mode": "affine"
1732
+ },
1733
+ "model.layers.40.mlp.shared_expert.gate_proj": {
1734
+ "bits": 6,
1735
+ "group_size": 64,
1736
+ "mode": "affine"
1737
+ },
1738
+ "model.layers.41.linear_attn.in_proj_ba": {
1739
+ "bits": 6,
1740
+ "group_size": 64,
1741
+ "mode": "affine"
1742
+ },
1743
+ "model.layers.40.linear_attn.out_proj": {
1744
+ "bits": 6,
1745
+ "group_size": 64,
1746
+ "mode": "affine"
1747
+ },
1748
+ "model.layers.40.mlp.shared_expert_gate": {
1749
+ "bits": 8,
1750
+ "group_size": 64,
1751
+ "mode": "affine"
1752
+ },
1753
+ "model.layers.41.mlp.shared_expert.down_proj": {
1754
+ "bits": 6,
1755
+ "group_size": 64,
1756
+ "mode": "affine"
1757
+ },
1758
+ "model.layers.40.mlp.shared_expert.up_proj": {
1759
+ "bits": 6,
1760
+ "group_size": 64,
1761
+ "mode": "affine"
1762
+ },
1763
+ "model.layers.41.mlp.shared_expert_gate": {
1764
+ "bits": 8,
1765
+ "group_size": 64,
1766
+ "mode": "affine"
1767
+ },
1768
+ "model.layers.40.linear_attn.in_proj_ba": {
1769
+ "bits": 6,
1770
+ "group_size": 64,
1771
+ "mode": "affine"
1772
+ },
1773
+ "model.layers.40.linear_attn.in_proj_qkvz": {
1774
+ "bits": 6,
1775
+ "group_size": 64,
1776
+ "mode": "affine"
1777
+ },
1778
+ "model.layers.42.mlp.shared_expert.gate_proj": {
1779
+ "bits": 6,
1780
+ "group_size": 64,
1781
+ "mode": "affine"
1782
+ },
1783
+ "model.layers.42.mlp.gate": {
1784
+ "bits": 8,
1785
+ "group_size": 64,
1786
+ "mode": "affine"
1787
+ },
1788
+ "model.layers.42.mlp.shared_expert.down_proj": {
1789
+ "bits": 6,
1790
+ "group_size": 64,
1791
+ "mode": "affine"
1792
+ },
1793
+ "model.layers.42.linear_attn.in_proj_qkvz": {
1794
+ "bits": 6,
1795
+ "group_size": 64,
1796
+ "mode": "affine"
1797
+ },
1798
+ "model.layers.42.linear_attn.in_proj_ba": {
1799
+ "bits": 6,
1800
+ "group_size": 64,
1801
+ "mode": "affine"
1802
+ },
1803
+ "model.layers.42.mlp.shared_expert.up_proj": {
1804
+ "bits": 6,
1805
+ "group_size": 64,
1806
+ "mode": "affine"
1807
+ },
1808
+ "model.layers.42.linear_attn.out_proj": {
1809
+ "bits": 6,
1810
+ "group_size": 64,
1811
+ "mode": "affine"
1812
+ },
1813
+ "model.layers.42.mlp.shared_expert_gate": {
1814
+ "bits": 8,
1815
+ "group_size": 64,
1816
+ "mode": "affine"
1817
+ },
1818
+ "model.layers.43.self_attn.q_proj": {
1819
+ "bits": 8,
1820
+ "group_size": 32,
1821
+ "mode": "mxfp8"
1822
+ },
1823
+ "model.layers.43.self_attn.o_proj": {
1824
+ "bits": 8,
1825
+ "group_size": 32,
1826
+ "mode": "mxfp8"
1827
+ },
1828
+ "model.layers.43.mlp.shared_expert.up_proj": {
1829
+ "bits": 8,
1830
+ "group_size": 32,
1831
+ "mode": "mxfp8"
1832
+ },
1833
+ "model.layers.43.mlp.shared_expert.gate_proj": {
1834
+ "bits": 8,
1835
+ "group_size": 32,
1836
+ "mode": "mxfp8"
1837
+ },
1838
+ "model.layers.43.mlp.shared_expert.down_proj": {
1839
+ "bits": 8,
1840
+ "group_size": 32,
1841
+ "mode": "mxfp8"
1842
+ },
1843
+ "model.layers.43.self_attn.k_proj": {
1844
+ "bits": 8,
1845
+ "group_size": 32,
1846
+ "mode": "mxfp8"
1847
+ },
1848
+ "model.layers.43.self_attn.v_proj": {
1849
+ "bits": 8,
1850
+ "group_size": 32,
1851
+ "mode": "mxfp8"
1852
+ },
1853
+ "model.layers.43.mlp.shared_expert_gate": {
1854
+ "bits": 8,
1855
+ "group_size": 64,
1856
+ "mode": "affine"
1857
+ },
1858
+ "model.layers.43.mlp.gate": {
1859
+ "bits": 8,
1860
+ "group_size": 64,
1861
+ "mode": "affine"
1862
+ },
1863
+ "model.layers.44.mlp.shared_expert_gate": {
1864
+ "bits": 8,
1865
+ "group_size": 64,
1866
+ "mode": "affine"
1867
+ },
1868
+ "model.layers.44.mlp.shared_expert.gate_proj": {
1869
+ "bits": 6,
1870
+ "group_size": 64,
1871
+ "mode": "affine"
1872
+ },
1873
+ "model.layers.44.mlp.shared_expert.down_proj": {
1874
+ "bits": 6,
1875
+ "group_size": 64,
1876
+ "mode": "affine"
1877
+ },
1878
+ "model.layers.44.mlp.gate": {
1879
+ "bits": 8,
1880
+ "group_size": 64,
1881
+ "mode": "affine"
1882
+ },
1883
+ "model.layers.44.linear_attn.in_proj_qkvz": {
1884
+ "bits": 6,
1885
+ "group_size": 64,
1886
+ "mode": "affine"
1887
+ },
1888
+ "model.layers.44.linear_attn.out_proj": {
1889
+ "bits": 6,
1890
+ "group_size": 64,
1891
+ "mode": "affine"
1892
+ },
1893
+ "model.layers.44.linear_attn.in_proj_ba": {
1894
+ "bits": 6,
1895
+ "group_size": 64,
1896
+ "mode": "affine"
1897
+ },
1898
+ "model.layers.44.mlp.shared_expert.up_proj": {
1899
+ "bits": 6,
1900
+ "group_size": 64,
1901
+ "mode": "affine"
1902
+ },
1903
+ "model.layers.45.mlp.shared_expert.down_proj": {
1904
+ "bits": 8,
1905
+ "group_size": 32,
1906
+ "mode": "mxfp8"
1907
+ },
1908
+ "model.layers.45.mlp.gate": {
1909
+ "bits": 8,
1910
+ "group_size": 64,
1911
+ "mode": "affine"
1912
+ },
1913
+ "model.layers.45.mlp.shared_expert.gate_proj": {
1914
+ "bits": 8,
1915
+ "group_size": 32,
1916
+ "mode": "mxfp8"
1917
+ },
1918
+ "model.layers.45.mlp.shared_expert_gate": {
1919
+ "bits": 8,
1920
+ "group_size": 64,
1921
+ "mode": "affine"
1922
+ },
1923
+ "model.layers.45.mlp.shared_expert.up_proj": {
1924
+ "bits": 8,
1925
+ "group_size": 32,
1926
+ "mode": "mxfp8"
1927
+ },
1928
+ "model.layers.45.linear_attn.out_proj": {
1929
+ "bits": 8,
1930
+ "group_size": 32,
1931
+ "mode": "mxfp8"
1932
+ },
1933
+ "model.layers.45.linear_attn.in_proj_ba": {
1934
+ "bits": 8,
1935
+ "group_size": 32,
1936
+ "mode": "mxfp8"
1937
+ },
1938
+ "model.layers.45.linear_attn.in_proj_qkvz": {
1939
+ "bits": 8,
1940
+ "group_size": 32,
1941
+ "mode": "mxfp8"
1942
+ },
1943
+ "model.layers.47.self_attn.v_proj": {
1944
+ "bits": 8,
1945
+ "group_size": 32,
1946
+ "mode": "mxfp8"
1947
+ },
1948
+ "model.layers.47.self_attn.q_proj": {
1949
+ "bits": 8,
1950
+ "group_size": 32,
1951
+ "mode": "mxfp8"
1952
+ },
1953
+ "model.layers.47.self_attn.k_proj": {
1954
+ "bits": 8,
1955
+ "group_size": 32,
1956
+ "mode": "mxfp8"
1957
+ },
1958
+ "model.layers.47.mlp.shared_expert_gate": {
1959
+ "bits": 8,
1960
+ "group_size": 64,
1961
+ "mode": "affine"
1962
+ },
1963
+ "model.layers.47.mlp.shared_expert.up_proj": {
1964
+ "bits": 8,
1965
+ "group_size": 32,
1966
+ "mode": "mxfp8"
1967
+ },
1968
+ "model.layers.46.mlp.shared_expert_gate": {
1969
+ "bits": 8,
1970
+ "group_size": 64,
1971
+ "mode": "affine"
1972
+ },
1973
+ "model.layers.46.mlp.shared_expert.gate_proj": {
1974
+ "bits": 8,
1975
+ "group_size": 32,
1976
+ "mode": "mxfp8"
1977
+ },
1978
+ "model.layers.46.mlp.shared_expert.down_proj": {
1979
+ "bits": 8,
1980
+ "group_size": 32,
1981
+ "mode": "mxfp8"
1982
+ },
1983
+ "model.layers.47.mlp.shared_expert.down_proj": {
1984
+ "bits": 8,
1985
+ "group_size": 32,
1986
+ "mode": "mxfp8"
1987
+ },
1988
+ "model.layers.47.self_attn.o_proj": {
1989
+ "bits": 8,
1990
+ "group_size": 32,
1991
+ "mode": "mxfp8"
1992
+ },
1993
+ "model.layers.47.mlp.shared_expert.gate_proj": {
1994
+ "bits": 8,
1995
+ "group_size": 32,
1996
+ "mode": "mxfp8"
1997
+ },
1998
+ "model.layers.46.mlp.gate": {
1999
+ "bits": 8,
2000
+ "group_size": 64,
2001
+ "mode": "affine"
2002
+ },
2003
+ "model.layers.46.linear_attn.in_proj_qkvz": {
2004
+ "bits": 8,
2005
+ "group_size": 32,
2006
+ "mode": "mxfp8"
2007
+ },
2008
+ "model.layers.46.linear_attn.in_proj_ba": {
2009
+ "bits": 8,
2010
+ "group_size": 32,
2011
+ "mode": "mxfp8"
2012
+ },
2013
+ "model.layers.46.linear_attn.out_proj": {
2014
+ "bits": 8,
2015
+ "group_size": 32,
2016
+ "mode": "mxfp8"
2017
+ },
2018
+ "model.layers.47.mlp.gate": {
2019
+ "bits": 8,
2020
+ "group_size": 64,
2021
+ "mode": "affine"
2022
+ },
2023
+ "model.layers.46.mlp.shared_expert.up_proj": {
2024
+ "bits": 8,
2025
+ "group_size": 32,
2026
+ "mode": "mxfp8"
2027
+ },
2028
+ "lm_head": {
2029
+ "bits": 8,
2030
+ "group_size": 32,
2031
+ "mode": "mxfp8"
2032
+ }
2033
+ },
2034
+ "quantization_config": {
2035
+ "group_size": 64,
2036
+ "bits": 4,
2037
+ "mode": "affine",
2038
+ "model.layers.1.mlp.shared_expert_gate": {
2039
+ "bits": 8,
2040
+ "group_size": 64,
2041
+ "mode": "affine"
2042
+ },
2043
+ "model.layers.1.mlp.shared_expert.up_proj": {
2044
+ "bits": 6,
2045
+ "group_size": 64,
2046
+ "mode": "affine"
2047
+ },
2048
+ "model.layers.1.mlp.shared_expert.gate_proj": {
2049
+ "bits": 6,
2050
+ "group_size": 64,
2051
+ "mode": "affine"
2052
+ },
2053
+ "model.layers.1.linear_attn.in_proj_qkvz": {
2054
+ "bits": 6,
2055
+ "group_size": 64,
2056
+ "mode": "affine"
2057
+ },
2058
+ "model.layers.0.mlp.shared_expert.up_proj": {
2059
+ "bits": 6,
2060
+ "group_size": 64,
2061
+ "mode": "affine"
2062
+ },
2063
+ "model.layers.0.mlp.shared_expert.gate_proj": {
2064
+ "bits": 6,
2065
+ "group_size": 64,
2066
+ "mode": "affine"
2067
+ },
2068
+ "model.layers.0.mlp.gate": {
2069
+ "bits": 8,
2070
+ "group_size": 64,
2071
+ "mode": "affine"
2072
+ },
2073
+ "model.layers.1.mlp.shared_expert.down_proj": {
2074
+ "bits": 6,
2075
+ "group_size": 64,
2076
+ "mode": "affine"
2077
+ },
2078
+ "model.layers.1.mlp.gate": {
2079
+ "bits": 8,
2080
+ "group_size": 64,
2081
+ "mode": "affine"
2082
+ },
2083
+ "model.layers.1.linear_attn.in_proj_ba": {
2084
+ "bits": 6,
2085
+ "group_size": 64,
2086
+ "mode": "affine"
2087
+ },
2088
+ "model.layers.0.mlp.shared_expert_gate": {
2089
+ "bits": 8,
2090
+ "group_size": 64,
2091
+ "mode": "affine"
2092
+ },
2093
+ "model.layers.0.linear_attn.in_proj_ba": {
2094
+ "bits": 6,
2095
+ "group_size": 64,
2096
+ "mode": "affine"
2097
+ },
2098
+ "model.layers.0.linear_attn.out_proj": {
2099
+ "bits": 6,
2100
+ "group_size": 64,
2101
+ "mode": "affine"
2102
+ },
2103
+ "model.layers.1.linear_attn.out_proj": {
2104
+ "bits": 6,
2105
+ "group_size": 64,
2106
+ "mode": "affine"
2107
+ },
2108
+ "model.layers.0.mlp.shared_expert.down_proj": {
2109
+ "bits": 6,
2110
+ "group_size": 64,
2111
+ "mode": "affine"
2112
+ },
2113
+ "model.layers.0.linear_attn.in_proj_qkvz": {
2114
+ "bits": 6,
2115
+ "group_size": 64,
2116
+ "mode": "affine"
2117
+ },
2118
+ "model.layers.2.mlp.shared_expert_gate": {
2119
+ "bits": 8,
2120
+ "group_size": 64,
2121
+ "mode": "affine"
2122
+ },
2123
+ "model.layers.2.mlp.gate": {
2124
+ "bits": 8,
2125
+ "group_size": 64,
2126
+ "mode": "affine"
2127
+ },
2128
+ "model.layers.2.mlp.shared_expert.up_proj": {
2129
+ "bits": 6,
2130
+ "group_size": 64,
2131
+ "mode": "affine"
2132
+ },
2133
+ "model.layers.2.linear_attn.out_proj": {
2134
+ "bits": 6,
2135
+ "group_size": 64,
2136
+ "mode": "affine"
2137
+ },
2138
+ "model.layers.2.linear_attn.in_proj_ba": {
2139
+ "bits": 6,
2140
+ "group_size": 64,
2141
+ "mode": "affine"
2142
+ },
2143
+ "model.layers.2.linear_attn.in_proj_qkvz": {
2144
+ "bits": 6,
2145
+ "group_size": 64,
2146
+ "mode": "affine"
2147
+ },
2148
+ "model.layers.2.mlp.shared_expert.gate_proj": {
2149
+ "bits": 6,
2150
+ "group_size": 64,
2151
+ "mode": "affine"
2152
+ },
2153
+ "model.layers.2.mlp.shared_expert.down_proj": {
2154
+ "bits": 6,
2155
+ "group_size": 64,
2156
+ "mode": "affine"
2157
+ },
2158
+ "model.layers.3.self_attn.v_proj": {
2159
+ "bits": 8,
2160
+ "group_size": 32,
2161
+ "mode": "mxfp8"
2162
+ },
2163
+ "model.layers.3.self_attn.q_proj": {
2164
+ "bits": 8,
2165
+ "group_size": 32,
2166
+ "mode": "mxfp8"
2167
+ },
2168
+ "model.layers.3.self_attn.o_proj": {
2169
+ "bits": 8,
2170
+ "group_size": 32,
2171
+ "mode": "mxfp8"
2172
+ },
2173
+ "model.layers.3.mlp.gate": {
2174
+ "bits": 8,
2175
+ "group_size": 64,
2176
+ "mode": "affine"
2177
+ },
2178
+ "model.layers.3.mlp.shared_expert.gate_proj": {
2179
+ "bits": 8,
2180
+ "group_size": 32,
2181
+ "mode": "mxfp8"
2182
+ },
2183
+ "model.layers.3.mlp.shared_expert.up_proj": {
2184
+ "bits": 8,
2185
+ "group_size": 32,
2186
+ "mode": "mxfp8"
2187
+ },
2188
+ "model.layers.3.mlp.shared_expert_gate": {
2189
+ "bits": 8,
2190
+ "group_size": 64,
2191
+ "mode": "affine"
2192
+ },
2193
+ "model.layers.3.mlp.shared_expert.down_proj": {
2194
+ "bits": 8,
2195
+ "group_size": 32,
2196
+ "mode": "mxfp8"
2197
+ },
2198
