litellm-gateway/config.yaml

72 lines
2.3 KiB
YAML
Raw Normal View History

# ============================================
# LiteLLM Gateway — config.yaml
# ============================================
# MiniMax 模型路由配置
# 完整文档: https://docs.litellm.ai/docs/proxy/configs
model_list:
# ── MiniMax M2.7 (优先) ──────────────────────
- model_name: MiniMax-M2.7
litellm_params:
model: minimax/MiniMax-M2.7
api_key: os.environ/MINIMAX_API_KEY
api_base: https://api.minimax.io/v1
rpm: 60
model_info:
mode: chat
supports_function_calling: true
# ── MiniMax M2.5 (备用) ──────────────────────
- model_name: MiniMax-M2.5
litellm_params:
model: minimax/MiniMax-M2.5
api_key: os.environ/MINIMAX_API_KEY
api_base: https://api.minimax.io/v1
rpm: 60
model_info:
mode: chat
supports_function_calling: true
# ── MiniMax M2.5-Lightning (兜底) ───────────
- model_name: MiniMax-M2.5-Lightning
litellm_params:
model: minimax/MiniMax-M2.5-Lightning
api_key: os.environ/MINIMAX_API_KEY
api_base: https://api.minimax.io/v1
rpm: 120
model_info:
mode: chat
# ── 路由设置 ───────────────────────────────────
router_settings:
# 失败后按顺序回落
fallback_params:
- model:
- MiniMax-M2.7
- MiniMax-M2.5
- MiniMax-M2.5-Lightning
# 每个模型重试次数
num_retries: 3
# 重试前等待秒数
retry_after: 5
# 单次请求超时(秒)
timeout: 30
# 路由策略:优先用最好的模型
routing_strategy: latency-based-routing
# ── 全局设置 ───────────────────────────────────
litellm_settings:
drop_params: true
set_verbose: false
# 每次请求记录到数据库
success_callback: ["database"]
# ── 服务器设置 ─────────────────────────────────
general_settings:
# 管理员 master key必须以 sk- 开头)
master_key: os.environ/LITELLM_MASTER_KEY
# 数据库SQLite 轻量版,生产推荐 PostgreSQL
database_url: "sqlite:///./litellm.db"
# UI 管理界面
ui_access_mode: "admin"