omni-route/config.example.yaml at main · attnstack/omni-route · GitHub

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
# omni-route 生产示例配置：OpenAI + DeepSeek + Anthropic + 本地 Ollama。
# key 一律走环境变量（key_env），禁止写进配置文件。
listen: ":8080"
# 网关自身鉴权：客户端请求需带 Authorization: Bearer <以下之一>
api_keys: ["sk-or-local-dev-key"]
admin_key_env: OMNI_ADMIN_KEY   # /admin/reload 需要该环境变量的值
seed: 0                         # 0 = 每次启动随机；>0 则路由决策可复现

log:
  level: info
  format: json

store:
  path: data/omni.db
  flush_interval_ms: 200
  batch_size: 64

providers:
  openai:
    kind: openai
    base_url: https://api.openai.com/v1
    key_env: OPENAI_API_KEY
    rpm: 500
  deepseek:
    kind: openai
    base_url: https://api.deepseek.com/v1
    key_env: DEEPSEEK_API_KEY
    rpm: 300
  anthropic:
    kind: anthropic
    base_url: https://api.anthropic.com
    key_env: ANTHROPIC_API_KEY
    rpm: 100
    default_max_tokens: 8192
  ollama:
    kind: openai
    base_url: http://localhost:11434/v1
    key_env: ""          # 本地服务无鉴权
    rpm: 600

models:
  gpt-4o:
    provider: openai
    price_in: 2.50
    price_out: 10.00
    caps: { streaming: true, tools: true, json_mode: true, max_context_tokens: 128000 }
  gpt-4o-mini:
    provider: openai
    price_in: 0.15
    price_out: 0.60
    caps: { streaming: true, tools: true, json_mode: true, max_context_tokens: 128000 }
  deepseek-chat:
    provider: deepseek
    price_in: 0.27
    price_out: 1.10
    caps: { streaming: true, tools: true, json_mode: true, max_context_tokens: 64000 }
  deepseek-reasoner:
    provider: deepseek
    price_in: 0.55
    price_out: 2.19
    caps: { streaming: true, tools: false, json_mode: false, max_context_tokens: 64000 }
  claude-sonnet:
    provider: anthropic
    price_in: 3.00
    price_out: 15.00
    caps: { streaming: true, tools: true, json_mode: false, max_context_tokens: 200000 }
  qwen2.5:7b:
    provider: ollama
    price_in: 0            # 本地模型无边际成本
    price_out: 0
    caps: { streaming: true, tools: false, json_mode: true, max_context_tokens: 32768 }

routing:
  mode: cascade
  default_model: gpt-4o-mini
  task_defaults:
    reasoning: deepseek-reasoner
  budget_per_task:
    code: 5.00
    reasoning: 8.00
    summarization: 1.50
    translation: 1.50
    extraction: 1.50
    chat: 1.00
  baseline:
    code: claude-sonnet
    reasoning: claude-sonnet
    summarization: claude-sonnet
    translation: claude-sonnet
    extraction: claude-sonnet
    chat: claude-sonnet
  # 数据合规组：请求带 x-or-data-class: sensitive 时只允许发往组内 provider
  compliance_groups:
    sensitive: [ollama]
  latency_budget_ms: 30000
  spend_window: 1000
  lambda: { init: 50, min: 1, max: 5000, eta: 0.05 }

cascades:
  code: [deepseek-chat, claude-sonnet]
  reasoning: [deepseek-reasoner, claude-sonnet]
  summarization: [gpt-4o-mini, deepseek-chat]
  translation: [deepseek-chat, gpt-4o]
  extraction: [gpt-4o-mini, deepseek-chat]
  chat: [qwen2.5:7b, gpt-4o-mini, gpt-4o]

cascade:
  max_upgrades: 2
  max_latency_ms: 15000
  target_upgrade_rate: 0.30
  calibration_step: 0.05
  calibration_window: 50
  confidence:
    threshold: 0.60
    use_format: true
    use_heuristic: true
    llm_judge:
      enabled: true
      model: gpt-4o-mini
      sample_rate: 0.05
      timeout_ms: 8000
    self_consistency:
      enabled: false
      k: 3

reliability:
  retries: 2
cache:
  enabled: true
  mode: zero_temp        # 只缓存显式 temperature=0 的请求
  ttl_seconds: 3600
  max_entries: 4096

  backoff_base_ms: 200
  breaker:
    window_sec: 300
    min_samples: 20
    error_rate: 0.20
    open_sec: 30

timeouts:
  connect_ms: 5000
  first_byte_ms: 60000
  total_ms: 300000