+ "model.layers.3.self_attn.k_proj": {
2199
+ "bits": 8,
2200
+ "group_size": 32,
2201
+ "mode": "mxfp8"
2202
+ },
2203
+ "model.layers.4.mlp.shared_expert.up_proj": {
2204
+ "bits": 6,
2205
+ "group_size": 64,
2206
+ "mode": "affine"
2207
+ },
2208
+ "model.layers.4.mlp.shared_expert.gate_proj": {
2209
+ "bits": 6,
2210
+ "group_size": 64,
2211
+ "mode": "affine"
2212
+ },
2213
+ "model.layers.4.mlp.shared_expert.down_proj": {
2214
+ "bits": 6,
2215
+ "group_size": 64,
2216
+ "mode": "affine"
2217
+ },
2218
+ "model.layers.4.mlp.gate": {
2219
+ "bits": 8,
2220
+ "group_size": 64,
2221
+ "mode": "affine"
2222
+ },
2223
+ "model.layers.4.linear_attn.in_proj_qkvz": {
2224
+ "bits": 6,
2225
+ "group_size": 64,
2226
+ "mode": "affine"
2227
+ },
2228
+ "model.layers.4.linear_attn.in_proj_ba": {
2229
+ "bits": 6,
2230
+ "group_size": 64,
2231
+ "mode": "affine"
2232
+ },
2233
+ "model.layers.4.linear_attn.out_proj": {
2234
+ "bits": 6,
2235
+ "group_size": 64,
2236
+ "mode": "affine"
2237
+ },
2238
+ "model.layers.4.mlp.shared_expert_gate": {
2239
+ "bits": 8,
2240
+ "group_size": 64,
2241
+ "mode": "affine"
2242
+ },
2243
+ "model.layers.5.mlp.shared_expert.up_proj": {
2244
+ "bits": 5,
2245
+ "group_size": 64,
2246
+ "mode": "affine"
2247
+ },
2248
+ "model.layers.5.mlp.shared_expert.down_proj": {
2249
+ "bits": 5,
2250
+ "group_size": 64,
2251
+ "mode": "affine"
2252
+ },
2253
+ "model.layers.5.mlp.shared_expert.gate_proj": {
2254
+ "bits": 5,
2255
+ "group_size": 64,
2256
+ "mode": "affine"
2257
+ },
2258
+ "model.layers.5.mlp.shared_expert_gate": {
2259
+ "bits": 8,
2260
+ "group_size": 64,
2261
+ "mode": "affine"
2262
+ },
2263
+ "model.layers.5.linear_attn.out_proj": {
2264
+ "bits": 5,
2265
+ "group_size": 64,
2266
+ "mode": "affine"
2267
+ },
2268
+ "model.layers.5.linear_attn.in_proj_qkvz": {
2269
+ "bits": 5,
2270
+ "group_size": 64,
2271
+ "mode": "affine"
2272
+ },
2273
+ "model.layers.5.linear_attn.in_proj_ba": {
2274
+ "bits": 5,
2275
+ "group_size": 64,
2276
+ "mode": "affine"
2277
+ },
2278
+ "model.layers.5.mlp.gate": {
2279
+ "bits": 8,
2280
+ "group_size": 64,
2281
+ "mode": "affine"
2282
+ },
2283
+ "model.layers.7.self_attn.v_proj": {
2284
+ "bits": 6,
2285
+ "group_size": 64,
2286
+ "mode": "affine"
2287
+ },
2288
+ "model.layers.7.self_attn.q_proj": {
2289
+ "bits": 6,
2290
+ "group_size": 64,
2291
+ "mode": "affine"
2292
+ },
2293
+ "model.layers.7.self_attn.o_proj": {
2294
+ "bits": 6,
2295
+ "group_size": 64,
2296
+ "mode": "affine"
2297
+ },
2298
+ "model.layers.7.self_attn.k_proj": {
2299
+ "bits": 6,
2300
+ "group_size": 64,
2301
+ "mode": "affine"
2302
+ },
2303
+ "model.layers.7.mlp.shared_expert.up_proj": {
2304
+ "bits": 6,
2305
+ "group_size": 64,
2306
+ "mode": "affine"
2307
+ },
2308
+ "model.layers.7.mlp.shared_expert.down_proj": {
2309
+ "bits": 6,
2310
+ "group_size": 64,
2311
+ "mode": "affine"
2312
+ },
2313
+ "model.layers.6.mlp.gate": {
2314
+ "bits": 8,
2315
+ "group_size": 64,
2316
+ "mode": "affine"
2317
+ },
2318
+ "model.layers.7.mlp.gate": {
2319
+ "bits": 8,
2320
+ "group_size": 64,
2321
+ "mode": "affine"
2322
+ },
2323
+ "model.layers.6.mlp.shared_expert.gate_proj": {
2324
+ "bits": 6,
2325
+ "group_size": 64,
2326
+ "mode": "affine"
2327
+ },
2328
+ "model.layers.7.mlp.shared_expert_gate": {
2329
+ "bits": 8,
2330
+ "group_size": 64,
2331
+ "mode": "affine"
2332
+ },
2333
+ "model.layers.6.mlp.shared_expert_gate": {
2334
+ "bits": 8,
2335
+ "group_size": 64,
2336
+ "mode": "affine"
2337
+ },
2338
+ "model.layers.6.linear_attn.in_proj_ba": {
2339
+ "bits": 6,
2340
+ "group_size": 64,
2341
+ "mode": "affine"
2342
+ },
2343
+ "model.layers.6.linear_attn.out_proj": {
2344
+ "bits": 6,
2345
+ "group_size": 64,
2346
+ "mode": "affine"
2347
+ },
2348
+ "model.layers.6.mlp.shared_expert.up_proj": {
2349
+ "bits": 6,
2350
+ "group_size": 64,
2351
+ "mode": "affine"
2352
+ },
2353
+ "model.layers.7.mlp.shared_expert.gate_proj": {
2354
+ "bits": 6,
2355
+ "group_size": 64,
2356
+ "mode": "affine"
2357
+ },
2358
+ "model.layers.6.mlp.shared_expert.down_proj": {
2359
+ "bits": 6,
2360
+ "group_size": 64,
2361
+ "mode": "affine"
2362
+ },
2363
+ "model.layers.6.linear_attn.in_proj_qkvz": {
2364
+ "bits": 6,
2365
+ "group_size": 64,
2366
+ "mode": "affine"
2367
+ },
2368
+ "model.layers.8.mlp.shared_expert_gate": {
2369
+ "bits": 8,
2370
+ "group_size": 64,
2371
+ "mode": "affine"
2372
+ },
2373
+ "model.layers.8.mlp.shared_expert.gate_proj": {
2374
+ "bits": 6,
2375
+ "group_size": 64,
2376
+ "mode": "affine"
2377
+ },
2378
+ "model.layers.8.mlp.gate": {
2379
+ "bits": 8,
2380
+ "group_size": 64,
2381
+ "mode": "affine"
2382
+ },
2383
+ "model.layers.8.linear_attn.in_proj_ba": {
2384
+ "bits": 6,
2385
+ "group_size": 64,
2386
+ "mode": "affine"
2387
+ },
2388
+ "model.layers.8.mlp.shared_expert.up_proj": {
2389
+ "bits": 6,
2390
+ "group_size": 64,
2391
+ "mode": "affine"
2392
+ },
2393
+ "model.layers.8.linear_attn.in_proj_qkvz": {
2394
+ "bits": 6,
2395
+ "group_size": 64,
2396
+ "mode": "affine"
2397
+ },
2398
+ "model.layers.8.linear_attn.out_proj": {
2399
+ "bits": 6,
2400
+ "group_size": 64,
2401
+ "mode": "affine"
2402
+ },
2403
+ "model.layers.8.mlp.shared_expert.down_proj": {
2404
+ "bits": 6,
2405
+ "group_size": 64,
2406
+ "mode": "affine"
2407
+ },
2408
+ "model.layers.9.mlp.shared_expert.up_proj": {
2409
+ "bits": 6,
2410
+ "group_size": 64,
2411
+ "mode": "affine"
2412
+ },
2413
+ "model.layers.9.mlp.shared_expert.gate_proj": {
2414
+ "bits": 6,
2415
+ "group_size": 64,
2416
+ "mode": "affine"
2417
+ },
2418
+ "model.layers.9.mlp.shared_expert.down_proj": {
2419
+ "bits": 6,
2420
+ "group_size": 64,
2421
+ "mode": "affine"
2422
+ },
2423
+ "model.layers.9.mlp.gate": {
2424
+ "bits": 8,
2425
+ "group_size": 64,
2426
+ "mode": "affine"
2427
+ },
2428
+ "model.layers.9.linear_attn.in_proj_ba": {
2429
+ "bits": 6,
2430
+ "group_size": 64,
2431
+ "mode": "affine"
2432
+ },
2433
+ "model.layers.9.linear_attn.in_proj_qkvz": {
2434
+ "bits": 6,
2435
+ "group_size": 64,
2436
+ "mode": "affine"
2437
+ },
2438
+ "model.layers.9.mlp.shared_expert_gate": {
2439
+ "bits": 8,
2440
+ "group_size": 64,
2441
+ "mode": "affine"
2442
+ },
2443
+ "model.layers.9.linear_attn.out_proj": {
2444
+ "bits": 6,
2445
+ "group_size": 64,
2446
+ "mode": "affine"
2447
+ },
2448
+ "model.layers.10.mlp.shared_expert_gate": {
2449
+ "bits": 8,
2450
+ "group_size": 64,
2451
+ "mode": "affine"
2452
+ },
2453
+ "model.layers.10.mlp.shared_expert.gate_proj": {
2454
+ "bits": 6,
2455
+ "group_size": 64,
2456
+ "mode": "affine"
2457
+ },
2458
+ "model.layers.10.mlp.shared_expert.down_proj": {
2459
+ "bits": 6,
2460
+ "group_size": 64,
2461
+ "mode": "affine"
2462
+ },
2463
+ "model.layers.10.mlp.gate": {
2464
+ "bits": 8,
2465
+ "group_size": 64,
2466
+ "mode": "affine"
2467
+ },
2468
+ "model.layers.10.linear_attn.in_proj_ba": {
2469
+ "bits": 6,
2470
+ "group_size": 64,
2471
+ "mode": "affine"
2472
+ },
2473
+ "model.layers.10.mlp.shared_expert.up_proj": {
2474
+ "bits": 6,
2475
+ "group_size": 64,
2476
+ "mode": "affine"
2477
+ },
2478
+ "model.layers.10.linear_attn.in_proj_qkvz": {
2479
+ "bits": 6,
2480
+ "group_size": 64,
2481
+ "mode": "affine"
2482
+ },
2483
+ "model.layers.10.linear_attn.out_proj": {
2484
+ "bits": 6,
2485
+ "group_size": 64,
2486
+ "mode": "affine"
2487
+ },
2488
+ "model.layers.11.self_attn.v_proj": {
2489
+ "bits": 6,
2490
+ "group_size": 64,
2491
+ "mode": "affine"
2492
+ },
2493
+ "model.layers.11.self_attn.o_proj": {
2494
+ "bits": 6,
2495
+ "group_size": 64,
2496
+ "mode": "affine"
2497
+ },
2498
+ "model.layers.11.mlp.shared_expert.gate_proj": {
2499
+ "bits": 6,
2500
+ "group_size": 64,
2501
+ "mode": "affine"
2502
+ },
2503
+ "model.layers.11.mlp.shared_expert.down_proj": {
2504
+ "bits": 6,
2505
+ "group_size": 64,
2506
+ "mode": "affine"
2507
+ },
2508
+ "model.layers.11.mlp.shared_expert.up_proj": {
2509
+ "bits": 6,
2510
+ "group_size": 64,
2511
+ "mode": "affine"
2512
+ },
2513
+ "model.layers.11.mlp.shared_expert_gate": {
2514
+ "bits": 8,
2515
+ "group_size": 64,
2516
+ "mode": "affine"
2517
+ },
2518
+ "model.layers.11.self_attn.k_proj": {
2519
+ "bits": 6,
2520
+ "group_size": 64,
2521
+ "mode": "affine"
2522
+ },
2523
+ "model.layers.11.mlp.gate": {
2524
+ "bits": 8,
2525
+ "group_size": 64,
2526
+ "mode": "affine"
2527
+ },
2528
+ "model.layers.11.self_attn.q_proj": {
2529
+ "bits": 6,
2530
+ "group_size": 64,
2531
+ "mode": "affine"
2532
+ },
2533
+ "model.layers.13.mlp.shared_expert.up_proj": {
2534
+ "bits": 5,
2535
+ "group_size": 64,
2536
+ "mode": "affine"
2537
+ },
2538
+ "model.layers.13.mlp.shared_expert.down_proj": {
2539
+ "bits": 5,
2540
+ "group_size": 64,
2541
+ "mode": "affine"
2542
+ },
2543
+ "model.layers.13.linear_attn.out_proj": {
2544
+ "bits": 5,
2545
+ "group_size": 64,
2546
+ "mode": "affine"
2547
+ },
2548
+ "model.layers.13.linear_attn.in_proj_qkvz": {
2549
+ "bits": 5,
2550
+ "group_size": 64,
2551
+ "mode": "affine"
2552
+ },
2553
+ "model.layers.13.linear_attn.in_proj_ba": {
2554
+ "bits": 5,
2555
+ "group_size": 64,
2556
+ "mode": "affine"
2557
+ },
2558
+ "model.layers.12.mlp.shared_expert_gate": {
2559
+ "bits": 8,
2560
+ "group_size": 64,
2561
+ "mode": "affine"
2562
+ },
2563
+ "model.layers.12.mlp.shared_expert.up_proj": {
2564
+ "bits": 6,
2565
+ "group_size": 64,
2566
+ "mode": "affine"
2567
+ },
2568
+ "model.layers.12.mlp.shared_expert.gate_proj": {
2569
+ "bits": 6,
2570
+ "group_size": 64,
2571
+ "mode": "affine"
2572
+ },
2573
+ "model.layers.12.mlp.shared_expert.down_proj": {
2574
+ "bits": 6,
2575
+ "group_size": 64,
2576
+ "mode": "affine"
2577
+ },
2578
+ "model.layers.12.mlp.gate": {
2579
+ "bits": 8,
2580
+ "group_size": 64,
2581
+ "mode": "affine"
2582
+ },
2583
+ "model.layers.13.mlp.shared_expert_gate": {
2584
+ "bits": 8,
2585
+ "group_size": 64,
2586
+ "mode": "affine"
2587
+ },
2588
+ "model.layers.12.linear_attn.in_proj_ba": {
2589
+ "bits": 6,
2590
+ "group_size": 64,
2591
+ "mode": "affine"
2592
+ },
2593
+ "model.layers.13.mlp.gate": {
2594
+ "bits": 8,
2595
+ "group_size": 64,
2596
+ "mode": "affine"
2597
+ },
2598
+ "model.layers.13.mlp.shared_expert.gate_proj": {
2599
+ "bits": 5,
2600
+ "group_size": 64,
2601
+ "mode": "affine"
2602
+ },
2603
+ "model.layers.12.linear_attn.in_proj_qkvz": {
2604
+ "bits": 6,
2605
+ "group_size": 64,
2606
+ "mode": "affine"
2607
+ },
2608
+ "model.layers.12.linear_attn.out_proj": {
2609
+ "bits": 6,
2610
+ "group_size": 64,
2611
+ "mode": "affine"
2612
+ },
2613
+ "model.layers.14.mlp.shared_expert.up_proj": {
2614
+ "bits": 5,
2615
+ "group_size": 64,
2616
+ "mode": "affine"
2617
+ },
2618
+ "model.layers.14.mlp.shared_expert.gate_proj": {
2619
+ "bits": 5,
2620
+ "group_size": 64,
2621
+ "mode": "affine"
2622
+ },
2623
+ "model.layers.14.linear_attn.in_proj_qkvz": {
2624
+ "bits": 5,
2625
+ "group_size": 64,
2626
+ "mode": "affine"
2627
+ },
2628
+ "model.layers.14.linear_attn.out_proj": {
2629
+ "bits": 5,
2630
+ "group_size": 64,
2631
+ "mode": "affine"
2632
+ },
2633
+ "model.layers.14.mlp.gate": {
2634
+ "bits": 8,
2635
+ "group_size": 64,
2636
+ "mode": "affine"
2637
+ },
2638
+ "model.layers.14.linear_attn.in_proj_ba": {
2639
+ "bits": 5,
2640
+ "group_size": 64,
2641
+ "mode": "affine"
2642
+ },
2643
+ "model.layers.14.mlp.shared_expert_gate": {
2644
+ "bits": 8,
2645
+ "group_size": 64,
2646
+ "mode": "affine"
2647
+ },
2648
+ "model.layers.14.mlp.shared_expert.down_proj": {
2649
+ "bits": 5,
2650
+ "group_size": 64,
2651
+ "mode": "affine"
2652
+ },
2653
+ "model.layers.15.self_attn.o_proj": {
2654
+ "bits": 6,
2655
+ "group_size": 64,
2656
+ "mode": "affine"
2657
+ },
2658
+ "model.layers.15.self_attn.k_proj": {
2659
+ "bits": 6,
2660
+ "group_size": 64,
2661
+ "mode": "affine"
2662
+ },
2663
