# LiteLLM 网关配置文件
# 所有渠道和模型优先级、Failover 规则在此定义
# API Key 通过环境变量注入，见 .env 文件

model_list:
  # 聚合模型1: auto-vision (图文混合问答)
  # 按优先级依次尝试: Gemini(1) → NVIDIA(1) → OpenRouter(2) → Mistral(3)
  # (Groq 不支持视觉，故不在此列表)
  - model_name: auto-vision
    litellm_params:
      model: gemini/gemini-flash-latest
      api_key: ${GEMINI_API_KEY}
      api_base: https://generativelanguage.googleapis.com/v1beta
    fallbacks:
      - nvidia-nim-vision
      - openrouter-vision
      - mistral-vision
    tpm_limit: 10000
    rpm_limit: 100

  - model_name: nvidia-nim-vision
    litellm_params:
      model: openai/meta/llama-3.2-11b-vision-instruct
      api_key: ${NVIDIA_API_KEY}
      api_base: https://integrate.api.nvidia.com/v1

  - model_name: openrouter-vision
    litellm_params:
      model: openrouter/google/gemini-2.0-flash-exp:free
      api_key: ${OPENROUTER_API_KEY}
      api_base: https://openrouter.ai/api/v1

  - model_name: mistral-vision
    litellm_params:
      model: openai/pixtral-12b-2409
      api_key: ${MISTRAL_API_KEY}
      api_base: https://api.mistral.ai/v1

  # 聚合模型2: auto-text (纯文本问答)
  # 按优先级依次尝试: Gemini(1) → NVIDIA(1) → OpenRouter(2) → Groq(2) → Mistral(3)
  - model_name: auto-text
    litellm_params:
      model: gemini/gemini-flash-latest
      api_key: ${GEMINI_API_KEY}
      api_base: https://generativelanguage.googleapis.com/v1beta
    fallbacks:
      - nvidia-nim-text
      - openrouter-text
      - groq-text
      - mistral-text
    tpm_limit: 10000
    rpm_limit: 100

  - model_name: nvidia-nim-text
    litellm_params:
      model: openai/meta/llama-3.2-11b-vision-instruct
      api_key: ${NVIDIA_API_KEY}
      api_base: https://integrate.api.nvidia.com/v1

  - model_name: openrouter-text
    litellm_params:
      model: openrouter/google/gemini-2.0-flash-exp:free
      api_key: ${OPENROUTER_API_KEY}
      api_base: https://openrouter.ai/api/v1

  - model_name: groq-text
    litellm_params:
      model: openai/llama-3.3-70b-versatile
      api_key: ${GROQ_API_KEY}
      api_base: https://api.groq.com/openai/v1

  - model_name: mistral-text
    litellm_params:
      model: openai/mistral-small-latest
      api_key: ${MISTRAL_API_KEY}
      api_base: https://api.mistral.ai/v1

# 网关全局配置
router_settings:
  enable_cooldowns: true
  cooldown_threshold: 5  # 连续失败 5 次进入冷却
  cooldown_duration: 600  # 冷却期 10 分钟
  enable_retry: true
  max_retries: 2  # 最多重试 2 次

# 日志级别
logging:
  level: INFO

# Fallback 行为
# 上游错误(429/5xx/timeout)时自动切换到下一个渠道
fallback_error_codes:
  - 429
  - 502
  - 503
  - 504
  - timeout