server_openai_definitions.yaml

# OpenAI 兼容层 schema，供 server.yaml 通过 $ref 引用
swagger: '2.0'
info:
  title: OpenAPI fragments — OpenAI-compatible schemas
  version: "0.0"
definitions:

  # --- Completions /v1/completions ---

  # POST /v1/completions/prompt：将用户原文套用 chat template，返回实际送入续写接口的字符串
  OpenAICompletionsPromptRequest:
    type: object
    additionalProperties: true
    required:
      - model
      - prompt
    properties:
      model:
        type: string
        description: Model ID
      prompt:
        description: 用户输入原文（非已格式化全文）
        type: string
      system:
        description: 可选；chat template 的 system 段。省略时不拼装 system（仅 user）；传入字符串时原样作为 system 段（含空字符串、仅空白等）
        type: string
      enable_thinking:
        description: 可选；是否启用 Qwen3 chat template 的 thinking 模式；省略或 false 时不插入 thinking 块
        type: boolean

  OpenAICompletionsPromptResponse:
    type: object
    required:
      - prompt_used
    properties:
      prompt_used:
        type: string
        description: 套用 chat template 后、与 POST /v1/completions 的 prompt 字段同义的完整模型输入

  OpenAICompletionsRequest:
    type: object
    additionalProperties: true
    required:
      - model
      - prompt
    properties:
      model:
        type: string
        description: Model ID
      prompt:
        description: 已确定的完整模型输入（需 chat template 时请先 POST /v1/completions/prompt）
        type: string
      best_of:
        type: integer
        minimum: 0
        maximum: 20
      echo:
        type: boolean
      frequency_penalty:
        type: number
        minimum: -2
        maximum: 2
      logit_bias:
        type: object
        additionalProperties:
          type: number
      logprobs:
        type: integer
        minimum: 0
        maximum: 5
      max_tokens:
        description: 可选；正整数（> 0），且与 prompt 合计不超过续写接口的上下文 token 上限；省略则用满剩余额度
        type: integer
        minimum: 1
      n:
        type: integer
        minimum: 1
        maximum: 128
      presence_penalty:
        type: number
        minimum: -2
        maximum: 2
      seed:
        type: integer
        format: int64
      stop:
        description: 停止序列，最多 4 个
        type:
          - string
          - array
        items:
          type: string
        maxItems: 4
      stream:
        type: boolean
        description: 本服务忽略；响应恒为 SSE
      stream_options:
        type: object
        properties:
          include_usage:
            type: boolean
      suffix:
        type: string
      temperature:
        type: number
        minimum: 0
        maximum: 2
      top_p:
        type: number
        minimum: 0
        maximum: 1
      user:
        type: string

  OpenAICompletionUsage:
    type: object
    properties:
      prompt_tokens:
        type: integer
      completion_tokens:
        type: integer
      total_tokens:
        type: integer
      completion_tokens_details:
        type: object
        additionalProperties: true
      prompt_tokens_details:
        type: object
        additionalProperties: true

  OpenAICompletionChoice:
    type: object
    properties:
      text:
        type: string
      index:
        type: integer
      logprobs:
        description: 无 logprobs 时为 null；否则含 text_offset、token_logprobs、tokens、top_logprobs
        type: object
        x-nullable: true
      finish_reason:
        type: string
        enum:
          - stop
          - length
          - content_filter
        x-nullable: true

  # bpe_strings 项与 server.yaml#/definitions/TokenWithOffset 一致（跨文件复用）
  InfoRadarCompletionPayload:
    type: object
    properties:
      bpe_strings:
        type: array
        items:
          $ref: "server.yaml#/definitions/TokenWithOffset"

  OpenAICompletionsResponse:
    type: object
    required:
      - id
      - object
      - created
      - model
      - choices
    properties:
      id:
        type: string
      object:
        type: string
        enum:
          - text_completion
      created:
        type: integer
        description: Unix 时间戳（秒）
      model:
        type: string
      choices:
        type: array
        items:
          $ref: "#/definitions/OpenAICompletionChoice"
      system_fingerprint:
        type: string
      usage:
        $ref: "#/definitions/OpenAICompletionUsage"
      info_radar:
        $ref: "#/definitions/InfoRadarCompletionPayload"
        description: 续写 token 级概率与 top-N，便于与信息密度可视化对齐