+ "model.layers.15.mlp.gate": {
2664
+ "bits": 8,
2665
+ "group_size": 64,
2666
+ "mode": "affine"
2667
+ },
2668
+ "model.layers.15.mlp.shared_expert_gate": {
2669
+ "bits": 8,
2670
+ "group_size": 64,
2671
+ "mode": "affine"
2672
+ },
2673
+ "model.layers.15.mlp.shared_expert.down_proj": {
2674
+ "bits": 6,
2675
+ "group_size": 64,
2676
+ "mode": "affine"
2677
+ },
2678
+ "model.layers.15.self_attn.v_proj": {
2679
+ "bits": 6,
2680
+ "group_size": 64,
2681
+ "mode": "affine"
2682
+ },
2683
+ "model.layers.15.mlp.shared_expert.up_proj": {
2684
+ "bits": 6,
2685
+ "group_size": 64,
2686
+ "mode": "affine"
2687
+ },
2688
+ "model.layers.15.self_attn.q_proj": {
2689
+ "bits": 6,
2690
+ "group_size": 64,
2691
+ "mode": "affine"
2692
+ },
2693
+ "model.layers.15.mlp.shared_expert.gate_proj": {
2694
+ "bits": 6,
2695
+ "group_size": 64,
2696
+ "mode": "affine"
2697
+ },
2698
+ "model.layers.16.mlp.shared_expert.up_proj": {
2699
+ "bits": 5,
2700
+ "group_size": 64,
2701
+ "mode": "affine"
2702
+ },
2703
+ "model.layers.16.mlp.shared_expert.down_proj": {
2704
+ "bits": 5,
2705
+ "group_size": 64,
2706
+ "mode": "affine"
2707
+ },
2708
+ "model.layers.16.mlp.gate": {
2709
+ "bits": 8,
2710
+ "group_size": 64,
2711
+ "mode": "affine"
2712
+ },
2713
+ "model.layers.16.mlp.shared_expert_gate": {
2714
+ "bits": 8,
2715
+ "group_size": 64,
2716
+ "mode": "affine"
2717
+ },
2718
+ "model.layers.16.linear_attn.in_proj_ba": {
2719
+ "bits": 5,
2720
+ "group_size": 64,
2721
+ "mode": "affine"
2722
+ },
2723
+ "model.layers.16.linear_attn.in_proj_qkvz": {
2724
+ "bits": 5,
2725
+ "group_size": 64,
2726
+ "mode": "affine"
2727
+ },
2728
+ "model.layers.16.mlp.shared_expert.gate_proj": {
2729
+ "bits": 5,
2730
+ "group_size": 64,
2731
+ "mode": "affine"
2732
+ },
2733
+ "model.layers.16.linear_attn.out_proj": {
2734
+ "bits": 5,
2735
+ "group_size": 64,
2736
+ "mode": "affine"
2737
+ },
2738
+ "model.layers.18.mlp.shared_expert_gate": {
2739
+ "bits": 8,
2740
+ "group_size": 64,
2741
+ "mode": "affine"
2742
+ },
2743
+ "model.layers.18.mlp.shared_expert.up_proj": {
2744
+ "bits": 5,
2745
+ "group_size": 64,
2746
+ "mode": "affine"
2747
+ },
2748
+ "model.layers.18.linear_attn.out_proj": {
2749
+ "bits": 5,
2750
+ "group_size": 64,
2751
+ "mode": "affine"
2752
+ },
2753
+ "model.layers.18.linear_attn.in_proj_qkvz": {
2754
+ "bits": 5,
2755
+ "group_size": 64,
2756
+ "mode": "affine"
2757
+ },
2758
+ "model.layers.18.linear_attn.in_proj_ba": {
2759
+ "bits": 5,
2760
+ "group_size": 64,
2761
+ "mode": "affine"
2762
+ },
2763
+ "model.layers.17.mlp.shared_expert.gate_proj": {
2764
+ "bits": 5,
2765
+ "group_size": 64,
2766
+ "mode": "affine"
2767
+ },
2768
+ "model.layers.17.mlp.shared_expert.down_proj": {
2769
+ "bits": 5,
2770
+ "group_size": 64,
2771
+ "mode": "affine"
2772
+ },
2773
+ "model.layers.17.mlp.gate": {
2774
+ "bits": 8,
2775
+ "group_size": 64,
2776
+ "mode": "affine"
2777
+ },
2778
+ "model.layers.18.mlp.shared_expert.gate_proj": {
2779
+ "bits": 5,
2780
+ "group_size": 64,
2781
+ "mode": "affine"
2782
+ },
2783
+ "model.layers.17.mlp.shared_expert.up_proj": {
2784
+ "bits": 5,
2785
+ "group_size": 64,
2786
+ "mode": "affine"
2787
+ },
2788
+ "model.layers.17.linear_attn.out_proj": {
2789
+ "bits": 5,
2790
+ "group_size": 64,
2791
+ "mode": "affine"
2792
+ },
2793
+ "model.layers.17.mlp.shared_expert_gate": {
2794
+ "bits": 8,
2795
+ "group_size": 64,
2796
+ "mode": "affine"
2797
+ },
2798
+ "model.layers.18.mlp.shared_expert.down_proj": {
2799
+ "bits": 5,
2800
+ "group_size": 64,
2801
+ "mode": "affine"
2802
+ },
2803
+ "model.layers.17.linear_attn.in_proj_ba": {
2804
+ "bits": 5,
2805
+ "group_size": 64,
2806
+ "mode": "affine"
2807
+ },
2808
+ "model.layers.17.linear_attn.in_proj_qkvz": {
2809
+ "bits": 5,
2810
+ "group_size": 64,
2811
+ "mode": "affine"
2812
+ },
2813
+ "model.layers.18.mlp.gate": {
2814
+ "bits": 8,
2815
+ "group_size": 64,
2816
+ "mode": "affine"
2817
+ },
2818
+ "model.layers.19.self_attn.q_proj": {
2819
+ "bits": 5,
2820
+ "group_size": 64,
2821
+ "mode": "affine"
2822
+ },
2823
+ "model.layers.19.self_attn.o_proj": {
2824
+ "bits": 5,
2825
+ "group_size": 64,
2826
+ "mode": "affine"
2827
+ },
2828
+ "model.layers.19.mlp.shared_expert.down_proj": {
2829
+ "bits": 5,
2830
+ "group_size": 64,
2831
+ "mode": "affine"
2832
+ },
2833
+ "model.layers.19.mlp.gate": {
2834
+ "bits": 8,
2835
+ "group_size": 64,
2836
+ "mode": "affine"
2837
+ },
2838
+ "model.layers.19.self_attn.v_proj": {
2839
+ "bits": 5,
2840
+ "group_size": 64,
2841
+ "mode": "affine"
2842
+ },
2843
+ "model.layers.19.mlp.shared_expert.gate_proj": {
2844
+ "bits": 5,
2845
+ "group_size": 64,
2846
+ "mode": "affine"
2847
+ },
2848
+ "model.layers.19.mlp.shared_expert.up_proj": {
2849
+ "bits": 5,
2850
+ "group_size": 64,
2851
+ "mode": "affine"
2852
+ },
2853
+ "model.layers.19.self_attn.k_proj": {
2854
+ "bits": 5,
2855
+ "group_size": 64,
2856
+ "mode": "affine"
2857
+ },
2858
+ "model.layers.19.mlp.shared_expert_gate": {
2859
+ "bits": 8,
2860
+ "group_size": 64,
2861
+ "mode": "affine"
2862
+ },
2863
+ "model.layers.20.mlp.shared_expert.up_proj": {
2864
+ "bits": 6,
2865
+ "group_size": 64,
2866
+ "mode": "affine"
2867
+ },
2868
+ "model.layers.20.mlp.gate": {
2869
+ "bits": 8,
2870
+ "group_size": 64,
2871
+ "mode": "affine"
2872
+ },
2873
+ "model.layers.20.mlp.shared_expert.gate_proj": {
2874
+ "bits": 6,
2875
+ "group_size": 64,
2876
+ "mode": "affine"
2877
+ },
2878
+ "model.layers.20.mlp.shared_expert.down_proj": {
2879
+ "bits": 6,
2880
+ "group_size": 64,
2881
+ "mode": "affine"
2882
+ },
2883
+ "model.layers.20.linear_attn.in_proj_ba": {
2884
+ "bits": 6,
2885
+ "group_size": 64,
2886
+ "mode": "affine"
2887
+ },
2888
+ "model.layers.20.linear_attn.in_proj_qkvz": {
2889
+ "bits": 6,
2890
+ "group_size": 64,
2891
+ "mode": "affine"
2892
+ },
2893
+ "model.layers.20.linear_attn.out_proj": {
2894
+ "bits": 6,
2895
+ "group_size": 64,
2896
+ "mode": "affine"
2897
+ },
2898
+ "model.layers.20.mlp.shared_expert_gate": {
2899
+ "bits": 8,
2900
+ "group_size": 64,
2901
+ "mode": "affine"
2902
+ },
2903
+ "model.layers.21.mlp.shared_expert_gate": {
2904
+ "bits": 8,
2905
+ "group_size": 64,
2906
+ "mode": "affine"
2907
+ },
2908
+ "model.layers.21.mlp.shared_expert.up_proj": {
2909
+ "bits": 5,
2910
+ "group_size": 64,
2911
+ "mode": "affine"
2912
+ },
2913
+ "model.layers.21.mlp.shared_expert.gate_proj": {
2914
+ "bits": 5,
2915
+ "group_size": 64,
2916
+ "mode": "affine"
2917
+ },
2918
+ "model.layers.21.mlp.shared_expert.down_proj": {
2919
+ "bits": 5,
2920
+ "group_size": 64,
2921
+ "mode": "affine"
2922
+ },
2923
+ "model.layers.21.mlp.gate": {
2924
+ "bits": 8,
2925
+ "group_size": 64,
2926
+ "mode": "affine"
2927
+ },
2928
+ "model.layers.21.linear_attn.in_proj_qkvz": {
2929
+ "bits": 5,
2930
+ "group_size": 64,
2931
+ "mode": "affine"
2932
+ },
2933
+ "model.layers.21.linear_attn.in_proj_ba": {
2934
+ "bits": 5,
2935
+ "group_size": 64,
2936
+ "mode": "affine"
2937
+ },
2938
+ "model.layers.21.linear_attn.out_proj": {
2939
+ "bits": 5,
2940
+ "group_size": 64,
2941
+ "mode": "affine"
2942
+ },
2943
+ "model.layers.22.mlp.shared_expert.gate_proj": {
2944
+ "bits": 6,
2945
+ "group_size": 64,
2946
+ "mode": "affine"
2947
+ },
2948
+ "model.layers.22.mlp.shared_expert.down_proj": {
2949
+ "bits": 6,
2950
+ "group_size": 64,
2951
+ "mode": "affine"
2952
+ },
2953
+ "model.layers.22.linear_attn.in_proj_ba": {
2954
+ "bits": 6,
2955
+ "group_size": 64,
2956
+ "mode": "affine"
2957
+ },
2958
+ "model.layers.22.mlp.shared_expert_gate": {
2959
+ "bits": 8,
2960
+ "group_size": 64,
2961
+ "mode": "affine"
2962
+ },
2963
+ "model.layers.22.linear_attn.out_proj": {
2964
+ "bits": 6,
2965
+ "group_size": 64,
2966
+ "mode": "affine"
2967
+ },
2968
+ "model.layers.22.linear_attn.in_proj_qkvz": {
2969
+ "bits": 6,
2970
+ "group_size": 64,
2971
+ "mode": "affine"
2972
+ },
2973
+ "model.layers.22.mlp.shared_expert.up_proj": {
2974
+ "bits": 6,
2975
+ "group_size": 64,
2976
+ "mode": "affine"
2977
+ },
2978
+ "model.layers.22.mlp.gate": {
2979
+ "bits": 8,
2980
+ "group_size": 64,
2981
+ "mode": "affine"
2982
+ },
2983
+ "model.layers.24.mlp.shared_expert_gate": {
2984
+ "bits": 8,
2985
+ "group_size": 64,
2986
+ "mode": "affine"
2987
+ },
2988
+ "model.layers.24.mlp.shared_expert.up_proj": {
2989
+ "bits": 6,
2990
+ "group_size": 64,
2991
+ "mode": "affine"
2992
+ },
2993
+ "model.layers.24.mlp.shared_expert.gate_proj": {
2994
+ "bits": 6,
2995
+ "group_size": 64,
2996
+ "mode": "affine"
2997
+ },
2998
+ "model.layers.24.mlp.shared_expert.down_proj": {
2999
+ "bits": 6,
3000
+ "group_size": 64,
3001
+ "mode": "affine"
3002
+ },
3003
+ "model.layers.24.mlp.gate": {
3004
+ "bits": 8,
3005
+ "group_size": 64,
3006
+ "mode": "affine"
3007
+ },
3008
+ "model.layers.24.linear_attn.in_proj_qkvz": {
3009
+ "bits": 6,
3010
+ "group_size": 64,
3011
+ "mode": "affine"
3012
+ },
3013
+ "model.layers.24.linear_attn.in_proj_ba": {
3014
+ "bits": 6,
3015
+ "group_size": 64,
3016
+ "mode": "affine"
3017
+ },
3018
+ "model.layers.23.self_attn.v_proj": {
3019
+ "bits": 6,
3020
+ "group_size": 64,
3021
+ "mode": "affine"
3022
+ },
3023
+ "model.layers.23.self_attn.q_proj": {
3024
+ "bits": 6,
3025
+ "group_size": 64,
3026
+ "mode": "affine"
3027
+ },
3028
+ "model.layers.23.self_attn.o_proj": {
3029
+ "bits": 6,
3030
+ "group_size": 64,
3031
+ "mode": "affine"
3032
+ },
3033
+ "model.layers.23.mlp.shared_expert.up_proj": {
3034
+ "bits": 6,
3035
+ "group_size": 64,
3036
+ "mode": "affine"
3037
+ },
3038
+ "model.layers.23.mlp.shared_expert.gate_proj": {
3039
+ "bits": 6,
3040
+ "group_size": 64,
3041
+ "mode": "affine"
3042
+ },
3043
+ "model.layers.23.mlp.shared_expert.down_proj": {
3044
+ "bits": 6,
3045
+ "group_size": 64,
3046
+ "mode": "affine"
3047
+ },
3048
+ "model.layers.23.mlp.gate": {
3049
+ "bits": 8,
3050
+ "group_size": 64,
3051
+ "mode": "affine"
3052
+ },
3053
+ "model.layers.23.self_attn.k_proj": {
3054
+ "bits": 6,
3055
+ "group_size": 64,
3056
+ "mode": "affine"
3057
+ },
3058
+ "model.layers.23.mlp.shared_expert_gate": {
3059
+ "bits": 8,
3060
+ "group_size": 64,
3061
+ "mode": "affine"
3062
+ },
3063
+ "model.layers.24.linear_attn.out_proj": {
3064
+ "bits": 6,
3065
+ "group_size": 64,
3066
+ "mode": "affine"
3067
+ },
3068
+ "model.layers.25.mlp.shared_expert_gate": {
3069
+ "bits": 8,
3070
+ "group_size": 64,
3071
+ "mode": "affine"
3072
+ },
3073
+ "model.layers.25.mlp.shared_expert.up_proj": {
3074
+ "bits": 6,
3075
+ "group_size": 64,
3076
+ "mode": "affine"
3077
+ },
3078
+ "model.layers.25.mlp.shared_expert.down_proj": {
3079
+ "bits": 6,
3080
+ "group_size": 64,
3081
+ "mode": "affine"
3082
+ },
3083
+ "model.layers.25.mlp.gate": {
3084
+ "bits": 8,
3085
+ "group_size": 64,
3086
+ "mode": "affine"
3087
+ },
3088
+ "model.layers.25.linear_attn.out_proj": {
3089
+ "bits": 6,
3090
+ "group_size": 64,
3091
+ "mode": "affine"
3092
+ },
3093
+ "model.layers.25.linear_attn.in_proj_qkvz": {
3094
+ "bits": 6,
3095
+ "group_size": 64,
3096
+ "mode": "affine"
3097
+ },
3098
+ "model.layers.25.mlp.shared_expert.gate_proj": {
3099
+ "bits": 6,
3100
+ "group_size": 64,
3101
+ "mode": "affine"
3102
+ },
3103
+ "model.layers.25.linear_attn.in_proj_ba": {
3104
+ "bits": 6,
3105
+ "group_size": 64,
3106
+ "mode": "affine"
3107
+ },
3108
+ "model.layers.26.mlp.shared_expert_gate": {
3109
+ "bits": 8,
3110
+ "group_size": 64,
3111
+ "mode": "affine"
3112
+ },
3113
+ "model.layers.26.mlp.shared_expert.up_proj": {
3114
+ "bits": 5,
3115
+ "group_size": 64,
3116
+ "mode": "affine"
3117
+ },
3118
+ "model.layers.26.mlp.shared_expert.gate_proj": {
3119
+ "bits": 5,
3120
+ "group_size": 64,
3121
+ "mode": "affine"
3122
+ },
3123
+ "model.layers.26.mlp.shared_expert.down_proj": {
3124
+ "bits": 5,
