-
Notifications
You must be signed in to change notification settings - Fork 2.1k
Expand file tree
/
Copy pathvllm-qwen3.5-27b-fp8.provider.yml
More file actions
174 lines (161 loc) · 3.42 KB
/
vllm-qwen3.5-27b-fp8.provider.yml
File metadata and controls
174 lines (161 loc) · 3.42 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
# Qwen3.5-27B FP8 Provider Configuration - THINKING MODE (default)
# Based on official Qwen recommendations for vLLM inference
# Architecture: Hybrid 75% DeltaNet + 25% Full Attention (48+16 layers)
# Context: 262K native, expandable to 1M with YaRN
# Vision: VLM with Vision Encoder (uses VRAM even for text-only tasks)
#
# Thinking mode is enabled by default (no extra_body needed)
# Recommended sampling parameters:
# - General tasks: temp=1.0, top_p=0.95, top_k=20, min_p=0.0, pp=1.5, rp=1.0
# - Precise coding: temp=0.6, top_p=0.95, top_k=20, min_p=0.0, pp=0.0, rp=1.0
#
# Non-thinking mode is disabled via extra_body parameter
# Recommended sampling parameters:
# - General tasks: temp=0.7, top_p=0.8, top_k=20, min_p=0.0, pp=1.5, rp=1.0
# - Reasoning tasks: temp=1.0, top_p=0.95, top_k=20, min_p=0.0, pp=1.5, rp=1.0
simple:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.7
top_k: 20
top_p: 0.8
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
extra_body:
chat_template_kwargs:
enable_thinking: false
simple_json:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.7
top_k: 20
top_p: 0.8
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
json: true
extra_body:
chat_template_kwargs:
enable_thinking: false
primary_agent:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 1.0
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
assistant:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 1.0
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
generator:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 1.0
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
refiner:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 1.0
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
adviser:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 1.0
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
reflector:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 1.0
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
extra_body:
chat_template_kwargs:
enable_thinking: false
searcher:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.7
top_k: 20
top_p: 0.8
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
extra_body:
chat_template_kwargs:
enable_thinking: false
enricher:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.7
top_k: 20
top_p: 0.8
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
n: 1
max_tokens: 32768
extra_body:
chat_template_kwargs:
enable_thinking: false
coder:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.6
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 0.0
repetition_penalty: 1.0
n: 1
max_tokens: 32768
installer:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.6
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 0.0
repetition_penalty: 1.0
n: 1
max_tokens: 32768
pentester:
model: "Qwen/Qwen3.5-27B-FP8"
temperature: 0.6
top_k: 20
top_p: 0.95
min_p: 0.0
presence_penalty: 0.0
repetition_penalty: 1.0
n: 1
max_tokens: 32768