3125
+ "group_size": 64,
3126
+ "mode": "affine"
3127
+ },
3128
+ "model.layers.26.linear_attn.out_proj": {
3129
+ "bits": 5,
3130
+ "group_size": 64,
3131
+ "mode": "affine"
3132
+ },
3133
+ "model.layers.26.linear_attn.in_proj_qkvz": {
3134
+ "bits": 5,
3135
+ "group_size": 64,
3136
+ "mode": "affine"
3137
+ },
3138
+ "model.layers.26.linear_attn.in_proj_ba": {
3139
+ "bits": 5,
3140
+ "group_size": 64,
3141
+ "mode": "affine"
3142
+ },
3143
+ "model.layers.26.mlp.gate": {
3144
+ "bits": 8,
3145
+ "group_size": 64,
3146
+ "mode": "affine"
3147
+ },
3148
+ "model.layers.27.self_attn.v_proj": {
3149
+ "bits": 6,
3150
+ "group_size": 64,
3151
+ "mode": "affine"
3152
+ },
3153
+ "model.layers.27.self_attn.q_proj": {
3154
+ "bits": 6,
3155
+ "group_size": 64,
3156
+ "mode": "affine"
3157
+ },
3158
+ "model.layers.27.self_attn.o_proj": {
3159
+ "bits": 6,
3160
+ "group_size": 64,
3161
+ "mode": "affine"
3162
+ },
3163
+ "model.layers.27.self_attn.k_proj": {
3164
+ "bits": 6,
3165
+ "group_size": 64,
3166
+ "mode": "affine"
3167
+ },
3168
+ "model.layers.27.mlp.shared_expert_gate": {
3169
+ "bits": 8,
3170
+ "group_size": 64,
3171
+ "mode": "affine"
3172
+ },
3173
+ "model.layers.27.mlp.shared_expert.up_proj": {
3174
+ "bits": 6,
3175
+ "group_size": 64,
3176
+ "mode": "affine"
3177
+ },
3178
+ "model.layers.27.mlp.shared_expert.gate_proj": {
3179
+ "bits": 6,
3180
+ "group_size": 64,
3181
+ "mode": "affine"
3182
+ },
3183
+ "model.layers.27.mlp.gate": {
3184
+ "bits": 8,
3185
+ "group_size": 64,
3186
+ "mode": "affine"
3187
+ },
3188
+ "model.layers.27.mlp.shared_expert.down_proj": {
3189
+ "bits": 6,
3190
+ "group_size": 64,
3191
+ "mode": "affine"
3192
+ },
3193
+ "model.layers.28.mlp.shared_expert.up_proj": {
3194
+ "bits": 5,
3195
+ "group_size": 64,
3196
+ "mode": "affine"
3197
+ },
3198
+ "model.layers.28.mlp.shared_expert.down_proj": {
3199
+ "bits": 5,
3200
+ "group_size": 64,
3201
+ "mode": "affine"
3202
+ },
3203
+ "model.layers.28.mlp.gate": {
3204
+ "bits": 8,
3205
+ "group_size": 64,
3206
+ "mode": "affine"
3207
+ },
3208
+ "model.layers.28.linear_attn.out_proj": {
3209
+ "bits": 5,
3210
+ "group_size": 64,
3211
+ "mode": "affine"
3212
+ },
3213
+ "model.layers.28.linear_attn.in_proj_qkvz": {
3214
+ "bits": 5,
3215
+ "group_size": 64,
3216
+ "mode": "affine"
3217
+ },
3218
+ "model.layers.28.linear_attn.in_proj_ba": {
3219
+ "bits": 5,
3220
+ "group_size": 64,
3221
+ "mode": "affine"
3222
+ },
3223
+ "model.layers.28.mlp.shared_expert.gate_proj": {
3224
+ "bits": 5,
3225
+ "group_size": 64,
3226
+ "mode": "affine"
3227
+ },
3228
+ "model.layers.28.mlp.shared_expert_gate": {
3229
+ "bits": 8,
3230
+ "group_size": 64,
3231
+ "mode": "affine"
3232
+ },
3233
+ "model.layers.30.linear_attn.in_proj_qkvz": {
3234
+ "bits": 5,
3235
+ "group_size": 64,
3236
+ "mode": "affine"
3237
+ },
3238
+ "model.layers.30.linear_attn.in_proj_ba": {
3239
+ "bits": 5,
3240
+ "group_size": 64,
3241
+ "mode": "affine"
3242
+ },
3243
+ "model.layers.29.mlp.shared_expert_gate": {
3244
+ "bits": 8,
3245
+ "group_size": 64,
3246
+ "mode": "affine"
3247
+ },
3248
+ "model.layers.29.mlp.shared_expert.up_proj": {
3249
+ "bits": 5,
3250
+ "group_size": 64,
3251
+ "mode": "affine"
3252
+ },
3253
+ "model.layers.29.mlp.shared_expert.down_proj": {
3254
+ "bits": 5,
3255
+ "group_size": 64,
3256
+ "mode": "affine"
3257
+ },
3258
+ "model.layers.29.mlp.gate": {
3259
+ "bits": 8,
3260
+ "group_size": 64,
3261
+ "mode": "affine"
3262
+ },
3263
+ "model.layers.30.mlp.gate": {
3264
+ "bits": 8,
3265
+ "group_size": 64,
3266
+ "mode": "affine"
3267
+ },
3268
+ "model.layers.29.mlp.shared_expert.gate_proj": {
3269
+ "bits": 5,
3270
+ "group_size": 64,
3271
+ "mode": "affine"
3272
+ },
3273
+ "model.layers.30.mlp.shared_expert.gate_proj": {
3274
+ "bits": 5,
3275
+ "group_size": 64,
3276
+ "mode": "affine"
3277
+ },
3278
+ "model.layers.30.mlp.shared_expert.up_proj": {
3279
+ "bits": 5,
3280
+ "group_size": 64,
3281
+ "mode": "affine"
3282
+ },
3283
+ "model.layers.29.linear_attn.out_proj": {
3284
+ "bits": 5,
3285
+ "group_size": 64,
3286
+ "mode": "affine"
3287
+ },
3288
+ "model.layers.30.mlp.shared_expert.down_proj": {
3289
+ "bits": 5,
3290
+ "group_size": 64,
3291
+ "mode": "affine"
3292
+ },
3293
+ "model.layers.30.linear_attn.out_proj": {
3294
+ "bits": 5,
3295
+ "group_size": 64,
3296
+ "mode": "affine"
3297
+ },
3298
+ "model.layers.30.mlp.shared_expert_gate": {
3299
+ "bits": 8,
3300
+ "group_size": 64,
3301
+ "mode": "affine"
3302
+ },
3303
+ "model.layers.29.linear_attn.in_proj_qkvz": {
3304
+ "bits": 5,
3305
+ "group_size": 64,
3306
+ "mode": "affine"
3307
+ },
3308
+ "model.layers.29.linear_attn.in_proj_ba": {
3309
+ "bits": 5,
3310
+ "group_size": 64,
3311
+ "mode": "affine"
3312
+ },
3313
+ "model.layers.31.self_attn.v_proj": {
3314
+ "bits": 5,
3315
+ "group_size": 64,
3316
+ "mode": "affine"
3317
+ },
3318
+ "model.layers.31.self_attn.q_proj": {
3319
+ "bits": 5,
3320
+ "group_size": 64,
3321
+ "mode": "affine"
3322
+ },
3323
+ "model.layers.31.self_attn.k_proj": {
3324
+ "bits": 5,
3325
+ "group_size": 64,
3326
+ "mode": "affine"
3327
+ },
3328
+ "model.layers.31.mlp.shared_expert_gate": {
3329
+ "bits": 8,
3330
+ "group_size": 64,
3331
+ "mode": "affine"
3332
+ },
3333
+ "model.layers.31.mlp.shared_expert.up_proj": {
3334
+ "bits": 5,
3335
+ "group_size": 64,
3336
+ "mode": "affine"
3337
+ },
3338
+ "model.layers.31.mlp.shared_expert.gate_proj": {
3339
+ "bits": 5,
3340
+ "group_size": 64,
3341
+ "mode": "affine"
3342
+ },
3343
+ "model.layers.31.mlp.shared_expert.down_proj": {
3344
+ "bits": 5,
3345
+ "group_size": 64,
3346
+ "mode": "affine"
3347
+ },
3348
+ "model.layers.31.mlp.gate": {
3349
+ "bits": 8,
3350
+ "group_size": 64,
3351
+ "mode": "affine"
3352
+ },
3353
+ "model.layers.31.self_attn.o_proj": {
3354
+ "bits": 5,
3355
+ "group_size": 64,
3356
+ "mode": "affine"
3357
+ },
3358
+ "model.layers.32.mlp.shared_expert_gate": {
3359
+ "bits": 8,
3360
+ "group_size": 64,
3361
+ "mode": "affine"
3362
+ },
3363
+ "model.layers.32.mlp.shared_expert.down_proj": {
3364
+ "bits": 5,
3365
+ "group_size": 64,
3366
+ "mode": "affine"
3367
+ },
3368
+ "model.layers.32.linear_attn.out_proj": {
3369
+ "bits": 5,
3370
+ "group_size": 64,
3371
+ "mode": "affine"
3372
+ },
3373
+ "model.layers.32.linear_attn.in_proj_qkvz": {
3374
+ "bits": 5,
3375
+ "group_size": 64,
3376
+ "mode": "affine"
3377
+ },
3378
+ "model.layers.32.linear_attn.in_proj_ba": {
3379
+ "bits": 5,
3380
+ "group_size": 64,
3381
+ "mode": "affine"
3382
+ },
3383
+ "model.layers.32.mlp.shared_expert.up_proj": {
3384
+ "bits": 5,
3385
+ "group_size": 64,
3386
+ "mode": "affine"
3387
+ },
3388
+ "model.layers.32.mlp.gate": {
3389
+ "bits": 8,
3390
+ "group_size": 64,
3391
+ "mode": "affine"
3392
+ },
3393
+ "model.layers.32.mlp.shared_expert.gate_proj": {
3394
+ "bits": 5,
3395
+ "group_size": 64,
3396
+ "mode": "affine"
3397
+ },
3398
+ "model.layers.33.mlp.shared_expert_gate": {
3399
+ "bits": 8,
3400
+ "group_size": 64,
3401
+ "mode": "affine"
3402
+ },
3403
+ "model.layers.33.mlp.shared_expert.up_proj": {
3404
+ "bits": 5,
3405
+ "group_size": 64,
3406
+ "mode": "affine"
3407
+ },
3408
+ "model.layers.33.mlp.shared_expert.gate_proj": {
3409
+ "bits": 5,
3410
+ "group_size": 64,
3411
+ "mode": "affine"
3412
+ },
3413
+ "model.layers.33.mlp.shared_expert.down_proj": {
3414
+ "bits": 5,
3415
+ "group_size": 64,
3416
+ "mode": "affine"
3417
+ },
3418
+ "model.layers.33.mlp.gate": {
3419
+ "bits": 8,
3420
+ "group_size": 64,
3421
+ "mode": "affine"
3422
+ },
3423
+ "model.layers.33.linear_attn.out_proj": {
3424
+ "bits": 5,
3425
+ "group_size": 64,
3426
+ "mode": "affine"
3427
+ },
3428
+ "model.layers.33.linear_attn.in_proj_qkvz": {
3429
+ "bits": 5,
3430
+ "group_size": 64,
3431
+ "mode": "affine"
3432
+ },
3433
+ "model.layers.33.linear_attn.in_proj_ba": {
3434
+ "bits": 5,
3435
+ "group_size": 64,
3436
+ "mode": "affine"
3437
+ },
3438
+ "model.layers.35.self_attn.o_proj": {
3439
+ "bits": 6,
3440
+ "group_size": 64,
3441
+ "mode": "affine"
3442
+ },
3443
+ "model.layers.35.mlp.shared_expert_gate": {
3444
+ "bits": 8,
3445
+ "group_size": 64,
3446
+ "mode": "affine"
3447
+ },
3448
+ "model.layers.35.mlp.gate": {
3449
+ "bits": 8,
3450
+ "group_size": 64,
3451
+ "mode": "affine"
3452
+ },
3453
+ "model.layers.34.mlp.shared_expert_gate": {
3454
+ "bits": 8,
3455
+ "group_size": 64,
3456
+ "mode": "affine"
3457
+ },
3458
+ "model.layers.34.mlp.shared_expert.up_proj": {
3459
+ "bits": 5,
3460
+ "group_size": 64,
3461
+ "mode": "affine"
3462
+ },
3463
+ "model.layers.34.mlp.shared_expert.gate_proj": {
3464
+ "bits": 5,
3465
+ "group_size": 64,
3466
+ "mode": "affine"
3467
+ },
3468
+ "model.layers.35.mlp.shared_expert.gate_proj": {
3469
+ "bits": 6,
3470
+ "group_size": 64,
3471
+ "mode": "affine"
3472
+ },
3473
+ "model.layers.35.mlp.shared_expert.up_proj": {
3474
+ "bits": 6,
3475
+ "group_size": 64,
3476
+ "mode": "affine"
3477
+ },
3478
+ "model.layers.34.linear_attn.in_proj_ba": {
3479
+ "bits": 5,
3480
+ "group_size": 64,
3481
+ "mode": "affine"
3482
+ },
3483
+ "model.layers.35.self_attn.k_proj": {
3484
+ "bits": 6,
3485
+ "group_size": 64,
3486
+ "mode": "affine"
3487
+ },
3488
+ "model.layers.35.self_attn.v_proj": {
3489
+ "bits": 6,
3490
+ "group_size": 64,
3491
+ "mode": "affine"
3492
+ },
3493
+ "model.layers.34.linear_attn.out_proj": {
3494
+ "bits": 5,
3495
+ "group_size": 64,
3496
+ "mode": "affine"
3497
+ },
3498
+ "model.layers.35.mlp.shared_expert.down_proj": {
3499
+ "bits": 6,
3500
+ "group_size": 64,
3501
+ "mode": "affine"
3502
+ },
3503
+ "model.layers.34.linear_attn.in_proj_qkvz": {
3504
+ "bits": 5,
3505
+ "group_size": 64,
3506
+ "mode": "affine"
3507
+ },
3508
+ "model.layers.34.mlp.gate": {
3509
+ "bits": 8,
3510
+ "group_size": 64,
3511
+ "mode": "affine"
3512
+ },
3513
+ "model.layers.34.mlp.shared_expert.down_proj": {
3514
+ "bits": 5,
3515
+ "group_size": 64,
3516
+ "mode": "affine"
3517
+ },
3518
+ "model.layers.35.self_attn.q_proj": {
3519
+ "bits": 6,
3520
+ "group_size": 64,
3521
+ "mode": "affine"
3522
+ },
3523
+ "model.layers.36.mlp.shared_expert_gate": {
3524
+ "bits": 8,
3525
+ "group_size": 64,
3526
+ "mode": "affine"
3527
+ },
3528
+ "model.layers.36.mlp.shared_expert.up_proj": {
3529
+ "bits": 6,
3530
+ "group_size": 64,
3531
+ "mode": "affine"
3532
+ },
3533
+ "model.layers.36.mlp.shared_expert.gate_proj": {
3534
+ "bits": 6,
3535
+ "group_size": 64,
3536
+ "mode": "affine"
3537
+ },
3538
+ "model.layers.36.linear_attn.in_proj_ba": {
3539
+ "bits": 6,
3540
+ "group_size": 64,
3541
+ "mode": "affine"
3542
+ },
3543
+ "model.layers.36.linear_attn.in_proj_qkvz": {
3544
+ "bits": 6,
3545
+ "group_size": 64,
3546
+ "mode": "affine"
3547
+ },
3548
+ "model.layers.36.linear_attn.out_proj": {
3549
+ "bits": 6,
3550
+ "group_size": 64,
3551
+ "mode": "affine"
3552
+ },
3553
+ "model.layers.36.mlp.shared_expert.down_proj": {
3554
+ "bits": 6,
3555
+ "group_size": 64,
3556
+ "mode": "affine"
3557
+ },
3558
+ "model.layers.36.mlp.gate": {
3559
+ "bits": 8,
3560
+ "group_size": 64,
3561
+ "mode": "affine"
3562
+ },
3563
+ "model.layers.37.mlp.shared_expert_gate": {
3564
+ "bits": 8,
3565
+ "group_size": 64,
3566
+ "mode": "affine"
3567
+ },
3568
+ "model.layers.37.mlp.shared_expert.up_proj": {
3569
+ "bits": 5,
3570
+ "group_size": 64,
3571
+ "mode": "affine"
3572
+ },
3573
+ "model.layers.37.mlp.shared_expert.gate_proj": {
3574
+ "bits": 5,
3575
+ "group_size": 64,
3576
+ "mode": "affine"
3577
+ },
3578
+ "model.layers.37.mlp.gate": {
3579
+ "bits": 8,
3580
+ "group_size": 64,
3581
+ "mode": "affine"
3582
+ },
3583
+ "model.layers.37.linear_attn.out_proj": {
3584
+ "bits": 5,
3585
+ "group_size": 64,
3586
+ "mode": "affine"
3587
+ },
3588
+ "model.layers.37.linear_attn.in_proj_qkvz": {
3589
+ "bits": 5,
3590
+ "group_size": 64,
3591
+ "mode": "affine"
3592
+ },
3593
+ "model.layers.37.linear_attn.in_proj_ba": {
3594
+ "bits": 5,
3595
+ "group_size": 64,
3596
+ "mode": "affine"
3597
+ },
3598
+ "model.layers.37.mlp.shared_expert.down_proj": {
3599
+ "bits": 5,
3600
+ "group_size": 64,
3601
+ "mode": "affine"
3602
+ },
3603
+ "model.layers.38.mlp.shared_expert.down_proj": {
3604
+ "bits": 5,
3605
+ "group_size": 64,
3606
+ "mode": "affine"
3607
+ },
3608
+ "model.layers.38.mlp.gate": {
3609
+ "bits": 8,
3610
+ "group_size": 64,
3611
+ "mode": "affine"
3612
+ },
3613
+ "model.layers.38.linear_attn.out_proj": {
3614
+ "bits": 5,
3615
+ "group_size": 64,
3616
+ "mode": "affine"
3617
+ },
3618
+ "model.layers.38.mlp.shared_expert.gate_proj": {
3619
+ "bits": 5,
3620
+ "group_size": 64,
3621
+ "mode": "affine"
3622
+ },
3623
+ "model.layers.38.mlp.shared_expert_gate": {
3624
+ "bits": 8,
3625
+ "group_size": 64,
3626
+ "mode": "affine"
3627
+ },
3628
+ "model.layers.38.linear_attn.in_proj_qkvz": {
3629
+ "bits": 5,
3630
+ "group_size": 64,
3631
+ "mode": "affine"
3632
+ },
3633
+ "model.layers.38.linear_attn.in_proj_ba": {
3634
+ "bits": 5,
3635
+ "group_size": 64,
3636
+ "mode": "affine"
3637
+ },
3638
+ "model.layers.38.mlp.shared_expert.up_proj": {
3639
+ "bits": 5,
3640
+ "group_size": 64,
3641
+ "mode": "affine"
3642
+ },
3643
+ "model.layers.39.self_attn.v_proj": {
3644
+ "bits": 6,
3645
+ "group_size": 64,
3646
+ "mode": "affine"
3647
+ },
3648
+ "model.layers.39.self_attn.o_proj": {
3649
+ "bits": 6,
3650
+ "group_size": 64,
3651
+ "mode": "affine"
3652
+ },
3653
+ "model.layers.39.self_attn.k_proj": {
3654
+ "bits": 6,
3655
+ "group_size": 64,
3656
+ "mode": "affine"
3657
+ },
3658
+ "model.layers.39.mlp.shared_expert_gate": {
3659
+ "bits": 8,
3660
+ "group_size": 64,
3661
+ "mode": "affine"
3662
+ },
3663
+ "model.layers.39.mlp.shared_expert.gate_proj": {
3664
+ "bits": 6,
3665
+ "group_size": 64,
3666
+ "mode": "affine"
3667
+ },
3668
+ "model.layers.39.mlp.shared_expert.down_proj": {
3669
+ "bits": 6,
3670
+ "group_size": 64,
3671
+ "mode": "affine"
3672
+ },
3673
+ "model.layers.39.mlp.gate": {
3674
+ "bits": 8,
3675
+ "group_size": 64,
3676
+ "mode": "affine"
3677
+ },
3678
+ "model.layers.39.self_attn.q_proj": {
3679
+ "bits": 6,
3680
+ "group_size": 64,
3681
+ "mode": "affine"
3682
+ },
3683
+ "model.layers.39.mlp.shared_expert.up_proj": {
3684
+ "bits": 6,
3685
+ "group_size": 64,
3686
+ "mode": "affine"
3687
+ },
3688
+ "model.layers.41.mlp.shared_expert.up_proj": {
3689
+ "bits": 6,
3690
+ "group_size": 64,
3691
+ "mode": "affine"
3692
+ },
3693
+ "model.layers.41.mlp.shared_expert.gate_proj": {
3694
+ "bits": 6,
3695
+ "group_size": 64,
3696
+ "mode": "affine"
3697
+ },
3698
+ "model.layers.41.mlp.gate": {
3699
+ "bits": 8,
3700
+ "group_size": 64,
3701
+ "mode": "affine"
3702
+ },
3703
+ "model.layers.41.linear_attn.out_proj": {
3704
+ "bits": 6,
3705
+ "group_size": 64,
3706
+ "mode": "affine"
3707
+ },
3708
+ "model.layers.40.mlp.shared_expert.down_proj": {
3709
+ "bits": 6,
3710
+ "group_size": 64,
3711
+ "mode": "affine"
3712
+ },
3713
+ "model.layers.40.mlp.gate": {
3714
+ "bits": 8,
3715
+ "group_size": 64,
3716
+ "mode": "affine"
3717
+ },
3718
+ "model.layers.41.linear_attn.in_proj_qkvz": {
3719
+ "bits": 6,
3720
+ "group_size": 64,
3721
+ "mode": "affine"
3722
+ },
3723
+ "model.layers.40.mlp.shared_expert.gate_proj": {
3724
+ "bits": 6,
3725
+ "group_size": 64,
3726
+ "mode": "affine"
3727
+ },
3728
+ "model.layers.41.linear_attn.in_proj_ba": {
3729
+ "bits": 6,
3730
+ "group_size": 64,
3731
+ "mode": "affine"
3732
+ },
3733
+ "model.layers.40.linear_attn.out_proj": {
3734
+ "bits": 6,
3735
+ "group_size": 64,
3736
+ "mode": "affine"
3737
+ },
3738
+ "model.layers.40.mlp.shared_expert_gate": {
3739
+ "bits": 8,
3740
+ "group_size": 64,
3741
+ "mode": "affine"
3742
+ },
3743
+ "model.layers.41.mlp.shared_expert.down_proj": {
3744
+ "bits": 6,
3745
+ "group_size": 64,
3746
+ "mode": "affine"
3747
+ },
3748
+ "model.layers.40.mlp.shared_expert.up_proj": {
3749
+ "bits": 6,
3750
+ "group_size": 64,
3751
+ "mode": "affine"
3752
+ },
3753
+ "model.layers.41.mlp.shared_expert_gate": {
3754
+ "bits": 8,
3755
+ "group_size": 64,
3756
+ "mode": "affine"
3757
+ },
3758
+ "model.layers.40.linear_attn.in_proj_ba": {
3759
+ "bits": 6,
3760
+ "group_size": 64,
3761
+ "mode": "affine"
3762
+ },
3763
+ "model.layers.40.linear_attn.in_proj_qkvz": {
3764
+ "bits": 6,
3765
+ "group_size": 64,
3766
+ "mode": "affine"
3767
+ },
3768
+ "model.layers.42.mlp.shared_expert.gate_proj": {
3769
+ "bits": 6,
3770
+ "group_size": 64,
3771
+ "mode": "affine"
3772
+ },
3773
+ "model.layers.42.mlp.gate": {
3774
+ "bits": 8,
3775
+ "group_size": 64,
3776
+ "mode": "affine"
3777
+ },
3778
+ "model.layers.42.mlp.shared_expert.down_proj": {
3779
+ "bits": 6,
3780
+ "group_size": 64,
3781
+ "mode": "affine"
3782
+ },
3783
+ "model.layers.42.linear_attn.in_proj_qkvz": {
3784
+ "bits": 6,
3785
+ "group_size": 64,
3786
+ "mode": "affine"
3787
+ },
3788
+ "model.layers.42.linear_attn.in_proj_ba": {
3789
+ "bits": 6,
3790
+ "group_size": 64,
3791
+ "mode": "affine"
3792
+ },
3793
+ "model.layers.42.mlp.shared_expert.up_proj": {
3794
+ "bits": 6,
3795
+ "group_size": 64,
3796
+ "mode": "affine"
3797
+ },
3798
+ "model.layers.42.linear_attn.out_proj": {
3799
+ "bits": 6,
3800
+ "group_size": 64,
3801
+ "mode": "affine"
3802
+ },
3803
+ "model.layers.42.mlp.shared_expert_gate": {
3804
+ "bits": 8,
3805
+ "group_size": 64,
3806
+ "mode": "affine"
3807
+ },
3808
+ "model.layers.43.self_attn.q_proj": {
3809
+ "bits": 8,
3810
+ "group_size": 32,
3811
+ "mode": "mxfp8"
3812
+ },
3813
+ "model.layers.43.self_attn.o_proj": {
3814
+ "bits": 8,
3815
+ "group_size": 32,
3816
+ "mode": "mxfp8"
3817
+ },
3818
+ "model.layers.43.mlp.shared_expert.up_proj": {
3819
+ "bits": 8,
3820
+ "group_size": 32,
3821
+ "mode": "mxfp8"
3822
+ },
3823
+ "model.layers.43.mlp.shared_expert.gate_proj": {
3824
+ "bits": 8,
3825
+ "group_size": 32,
3826
+ "mode": "mxfp8"
3827
+ },
3828
+ "model.layers.43.mlp.shared_expert.down_proj": {
3829
+ "bits": 8,
3830
+ "group_size": 32,
3831
+ "mode": "mxfp8"
3832
+ },
3833
+ "model.layers.43.self_attn.k_proj": {
3834
+ "bits": 8,
3835
+ "group_size": 32,
3836
+ "mode": "mxfp8"
3837
+ },
3838
+ "model.layers.43.self_attn.v_proj": {
3839
+ "bits": 8,
3840
+ "group_size": 32,
3841
+ "mode": "mxfp8"
3842
+ },
3843
+ "model.layers.43.mlp.shared_expert_gate": {
3844
+ "bits": 8,
3845
+ "group_size": 64,
3846
+ "mode": "affine"
3847
+ },
3848
+ "model.layers.43.mlp.gate": {
3849
+ "bits": 8,
3850
+ "group_size": 64,
3851
+ "mode": "affine"
3852
+ },
3853
+ "model.layers.44.mlp.shared_expert_gate": {
3854
+ "bits": 8,
3855
+ "group_size": 64,
3856
+ "mode": "affine"
3857
+ },
3858
+ "model.layers.44.mlp.shared_expert.gate_proj": {
3859
+ "bits": 6,
3860
+ "group_size": 64,
3861
+ "mode": "affine"
3862
+ },
3863
+ "model.layers.44.mlp.shared_expert.down_proj": {
3864
+ "bits": 6,
3865
+ "group_size": 64,
3866
+ "mode": "affine"
3867
+ },
3868
+ "model.layers.44.mlp.gate": {
3869
+ "bits": 8,
3870
+ "group_size": 64,
3871
+ "mode": "affine"
3872
+ },
3873
+ "model.layers.44.linear_attn.in_proj_qkvz": {
3874
+ "bits": 6,
3875
+ "group_size": 64,
3876
+ "mode": "affine"
3877
+ },
3878
+ "model.layers.44.linear_attn.out_proj": {
3879
+ "bits": 6,
3880
+ "group_size": 64,
3881
+ "mode": "affine"
3882
+ },
3883
+ "model.layers.44.linear_attn.in_proj_ba": {
3884
+ "bits": 6,
3885
+ "group_size": 64,
3886
+ "mode": "affine"
3887
+ },
3888
+ "model.layers.44.mlp.shared_expert.up_proj": {
3889
+ "bits": 6,
3890
+ "group_size": 64,
3891
+ "mode": "affine"
3892
+ },
3893
+ "model.layers.45.mlp.shared_expert.down_proj": {
3894
+ "bits": 8,
3895
+ "group_size": 32,
3896
+ "mode": "mxfp8"
3897
+ },
3898
+ "model.layers.45.mlp.gate": {
3899
+ "bits": 8,
3900
+ "group_size": 64,
3901
+ "mode": "affine"
3902
+ },
3903
+ "model.layers.45.mlp.shared_expert.gate_proj": {
3904
+ "bits": 8,
3905
+ "group_size": 32,
3906
+ "mode": "mxfp8"
3907
+ },
3908
+ "model.layers.45.mlp.shared_expert_gate": {
3909
+ "bits": 8,
3910
+ "group_size": 64,
3911
+ "mode": "affine"
3912
+ },
3913
+ "model.layers.45.mlp.shared_expert.up_proj": {
3914
+ "bits": 8,
3915
+ "group_size": 32,
3916
+ "mode": "mxfp8"
3917
+ },
3918
+ "model.layers.45.linear_attn.out_proj": {
3919
+ "bits": 8,
3920
+ "group_size": 32,
3921
+ "mode": "mxfp8"
3922
+ },
3923
+ "model.layers.45.linear_attn.in_proj_ba": {
3924
+ "bits": 8,
3925
+ "group_size": 32,
3926
+ "mode": "mxfp8"
3927
+ },
3928
+ "model.layers.45.linear_attn.in_proj_qkvz": {
3929
+ "bits": 8,
3930
+ "group_size": 32,
3931
+ "mode": "mxfp8"
3932
+ },
3933
+ "model.layers.47.self_attn.v_proj": {
3934
+ "bits": 8,
3935
+ "group_size": 32,
3936
+ "mode": "mxfp8"
3937
+ },
3938
+ "model.layers.47.self_attn.q_proj": {
3939
+ "bits": 8,
3940
+ "group_size": 32,
3941
+ "mode": "mxfp8"
3942
+ },
3943
+ "model.layers.47.self_attn.k_proj": {
3944
+ "bits": 8,
3945
+ "group_size": 32,
3946
+ "mode": "mxfp8"
3947
+ },
3948
+ "model.layers.47.mlp.shared_expert_gate": {
3949
+ "bits": 8,
3950
+ "group_size": 64,
3951
+ "mode": "affine"
3952
+ },
3953
+ "model.layers.47.mlp.shared_expert.up_proj": {
3954
+ "bits": 8,
3955
+ "group_size": 32,
3956
+ "mode": "mxfp8"
3957
+ },
3958
+ "model.layers.46.mlp.shared_expert_gate": {
3959
+ "bits": 8,
3960
+ "group_size": 64,
3961
+ "mode": "affine"
3962
+ },
3963
+ "model.layers.46.mlp.shared_expert.gate_proj": {
3964
+ "bits": 8,
3965
+ "group_size": 32,
3966
+ "mode": "mxfp8"
3967
+ },
3968
+ "model.layers.46.mlp.shared_expert.down_proj": {
3969
+ "bits": 8,
3970
+ "group_size": 32,
3971
+ "mode": "mxfp8"
3972
+ },
3973
+ "model.layers.47.mlp.shared_expert.down_proj": {
3974
+ "bits": 8,
3975
+ "group_size": 32,
3976
+ "mode": "mxfp8"
3977
+ },
3978
+ "model.layers.47.self_attn.o_proj": {
3979
+ "bits": 8,
3980
+ "group_size": 32,
3981
+ "mode": "mxfp8"
3982
+ },
3983
+ "model.layers.47.mlp.shared_expert.gate_proj": {
3984
+ "bits": 8,
3985
+ "group_size": 32,
3986
+ "mode": "mxfp8"
3987
+ },
3988
+ "model.layers.46.mlp.gate": {
3989
+ "bits": 8,
3990
+ "group_size": 64,
3991
+ "mode": "affine"
3992
+ },
3993
+ "model.layers.46.linear_attn.in_proj_qkvz": {
3994
+ "bits": 8,
3995
+ "group_size": 32,
3996
+ "mode": "mxfp8"
3997
+ },
3998
+ "model.layers.46.linear_attn.in_proj_ba": {
3999
+ "bits": 8,
4000
+ "group_size": 32,
4001
+ "mode": "mxfp8"
4002
+ },
4003
+ "model.layers.46.linear_attn.out_proj": {
4004
+ "bits": 8,
4005
+ "group_size": 32,
4006
+ "mode": "mxfp8"
4007
+ },
4008
+ "model.layers.47.mlp.gate": {
4009
+ "bits": 8,
4010
+ "group_size": 64,
4011
+ "mode": "affine"
4012
+ },
4013
+ "model.layers.46.mlp.shared_expert.up_proj": {
4014
+ "bits": 8,
4015
+ "group_size": 32,
4016
+ "mode": "mxfp8"
4017
+ },
4018
+ "lm_head": {
4019
+ "bits": 8,
4020
+ "group_size": 32,
4021
+ "mode": "mxfp8"
4022
+ }
4023
+ }
4024
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 151643,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 151645,
6
+ 151643
7
+ ],
8
+ "pad_token_id": 151643,
9
+ "temperature": 1.0,
10
+ "top_k": 40,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.57.3"
13
+ }
model-00001-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aee4a4734d0836a7b3437787a05445faed71ed5a292b1b48a51d6203583d0602
3
+ size 5202948504
model-00002-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:020d0f2abe4573c4a866bad882a1518f225504891e6818f37b7795932309317c
3
+ size 5133834496
model-00003-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2825215611d8dcabecd46e78fb0d6a6fd3d369150ee7a744dfd1aeb9f41ae23a
3
+ size 5133834549
model-00004-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:17ba25908187e332d33e48a8272b0e6fb22c69b97ab26570b9459fee81cb651b
3
+ size 5133834545
model-00005-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:59910d501f65fed7b2dfe28b539cbcbc2adf377bbc0fc97b63d13a0e878b455a
3
+ size 5133834545
model-00006-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad74814e2f5d3ebfb32c0ed308a13b1dbbd93096c1a6bed00688f0416942bb04
3
+ size 5133834551
model-00007-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d4a355bab0994e1ad72519967e3c89a2d1ad535c927b8c87eab9d30f2a18c6a
3
+ size 5133834541
model-00008-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:876e982d4ca59c3b17137f0a960aa35f6dba44382bc69c7270e8a00ff90266a1
3
+ size 5133834537
model-00009-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e86741005a0d958a926efcd2ce687f526838074f411756d24e13dd97e326f985
3
+ size 4227863748
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3_coder_detector_sgl.py ADDED
@@ -0,0 +1,474 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import ast
2
+ import json
3
+ import logging
4
+ import re
5
+ from typing import Any, List, Optional
6
+
7
+ from sglang.srt.entrypoints.openai.protocol import Tool
8
+ from sglang.srt.function_call.base_format_detector import BaseFormatDetector
9
+ from sglang.srt.function_call.core_types import (
10
+ StreamingParseResult,
11
+ ToolCallItem,
12
+ _GetInfoFunc,
13
+ )
14
+
15
+ logger = logging.getLogger(__name__)
16
+
17
+
18
+ class Qwen3CoderDetector(BaseFormatDetector):
19
+ def __init__(self):
20
+ super().__init__()
21
+
22
+ # Sentinel tokens
23
+ self.tool_call_start_token: str = "<tool_call>"
24
+ self.tool_call_end_token: str = "</tool_call>"
25
+ self.tool_call_prefix: str = "<function="
26
+ self.function_end_token: str = "</function>"
27
+ self.parameter_prefix: str = "<parameter="
28
+ self.parameter_end_token: str = "</parameter>"
29
+
30
+ # Regex for non-streaming fallback
31
+ self.tool_call_regex = re.compile(r"<tool_call>(.*?)</tool_call>", re.DOTALL)
32
+ self.tool_call_function_regex = re.compile(
33
+ r"<function=(.*?)</function>|<function=(.*)$", re.DOTALL
34
+ )
35
+ self.tool_call_parameter_regex = re.compile(
36
+ r"<parameter=(.*?)(?:</parameter>|(?=<parameter=)|(?=</function>)|$)",
37
+ re.DOTALL,
38
+ )
39
+
40
+ # Streaming State
41
+ # Base class already initializes _buffer, we just use it directly
42
+ # No need to check with hasattr - we control the lifecycle through inheritance
43
+
44
+ # Index pointing to the next character to be processed in buffer
45
+ self.parsed_pos: int = 0
46
+ # Parameter count inside the current tool being processed, used to determine whether to add comma
47
+ self.current_tool_param_count: int = 0
48
+ # Flag indicating whether current tool has already sent '{'
49
+ self.json_started: bool = False
50
+
51
+ # [FIX] New state flag: mark whether inside tool_call structure block
52
+ self.is_inside_tool_call: bool = False
53
+
54
+ # Initialize attributes that were missing in the original PR
55
+ self.current_func_name: Optional[str] = None
56
+
57
+ def has_tool_call(self, text: str) -> bool:
58
+ return self.tool_call_start_token in text
59
+
60
+ def _get_arguments_config(
61
+ self, func_name: str, tools: Optional[list[Tool]]
62
+ ) -> dict:
63
+ """Extract argument configuration for a function."""
64
+ if tools is None:
65
+ return {}
66
+ for config in tools:
67
+ try:
68
+ config_type = config.type
69
+ config_function = config.function
70
+ config_function_name = config_function.name
71
+ except AttributeError:
72
+ continue
73
+
74
+ if config_type == "function" and config_function_name == func_name:
75
+ try:
76
+ params = config_function.parameters
77
+ except AttributeError:
78
+ return {}
79
+
80
+ if isinstance(params, dict) and "properties" in params:
81
+ return params["properties"]
82
+ elif isinstance(params, dict):
83
+ return params
84
+ else:
85
+ return {}
86
+ logger.warning(f"Tool '{func_name}' is not defined in the tools list.")
87
+ return {}
88
+
89
+ def _convert_param_value(
90
+ self, param_value: str, param_name: str, param_config: dict, func_name: str
91
+ ) -> Any:
92
+ """Convert parameter value based on its type in the schema."""
93
+ # Handle null value for any type
94
+ if param_value.lower() == "null":
95
+ return None
96
+
97
+ if param_name not in param_config:
98
+ if param_config != {}:
99
+ logger.warning(
100
+ f"Parsed parameter '{param_name}' is not defined in the tool "
101
+ f"parameters for tool '{func_name}', directly returning the string value."
102
+ )
103
+ return param_value
104
+
105
+ if (
106
+ isinstance(param_config[param_name], dict)
107
+ and "type" in param_config[param_name]
108
+ ):
109
+ param_type = str(param_config[param_name]["type"]).strip().lower()
110
+ else:
111
+ param_type = "string"
112
+ if param_type in ["string", "str", "text", "varchar", "char", "enum"]:
113
+ return param_value
114
+ elif (
115
+ param_type.startswith("int")
116
+ or param_type.startswith("uint")
117
+ or param_type.startswith("long")
118
+ or param_type.startswith("short")
119
+ or param_type.startswith("unsigned")
120
+ ):
121
+ try:
122
+ param_value = int(param_value)
123
+ except Exception:
124
+ logger.warning(
125
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not an integer in tool "
126
+ f"'{func_name}', degenerating to string."
127
+ )
128
+ return param_value
129
+ elif param_type.startswith("num") or param_type.startswith("float"):
130
+ try:
131
+ maybe_convert = (
132
+ False if "." in param_value or "e" in param_value.lower() else True
133
+ )
134
+ param_value: float = float(param_value)
135
+ if maybe_convert and param_value.is_integer():
136
+ param_value = int(param_value)
137
+ except Exception:
138
+ logger.warning(
139
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not a float in tool "
140
+ f"'{func_name}', degenerating to string."
141
+ )
142
+ return param_value
143
+ elif param_type in ["boolean", "bool", "binary"]:
144
+ param_value = param_value.lower()
145
+ if param_value not in ["true", "false"]:
146
+ logger.warning(
147
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not a boolean (`true` of `false`) in tool '{func_name}', degenerating to false."
148
+ )
149
+ return param_value == "true"
150
+ else:
151
+ if (
152
+ param_type in ["object", "array", "arr"]
153
+ or param_type.startswith("dict")
154
+ or param_type.startswith("list")
155
+ ):
156
+ try:
157
+ param_value = json.loads(param_value)
158
+ return param_value
159
+ except Exception:
160
+ logger.warning(
161
+ f"Parsed value '{param_value}' of parameter '{param_name}' cannot be parsed with json.loads in tool "
162
+ f"'{func_name}', will try other methods to parse it."
163
+ )
164
+ try:
165
+ param_value = ast.literal_eval(param_value) # safer
166
+ except Exception:
167
+ logger.warning(
168
+ f"Parsed value '{param_value}' of parameter '{param_name}' cannot be converted via Python `ast.literal_eval()` in tool '{func_name}', degenerating to string."
169
+ )
170
+ return param_value
171
+
172
+ def detect_and_parse(self, text: str, tools: List[Tool]) -> StreamingParseResult:
173
+ """One-shot parsing for non-streaming scenarios."""
174
+ if self.tool_call_start_token not in text:
175
+ return StreamingParseResult(normal_text=text)
176
+
177
+ calls = []
178
+ try:
179
+ # Simple cleanup of the text to find tool calls
180
+ # Note: This is a simplified regex approach consistent with vLLM
181
+ raw_tool_calls = self.tool_call_regex.findall(text)
182
+ if not raw_tool_calls:
183
+ # Fallback: maybe the whole text is inside the tag or tags are stripped
184
+ if self.tool_call_prefix in text:
185
+ raw_tool_calls = [text]
186
+
187
+ tool_idx = 0
188
+ for tool_content in raw_tool_calls:
189
+ # Find function calls
190
+ funcs = self.tool_call_function_regex.findall(tool_content)
191
+ for func_match in funcs:
192
+ func_body = func_match[0] or func_match[1]
193
+ if ">" not in func_body:
194
+ continue
195
+
196
+ name_end = func_body.index(">")
197
+ func_name = func_body[:name_end]
198
+ params_str = func_body[name_end + 1 :]
199
+
200
+ param_config = self._get_arguments_config(func_name, tools)
201
+ parsed_params = {}
202
+
203
+ for p_match in self.tool_call_parameter_regex.findall(params_str):
204
+ if ">" not in p_match:
205
+ continue
206
+ p_idx = p_match.index(">")
207
+ p_name = p_match[:p_idx]
208
+ p_val = p_match[p_idx + 1 :]
209
+ # Remove prefixing and trailing \n
210
+ if p_val.startswith("\n"):
211
+ p_val = p_val[1:]
212
+ if p_val.endswith("\n"):
213
+ p_val = p_val[:-1]
214
+
215
+ parsed_params[p_name] = self._convert_param_value(
216
+ p_val, p_name, param_config, func_name
217
+ )
218
+
219
+ calls.append(
220
+ ToolCallItem(
221
+ tool_index=tool_idx,
222
+ name=func_name,
223
+ parameters=json.dumps(parsed_params, ensure_ascii=False),
224
+ )
225
+ )
226
+ tool_idx += 1
227
+
228
+ # Determine normal text (text before the first tool call)
229
+ start_idx = text.find(self.tool_call_start_token)
230
+ if start_idx == -1:
231
+ start_idx = text.find(self.tool_call_prefix)
232
+ normal_text = text[:start_idx] if start_idx > 0 else ""
233
+
234
+ return StreamingParseResult(normal_text=normal_text, calls=calls)
235
+
236
+ except Exception as e:
237
+ logger.error(f"Error in detect_and_parse: {e}")
238
+ return StreamingParseResult(normal_text=text)
239
+
240
+ def parse_streaming_increment(
241
+ self, new_text: str, tools: List[Tool]
242
+ ) -> StreamingParseResult:
243
+ """
244
+ Robust cursor-based streaming parser.
245
+ """
246
+ self._buffer += new_text
247
+
248
+ # Guard against empty buffer
249
+ if not self._buffer:
250
+ return StreamingParseResult()
251
+
252
+ calls = []
253
+ normal_text_chunks = []
254
+
255
+ while True:
256
+ # Working text slice
257
+ current_slice = self._buffer[self.parsed_pos :]
258
+
259
+ # Optimization: If almost empty, wait for more
260
+ if not current_slice:
261
+ break
262
+
263
+ # -------------------------------------------------------
264
+ # 1. Priority detection: check if it's the start of Tool Call
265
+ # -------------------------------------------------------
266
+ if current_slice.startswith(self.tool_call_start_token):
267
+ self.parsed_pos += len(self.tool_call_start_token)
268
+ self.is_inside_tool_call = True
269
+ continue
270
+
271
+ # -------------------------------------------------------
272
+ # 2. Function Name: <function=name>
273
+ # -------------------------------------------------------
274
+ if current_slice.startswith(self.tool_call_prefix):
275
+ end_angle = current_slice.find(">")
276
+ if end_angle != -1:
277
+ func_name = current_slice[len(self.tool_call_prefix) : end_angle]
278
+
279
+ self.current_tool_id += 1
280
+ self.current_tool_name_sent = True
281
+ self.current_tool_param_count = 0
282
+ self.json_started = False
283
+ self.current_func_name = func_name
284
+
285
+ calls.append(
286
+ ToolCallItem(
287
+ tool_index=self.current_tool_id,
288
+ name=func_name,
289
+ parameters="",
290
+ )
291
+ )
292
+
293
+ self.parsed_pos += end_angle + 1
294
+ continue
295
+ else:
296
+ # Incomplete tag
297
+ break
298
+
299
+ # -------------------------------------------------------
300
+ # 3. Parameter: <parameter=name>value...
301
+ # -------------------------------------------------------
302
+ if current_slice.startswith(self.parameter_prefix):
303
+ name_end = current_slice.find(">")
304
+ if name_end != -1:
305
+ value_start_idx = name_end + 1
306
+ rest_of_slice = current_slice[value_start_idx:]
307
+
308
+ # A parameter can end in multiple ways:
309
+ # 1. [Normal] Encounter </parameter>
310
+ # 2. [Abnormal] Encounter next <parameter=
311
+ # 3. [Abnormal] Encounter </function>
312
+ # So we need to find the smallest one as the parameter end position.
313
+ cand_end_param = rest_of_slice.find(self.parameter_end_token)
314
+ cand_next_param = rest_of_slice.find(self.parameter_prefix)
315
+ cand_end_func = rest_of_slice.find(self.function_end_token)
316
+
317
+ candidates = []
318
+ if cand_end_param != -1:
319
+ candidates.append(
320
+ (cand_end_param, len(self.parameter_end_token))
321
+ )
322
+ if cand_next_param != -1:
323
+ candidates.append((cand_next_param, 0))
324
+ if cand_end_func != -1:
325
+ candidates.append((cand_end_func, 0))
326
+
327
+ if candidates:
328
+ best_cand = min(candidates, key=lambda x: x[0])
329
+ end_pos = best_cand[0]
330
+ end_token_len = best_cand[1]
331
+
332
+ param_name = current_slice[
333
+ len(self.parameter_prefix) : name_end
334
+ ]
335
+ raw_value = rest_of_slice[:end_pos]
336
+
337
+ # Cleanup value
338
+ if raw_value.startswith("\n"):
339
+ raw_value = raw_value[1:]
340
+ if raw_value.endswith("\n"):
341
+ raw_value = raw_value[:-1]
342
+
343
+ # JSON Construction
344
+ if not self.json_started:
345
+ calls.append(
346
+ ToolCallItem(
347
+ tool_index=self.current_tool_id, parameters="{"
348
+ )
349
+ )
350
+ self.json_started = True
351
+
352
+ param_config = self._get_arguments_config(
353
+ self.current_func_name, tools
354
+ )
355
+ converted_val = self._convert_param_value(
356
+ raw_value, param_name, param_config, self.current_func_name
357
+ )
358
+
359
+ # Construct JSON fragment: "key": value
360
+ # Note: We must be careful with json.dumps to ensure valid JSON streaming
361
+ json_key_val = f"{json.dumps(param_name)}: {json.dumps(converted_val, ensure_ascii=False)}"
362
+
363
+ if self.current_tool_param_count > 0:
364
+ fragment = f", {json_key_val}"
365
+ else:
366
+ fragment = json_key_val
367
+
368
+ calls.append(
369
+ ToolCallItem(
370
+ tool_index=self.current_tool_id, parameters=fragment
371
+ )
372
+ )
373
+ self.current_tool_param_count += 1
374
+
375
+ # Advance cursor
376
+ total_len = (name_end + 1) + end_pos + end_token_len
377
+ self.parsed_pos += total_len
378
+ continue
379
+
380
+ # Incomplete parameter tag or value
381
+ break
382
+
383
+ # -------------------------------------------------------
384
+ # 4. Function End: </function>
385
+ # -------------------------------------------------------
386
+ if current_slice.startswith(self.function_end_token):
387
+ if not self.json_started:
388
+ calls.append(
389
+ ToolCallItem(tool_index=self.current_tool_id, parameters="{")
390
+ )
391
+ self.json_started = True
392
+
393
+ calls.append(
394
+ ToolCallItem(tool_index=self.current_tool_id, parameters="}")
395
+ )
396
+ self.parsed_pos += len(self.function_end_token)
397
+ self.current_func_name = None
398
+ continue
399
+
400
+ # -------------------------------------------------------
401
+ # 5. Tool Call End: </tool_call>
402
+ # -------------------------------------------------------
403
+ if current_slice.startswith(self.tool_call_end_token):
404
+ self.parsed_pos += len(self.tool_call_end_token)
405
+ self.is_inside_tool_call = False # [FIX] Exit tool call region
406
+ continue
407
+
408
+ # -------------------------------------------------------
409
+ # 6. Handling content / whitespace / normal text
410
+ # -------------------------------------------------------
411
+ # If current position is not the start of a tag (i.e., doesn't start with <), it might be plain text,
412
+ # or a newline between two tags.
413
+ # But we need to be careful not to output truncated tags like "<fun" as text.
414
+
415
+ next_open_angle = current_slice.find("<")
416
+
417
+ if next_open_angle == -1:
418
+ # This entire segment is plain text
419
+ if not self.is_inside_tool_call:
420
+ normal_text_chunks.append(current_slice)
421
+ # [FIX] If inside tool call, discard this text (usually \n), don't append
422
+ self.parsed_pos += len(current_slice)
423
+ continue
424
+
425
+ elif next_open_angle == 0:
426
+ # Looks like a Tag, but doesn't match any known Tag above
427
+
428
+ possible_tags = [
429
+ self.tool_call_start_token,
430
+ self.tool_call_end_token,
431
+ self.tool_call_prefix,
432
+ self.function_end_token,
433
+ self.parameter_prefix,
434
+ self.parameter_end_token,
435
+ ]
436
+
437
+ is_potential_tag = False
438
+ for tag in possible_tags:
439
+ if tag.startswith(current_slice):
440
+ is_potential_tag = True
441
+ break
442
+
443
+ if is_potential_tag:
444
+ break # Wait for more
445
+ else:
446
+ # Just a plain '<' symbol
447
+ if not self.is_inside_tool_call:
448
+ normal_text_chunks.append("<")
449
+ self.parsed_pos += 1
450
+ continue
451
+
452
+ else:
453
+ # '<' is in the middle
454
+ text_segment = current_slice[:next_open_angle]
455
+ if not self.is_inside_tool_call:
456
+ normal_text_chunks.append(text_segment)
457
+ # [FIX] If inside tool call, discard whitespace/text before Tag
458
+ self.parsed_pos += next_open_angle
459
+ continue
460
+
461
+ # Memory Cleanup: Slice the buffer
462
+ # Keep unparsed part, discard parsed part
463
+ if self.parsed_pos > 0:
464
+ self._buffer = self._buffer[self.parsed_pos :]
465
+ self.parsed_pos = 0
466
+
467
+ normal_text = "".join(normal_text_chunks) if normal_text_chunks else ""
468
+ return StreamingParseResult(calls=calls, normal_text=normal_text)
469
+
470
+ def supports_structural_tag(self) -> bool:
471
+ return False
472
+
473
+ def structure_info(self) -> _GetInfoFunc:
474
+ raise NotImplementedError
qwen3coder_tool_parser_vllm.py ADDED
@@ -0,0 +1,690 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ maybe_convert = False if "." in param_value or "e" in param_value.lower() else True
160
+ param_value: float = float(param_value)
161
+ if maybe_convert and param_value.is_integer():
162
+ param_value = int(param_value)
163
+ except:
164
+ logger.warning(
165
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not a float in tool "
166
+ f"'{func_name}', degenerating to string.")
167
+ return param_value
168
+ elif param_type in ["boolean", "bool", "binary"]:
169
+ param_value = param_value.lower()
170
+ if param_value not in ["true", "false"]:
171
+ logger.warning(
172
+ f"Parsed value '{param_value}' of parameter '{param_name}' is not a boolean (`true` of `false`) in tool '{func_name}', degenerating to false."
173
+ )
174
+ return param_value == "true"
175
+ else:
176
+ if param_type in ["object", "array", "arr"
177
+ ] or param_type.startswith(
178
+ "dict") or param_type.startswith("list"):
179
+ try:
180
+ param_value = json.loads(param_value)
181
+ return param_value
182
+ except:
183
+ logger.warning(
184
+ f"Parsed value '{param_value}' of parameter '{param_name}' cannot be parsed with json.loads in tool "
185
+ f"'{func_name}', will try other methods to parse it.")
186
+ try:
187
+ param_value = ast.literal_eval(param_value) # safer
188
+ except:
189
+ logger.warning(
190
+ f"Parsed value '{param_value}' of parameter '{param_name}' cannot be converted via Python `ast.literal_eval()` in tool '{func_name}', degenerating to string."
191
+ )
192
+ return param_value
193
+
194
+ def _parse_xml_function_call(
195
+ self, function_call_str: str,
196
+ tools: Optional[list[ChatCompletionToolsParam]]
197
+ ) -> Optional[ToolCall]:
198
+
199
+ # Extract function name
200
+ end_index = function_call_str.index(">")
201
+ function_name = function_call_str[:end_index]
202
+ param_config = self._get_arguments_config(function_name, tools)
203
+ parameters = function_call_str[end_index + 1:]
204
+ param_dict = {}
205
+ for match_text in self.tool_call_parameter_regex.findall(parameters):
206
+ idx = match_text.index(">")
207
+ param_name = match_text[:idx]
208
+ param_value = str(match_text[idx + 1:])
209
+ # Remove prefix and trailing \n
210
+ if param_value.startswith("\n"):
211
+ param_value = param_value[1:]
212
+ if param_value.endswith("\n"):
213
+ param_value = param_value[:-1]
214
+
215
+ param_dict[param_name] = self._convert_param_value(
216
+ param_value, param_name, param_config, function_name)
217
+ return ToolCall(
218
+ type="function",
219
+ function=FunctionCall(name=function_name,
220
+ arguments=json.dumps(param_dict,
221
+ ensure_ascii=False)),
222
+ )
223
+
224
+ def _get_function_calls(self, model_output: str) -> List[str]:
225
+ # Find all tool calls
226
+ matched_ranges = self.tool_call_regex.findall(model_output)
227
+ raw_tool_calls = [
228
+ match[0] if match[0] else match[1] for match in matched_ranges
229
+ ]
230
+
231
+ # Back-off strategy if no tool_call tags found
232
+ if len(raw_tool_calls) == 0:
233
+ raw_tool_calls = [model_output]
234
+
235
+ raw_function_calls = []
236
+ for tool_call in raw_tool_calls:
237
+ raw_function_calls.extend(
238
+ self.tool_call_function_regex.findall(tool_call))
239
+
240
+ function_calls = [
241
+ match[0] if match[0] else match[1] for match in raw_function_calls
242
+ ]
243
+ return function_calls
244
+
245
+ def extract_tool_calls(
246
+ self,
247
+ model_output: str,
248
+ request: ChatCompletionRequest,
249
+ ) -> ExtractedToolCallInformation:
250
+ # Quick check to avoid unnecessary processing
251
+ if self.tool_call_prefix not in model_output:
252
+ return ExtractedToolCallInformation(tools_called=False,
253
+ tool_calls=[],
254
+ content=model_output)
255
+
256
+ try:
257
+ function_calls = self._get_function_calls(model_output)
258
+ if len(function_calls) == 0:
259
+ return ExtractedToolCallInformation(tools_called=False,
260
+ tool_calls=[],
261
+ content=model_output)
262
+
263
+ tool_calls = [
264
+ self._parse_xml_function_call(function_call_str, request.tools)
265
+ for function_call_str in function_calls
266
+ ]
267
+
268
+ # Populate prev_tool_call_arr for serving layer to set finish_reason
269
+ self.prev_tool_call_arr.clear() # Clear previous calls
270
+ for tool_call in tool_calls:
271
+ if tool_call:
272
+ self.prev_tool_call_arr.append({
273
+ "name":
274
+ tool_call.function.name,
275
+ "arguments":
276
+ tool_call.function.arguments,
277
+ })
278
+
279
+ # Extract content before tool calls
280
+ content_index = model_output.find(self.tool_call_start_token)
281
+ content_index = content_index if content_index >= 0 else model_output.find(
282
+ self.tool_call_prefix)
283
+ content = model_output[:content_index] # .rstrip()
284
+
285
+ return ExtractedToolCallInformation(
286
+ tools_called=(len(tool_calls) > 0),
287
+ tool_calls=tool_calls,
288
+ content=content if content else None,
289
+ )
290
+
291
+ except Exception:
292
+ logger.exception("Error in extracting tool call from response.")
293
+ return ExtractedToolCallInformation(tools_called=False,
294
+ tool_calls=[],
295
+ content=model_output)
296
+
297
+ def extract_tool_calls_streaming(
298
+ self,
299
+ previous_text: str,
300
+ current_text: str,
301
+ delta_text: str,
302
+ previous_token_ids: Sequence[int],
303
+ current_token_ids: Sequence[int],
304
+ delta_token_ids: Sequence[int],
305
+ request: ChatCompletionRequest,
306
+ ) -> Union[DeltaMessage, None]:
307
+ # Store request for type conversion
308
+ if not previous_text:
309
+ self._reset_streaming_state()
310
+ self.streaming_request = request
311
+
312
+ # If no delta text, return None unless it's an EOS token after tool calls
313
+ if not delta_text:
314
+ # Check if this is an EOS token after all tool calls are complete
315
+ # We check for tool calls in the text even if is_tool_call_started is False
316
+ # because it might have been reset after processing all tools
317
+ if delta_token_ids and self.tool_call_end_token_id not in delta_token_ids:
318
+ # Count complete tool calls
319
+ complete_calls = len(
320
+ self.tool_call_complete_regex.findall(current_text))
321
+
322
+ # If we have completed tool calls and populated prev_tool_call_arr
323
+ if complete_calls > 0 and len(self.prev_tool_call_arr) > 0:
324
+ # Check if all tool calls are closed
325
+ open_calls = current_text.count(
326
+ self.tool_call_start_token) - current_text.count(
327
+ self.tool_call_end_token)
328
+ if open_calls == 0:
329
+ # Return empty delta message to allow finish_reason processing
330
+ return DeltaMessage(content="")
331
+ elif not self.is_tool_call_started and current_text:
332
+ # This is a regular content response that's now complete
333
+ return DeltaMessage(content="")
334
+ return None
335
+
336
+ # Update accumulated text
337
+ self.accumulated_text = current_text
338
+
339
+ # Check if we need to advance to next tool
340
+ if self.json_closed and not self.in_function:
341
+ # Check if this tool call has ended
342
+ tool_ends = current_text.count(self.tool_call_end_token)
343
+ if tool_ends > self.current_tool_index:
344
+ # This tool has ended, advance to next
345
+ self.current_tool_index += 1
346
+ self.header_sent = False
347
+ self.param_count = 0
348
+ self.json_started = False
349
+ self.json_closed = False
350
+ self.accumulated_params = {}
351
+
352
+ # Check if there are more tool calls
353
+ tool_starts = current_text.count(self.tool_call_start_token)
354
+ if self.current_tool_index >= tool_starts:
355
+ # No more tool calls
356
+ self.is_tool_call_started = False
357
+ # Continue processing next tool
358
+ return None
359
+
360
+ # Handle normal content before tool calls
361
+ if not self.is_tool_call_started:
362
+ # Check if tool call is starting
363
+ if self.tool_call_start_token_id in delta_token_ids or self.tool_call_start_token in delta_text:
364
+ self.is_tool_call_started = True
365
+ # Return any content before the tool call
366
+ if self.tool_call_start_token in delta_text:
367
+ content_before = delta_text[:delta_text.index(
368
+ self.tool_call_start_token)]
369
+ if content_before:
370
+ return DeltaMessage(content=content_before)
371
+ return None
372
+ else:
373
+ # Check if we're between tool calls - skip whitespace
374
+ if current_text.rstrip().endswith(self.tool_call_end_token):
375
+ # We just ended a tool call, skip whitespace
376
+ if delta_text.strip() == "":
377
+ return None
378
+ # Normal content, no tool call
379
+ return DeltaMessage(content=delta_text)
380
+
381
+ # Check if we're between tool calls (waiting for next one)
382
+ # Count tool calls we've seen vs processed
383
+ tool_starts_count = current_text.count(self.tool_call_start_token)
384
+ if self.current_tool_index >= tool_starts_count:
385
+ # We're past all tool calls, shouldn't be here
386
+ return None
387
+
388
+ # We're in a tool call, find the current tool call portion
389
+ # Need to find the correct tool call based on current_tool_index
390
+ tool_starts = []
391
+ idx = 0
392
+ while True:
393
+ idx = current_text.find(self.tool_call_start_token, idx)
394
+ if idx == -1:
395
+ break
396
+ tool_starts.append(idx)
397
+ idx += len(self.tool_call_start_token)
398
+
399
+ if self.current_tool_index >= len(tool_starts):
400
+ # No more tool calls to process yet
401
+ return None
402
+
403
+ tool_start_idx = tool_starts[self.current_tool_index]
404
+ # Find where this tool call ends (or current position if not ended yet)
405
+ tool_end_idx = current_text.find(self.tool_call_end_token,
406
+ tool_start_idx)
407
+ if tool_end_idx == -1:
408
+ tool_text = current_text[tool_start_idx:]
409
+ else:
410
+ tool_text = current_text[tool_start_idx:tool_end_idx +
411
+ len(self.tool_call_end_token)]
412
+
413
+ # Looking for function header
414
+ if not self.header_sent:
415
+ if self.tool_call_prefix in tool_text:
416
+ func_start = tool_text.find(self.tool_call_prefix) + len(
417
+ self.tool_call_prefix)
418
+ func_end = tool_text.find(">", func_start)
419
+
420
+ if func_end != -1:
421
+ # Found complete function name
422
+ self.current_function_name = tool_text[func_start:func_end]
423
+ self.current_tool_id = self._generate_tool_call_id()
424
+ self.header_sent = True
425
+ self.in_function = True
426
+
427
+ # IMPORTANT: Add to prev_tool_call_arr immediately when we detect a tool call
428
+ # This ensures finish_reason="tool_calls" even if parsing isn't complete
429
+ already_added = any(
430
+ tool.get("name") == self.current_function_name
431
+ for tool in self.prev_tool_call_arr)
432
+ if not already_added:
433
+ self.prev_tool_call_arr.append({
434
+ "name": self.current_function_name,
435
+ "arguments":
436
+ "{}", # Placeholder, will be updated later
437
+ })
438
+
439
+ # Send header with function info
440
+ return DeltaMessage(tool_calls=[
441
+ DeltaToolCall(
442
+ index=self.current_tool_index,
443
+ id=self.current_tool_id,
444
+ function=DeltaFunctionCall(
445
+ name=self.current_function_name, arguments=""),
446
+ type="function",
447
+ )
448
+ ])
449
+ return None
450
+
451
+ # We've sent header, now handle function body
452
+ if self.in_function:
453
+ # Send opening brace if not sent yet
454
+ if not self.json_started and self.parameter_prefix not in delta_text:
455
+ self.json_started = True
456
+ return DeltaMessage(tool_calls=[
457
+ DeltaToolCall(
458
+ index=self.current_tool_index,
459
+ function=DeltaFunctionCall(arguments="{"),
460
+ )
461
+ ])
462
+
463
+ # Make sure json_started is set if we're processing parameters
464
+ if not self.json_started:
465
+ self.json_started = True
466
+
467
+ # Check for function end in accumulated text
468
+ if not self.json_closed and self.function_end_token in tool_text:
469
+ # Close JSON
470
+ self.json_closed = True
471
+
472
+ # Extract the complete tool call to update prev_tool_call_arr with final arguments
473
+ # Find the function content
474
+ func_start = tool_text.find(self.tool_call_prefix) + len(
475
+ self.tool_call_prefix)
476
+ func_content_end = tool_text.find(self.function_end_token,
477
+ func_start)
478
+ if func_content_end != -1:
479
+ func_content = tool_text[func_start:func_content_end]
480
+ # Parse to get the complete arguments
481
+ try:
482
+ parsed_tool = self._parse_xml_function_call(
483
+ func_content, self.streaming_request.tools
484
+ if self.streaming_request else None)
485
+ if parsed_tool:
486
+ # Update existing entry in prev_tool_call_arr with complete arguments
487
+ for i, tool in enumerate(self.prev_tool_call_arr):
488
+ if tool.get(
489
+ "name") == parsed_tool.function.name:
490
+ self.prev_tool_call_arr[i][
491
+ "arguments"] = parsed_tool.function.arguments
492
+ break
493
+ except Exception:
494
+ pass # Ignore parsing errors during streaming
495
+
496
+ result = DeltaMessage(tool_calls=[
497
+ DeltaToolCall(
498
+ index=self.current_tool_index,
499
+ function=DeltaFunctionCall(arguments="}"),
500
+ )
501
+ ])
502
+
503
+ # Reset state for next tool
504
+ self.in_function = False
505
+ self.json_closed = True
506
+ self.accumulated_params = {}
507
+
508
+ return result
509
+
510
+ # Look for parameters
511
+ # Find all parameter starts
512
+ param_starts = []
513
+ idx = 0
514
+ while True:
515
+ idx = tool_text.find(self.parameter_prefix, idx)
516
+ if idx == -1:
517
+ break
518
+ param_starts.append(idx)
519
+ idx += len(self.parameter_prefix)
520
+
521
+ # Check if we should start a new parameter
522
+ if not self.in_param and self.param_count < len(param_starts):
523
+
524
+ if len(param_starts) > self.param_count:
525
+ # Process the next parameter
526
+ param_idx = param_starts[self.param_count]
527
+ param_start = param_idx + len(self.parameter_prefix)
528
+ remaining = tool_text[param_start:]
529
+
530
+ if ">" in remaining:
531
+ # We have the complete parameter name
532
+ name_end = remaining.find(">")
533
+ self.current_param_name = remaining[:name_end]
534
+
535
+ # Find the parameter value
536
+ value_start = param_start + name_end + 1
537
+ value_text = tool_text[value_start:]
538
+ if value_text.startswith("\n"):
539
+ value_text = value_text[1:]
540
+
541
+ # Find where this parameter ends
542
+ param_end_idx = value_text.find(
543
+ self.parameter_end_token)
544
+ if param_end_idx == -1:
545
+ # No closing tag, look for next parameter or function end
546
+ next_param_idx = value_text.find(
547
+ self.parameter_prefix)
548
+ func_end_idx = value_text.find(
549
+ self.function_end_token)
550
+
551
+ if next_param_idx != -1 and (func_end_idx == -1
552
+ or next_param_idx
553
+ < func_end_idx):
554
+ param_end_idx = next_param_idx
555
+ elif func_end_idx != -1:
556
+ param_end_idx = func_end_idx
557
+ else:
558
+ # Neither found, check if tool call is complete
559
+ if self.tool_call_end_token in tool_text:
560
+ # Tool call is complete, so parameter must be complete too
561
+ # Use all remaining text before function end as value
562
+ param_end_idx = len(value_text)
563
+ else:
564
+ # Still streaming, wait for more content
565
+ return None
566
+
567
+ if param_end_idx != -1:
568
+ # Complete parameter found
569
+ param_value = value_text[:param_end_idx]
570
+ if param_value.endswith("\n"):
571
+ param_value = param_value[:-1]
572
+
573
+ # Store raw value for later processing
574
+ self.accumulated_params[
575
+ self.current_param_name] = param_value
576
+
577
+ # Get parameter configuration for type conversion
578
+ param_config = self._get_arguments_config(
579
+ self.current_function_name,
580
+ self.streaming_request.tools
581
+ if self.streaming_request else None)
582
+
583
+ # Convert the parameter value to the appropriate type
584
+ converted_value = self._convert_param_value(
585
+ param_value, self.current_param_name,
586
+ param_config, self.current_function_name)
587
+
588
+ # Build JSON fragment based on the converted type
589
+ # Use json.dumps to properly serialize the value
590
+ serialized_value = json.dumps(converted_value,
591
+ ensure_ascii=False)
592
+
593
+ if self.param_count == 0:
594
+ json_fragment = f'"{self.current_param_name}": {serialized_value}'
595
+ else:
596
+ json_fragment = f', "{self.current_param_name}": {serialized_value}'
597
+
598
+ self.param_count += 1
599
+
600
+ return DeltaMessage(tool_calls=[
601
+ DeltaToolCall(
602
+ index=self.current_tool_index,
603
+ function=DeltaFunctionCall(
604
+ arguments=json_fragment),
605
+ )
606
+ ])
607
+
608
+ # Continue parameter value - Not used in the current implementation
609
+ # since we process complete parameters above
610
+ if self.in_param:
611
+ if self.parameter_end_token in delta_text:
612
+ # End of parameter
613
+ end_idx = delta_text.find(self.parameter_end_token)
614
+ value_chunk = delta_text[:end_idx]
615
+
616
+ # Skip past > if at start
617
+ if not self.current_param_value and ">" in value_chunk:
618
+ gt_idx = value_chunk.find(">")
619
+ value_chunk = value_chunk[gt_idx + 1:]
620
+
621
+ if not self.current_param_value and value_chunk.startswith(
622
+ "\n"):
623
+ value_chunk = value_chunk[1:]
624
+
625
+ # Store complete value
626
+ full_value = self.current_param_value + value_chunk
627
+ self.accumulated_params[
628
+ self.current_param_name] = full_value
629
+
630
+ # Get parameter configuration for type conversion
631
+ param_config = self._get_arguments_config(
632
+ self.current_function_name,
633
+ self.streaming_request.tools
634
+ if self.streaming_request else None)
635
+
636
+ # Convert the parameter value to the appropriate type
637
+ converted_value = self._convert_param_value(
638
+ full_value, self.current_param_name, param_config,
639
+ self.current_function_name)
640
+
641
+ # Serialize the converted value
642
+ serialized_value = json.dumps(converted_value,
643
+ ensure_ascii=False)
644
+
645
+ # Since we've been streaming the quoted version, we need to close it properly
646
+ # This is complex - for now just complete the value
647
+ self.in_param = False
648
+ self.current_param_value = ""
649
+
650
+ # Just close the current parameter string
651
+ return DeltaMessage(tool_calls=[
652
+ DeltaToolCall(
653
+ index=self.current_tool_index,
654
+ function=DeltaFunctionCall(
655
+ arguments='"'), # Close the string quote
656
+ )
657
+ ])
658
+ else:
659
+ # Continue accumulating value
660
+ value_chunk = delta_text
661
+
662
+ # Handle first chunk after param name
663
+ if not self.current_param_value and ">" in value_chunk:
664
+ gt_idx = value_chunk.find(">")
665
+ value_chunk = value_chunk[gt_idx + 1:]
666
+
667
+ if not self.current_param_value and value_chunk.startswith(
668
+ "\n"):
669
+ value_chunk = value_chunk[1:]
670
+
671
+ if value_chunk:
672
+ # Stream the escaped delta
673
+ prev_escaped = json.dumps(
674
+ self.current_param_value, ensure_ascii=False
675
+ )[1:-1] if self.current_param_value else ""
676
+ self.current_param_value += value_chunk
677
+ full_escaped = json.dumps(self.current_param_value,
678
+ ensure_ascii=False)[1:-1]
679
+ delta_escaped = full_escaped[len(prev_escaped):]
680
+
681
+ if delta_escaped:
682
+ return DeltaMessage(tool_calls=[
683
+ DeltaToolCall(
684
+ index=self.current_tool_index,
685
+ function=DeltaFunctionCall(
686
+ arguments=delta_escaped),
687
+ )
688
+ ])
689
+
690
+ return None
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "151643": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "151644": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "151645": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "151646": {
29
+ "content": "<|object_ref_start|>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "151647": {
37
+ "content": "<|object_ref_end|>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "151648": {
45
+ "content": "<|box_start|>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "151649": {
53
+ "content": "<|box_end|>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "151650": {
61
+ "content": "<|quad_start|>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "151651": {
69
+ "content": "<|quad_end|>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "151652": {
77
+ "content": "<|vision_start|>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "151653": {
85
+ "content": "<|vision_end|>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "151654": {
93
+ "content": "<|vision_pad|>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "151655": {
101
+ "content": "<|image_pad|>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "151656": {
109
+ "content": "<|video_pad|>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "151657": {
117
+ "content": "<tool_call>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": false
123
+ },
124
+ "151658": {
125
+ "content": "</tool_call>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": false
131
+ },
132
+ "151659": {
133
+ "content": "<|fim_prefix|>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": false
139
+ },
140
+ "151660": {
141
+ "content": "<|fim_middle|>",
142
+ "lstrip": false,
143
+ "normalized": false,
144
+ "rstrip": false,
145
+ "single_word": false,
146
+ "special": false
147
+ },
148
+ "151661": {
149
+ "content": "<|fim_suffix|>",
150
+ "lstrip": false,
151
+ "normalized": false,
152
+ "rstrip": false,
153
+ "single_word": false,
154
+ "special": false
155
+ },
156
+ "151662": {
157
+ "content": "<|fim_pad|>",
158
+ "lstrip": false,
159
+ "normalized": false,
160
+ "rstrip": false,
161
+ "single_word": false,
162
+ "special": false
163
+ },
164
+ "151663": {
165
+ "content": "<|repo_name|>",
166
+ "lstrip": false,
167
+ "normalized": false,
168
+ "rstrip": false,
169
+ "single_word": false,
170
+ "special": false
171
+ },
172
+ "151664": {
173
+ "content": "<|file_sep|>",
174
+ "lstrip": false,
175
+ "normalized": false,
176
+ "rstrip": false,
177
+ "single_word": false,
178
+ "special": false
179
+ },
180
+ "151665": {
181
+ "content": "<tool_response>",
182
+ "lstrip": false,
183
+ "normalized": false,
184
+ "rstrip": false,
185
+ "single_word": false,
186
+ "special": false
187
+ },
188
+ "151666": {
189
+ "content": "</tool_response>",
190
+ "lstrip": false,
191
+ "normalized": false,
192
+ "rstrip": false,
193
+ "single_word": false,
194
+ "special": false
195
+ },
196
+ "151667": {
197
+ "content": "<think>",
198
+ "lstrip": false,
199
+ "normalized": false,
200
+ "rstrip": false,
201
+ "single_word": false,
202
+ "special": false
203
+ },
204
+ "151668": {
205
+ "content": "</think>",
206
+ "lstrip": false,
207
+ "normalized": false,
208
+ "rstrip": false,
209
+ "single_word": false,
210
+ "special": false
211
+ }
212
+ },
213
+ "additional_special_tokens": [
214
+ "<|im_start|>",
215
+ "<|im_end|>",
216
+ "<|object_ref_start|>",
217
+ "<|object_ref_end|>",
218
+ "<|box_start|>",
219
+ "<|box_end|>",
220
+ "<|quad_start|>",
221
+ "<|quad_end|>",
222
+ "<|vision_start|>",
223
+ "<|vision_end|>",
224
+ "<|vision_pad|>",
225
+ "<|image_pad|>",
226
+ "<|video_pad|>"
227
+ ],
228
+ "bos_token": null,
229
+ "chat_template": "{% macro render_extra_keys(json_dict, handled_keys) %}\n {%- if json_dict is mapping %}\n {%- for json_key in json_dict if json_key not in handled_keys %}\n {%- if json_dict[json_key] is string %}\n {{-'\\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' }}\n {%- else %}\n {{- '\\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ json_key ~ '>' }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n{%- endmacro %}\n\n{%- if messages[0][\"role\"] == \"system\" %}\n {%- set system_message = messages[0][\"content\"] %}\n {%- set loop_messages = messages[1:] %}\n{%- else %}\n {%- set loop_messages = messages %}\n{%- endif %}\n\n{%- if not tools is defined %}\n {%- set tools = [] %}\n{%- endif %}\n\n{%- if system_message is defined %}\n {{- \"<|im_start|>system\\n\" + system_message }}\n{%- else %}\n {%- if tools is iterable and tools | length > 0 %}\n {{- \"<|im_start|>system\\nYou are Qwen, a helpful AI assistant that can interact with a computer to solve tasks.\" }}\n {%- endif %}\n{%- endif %}\n{%- if tools is iterable and tools | length > 0 %}\n {{- \"\\n\\n# Tools\\n\\nYou have access to the following functions:\\n\\n\" }}\n {{- \"<tools>\" }}\n {%- for tool in tools %}\n {%- if tool.function is defined %}\n {%- set tool = tool.function %}\n {%- endif %}\n {{- \"\\n<function>\\n<name>\" ~ tool.name ~ \"</name>\" }}\n {%- if tool.description is defined %}\n {{- '\\n<description>' ~ (tool.description | trim) ~ '</description>' }}\n {%- endif %}\n {{- '\\n<parameters>' }}\n {%- if tool.parameters is defined and tool.parameters is mapping and tool.parameters.properties is defined and tool.parameters.properties is mapping %}\n {%- for param_name, param_fields in tool.parameters.properties|items %}\n {{- '\\n<parameter>' }}\n {{- '\\n<name>' ~ param_name ~ '</name>' }}\n {%- if param_fields.type is defined %}\n {{- '\\n<type>' ~ (param_fields.type | string) ~ '</type>' }}\n {%- endif %}\n {%- if param_fields.description is defined %}\n {{- '\\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}\n {%- endif %}\n {%- set handled_keys = ['name', 'type', 'description'] %}\n {{- render_extra_keys(param_fields, handled_keys) }}\n {{- '\\n</parameter>' }}\n {%- endfor %}\n {%- endif %}\n {%- set handled_keys = ['type', 'properties'] %}\n {{- render_extra_keys(tool.parameters, handled_keys) }}\n {{- '\\n</parameters>' }}\n {%- set handled_keys = ['type', 'name', 'description', 'parameters'] %}\n {{- render_extra_keys(tool, handled_keys) }}\n {{- '\\n</function>' }}\n {%- endfor %}\n {{- \"\\n</tools>\" }}\n {{- '\\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>' }}\n{%- endif %}\n{%- if system_message is defined %}\n {{- '<|im_end|>\\n' }}\n{%- else %}\n {%- if tools is iterable and tools | length > 0 %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- for message in loop_messages %}\n {%- if message.role == \"assistant\" and message.tool_calls is defined and message.tool_calls is iterable and message.tool_calls | length > 0 %}\n {{- '<|im_start|>' + message.role }}\n {%- if message.content is defined and message.content is string and message.content | trim | length > 0 %}\n {{- '\\n' + message.content | trim + '\\n' }}\n {%- endif %}\n {%- for tool_call in message.tool_calls %}\n {%- if tool_call.function is defined %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n {%- if tool_call.arguments is defined %}\n {%- for args_name, args_value in tool_call.arguments|items %}\n {{- '<parameter=' + args_name + '>\\n' }}\n {%- set args_value = args_value if args_value is string else args_value | tojson | safe %}\n {{- args_value }}\n {{- '\\n</parameter>\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '</function>\\n</tool_call>' }}\n {%- endfor %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"user\" or message.role == \"system\" or message.role == \"assistant\" %}\n {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {{- message.content }}\n {{- '\\n</tool_response>' }}\n {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n {{- '<|im_end|>\\n' }}\n {%- elif loop.last %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>\\n' }}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n{%- endif %}",
230
+ "clean_up_tokenization_spaces": false,
231
+ "eos_token": "<|im_end|>",
232
+ "errors": "replace",
233
+ "model_max_length": 1048576,
234
+ "pad_token": "<|endoftext|>",
235
+ "split_special_tokens": false,
236
+ "tokenizer_class": "Qwen2Tokenizer",
237
+ "unk_token": null,
238
+ "add_bos_token": false
239
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff