vLLM Scoring API

openapi: 3.1.0
info:
  title: vLLM OpenAI-Compatible Server Audio Scoring API
  version: '1'
  description: 'vLLM is a high-throughput open-source inference and serving engine for

    LLMs. Running `vllm serve` exposes an OpenAI-compatible REST API plus

    vLLM-specific endpoints. Authentication is via a server-startup

    `--api-key` flag; clients supply it as a Bearer token in the

    Authorization header (matching the OpenAI Python client).

    '
  contact:
    name: vLLM Documentation
    url: https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html
  license:
    name: Apache-2.0
    url: https://www.apache.org/licenses/LICENSE-2.0
servers:
- url: http://{host}:{port}
  description: Local vLLM server
  variables:
    host:
      default: localhost
    port:
      default: '8000'
security:
- bearerAuth: []
tags:
- name: Scoring
  description: vLLM-specific scoring and reranking endpoints
paths:
  /v1/score:
    post:
      tags:
      - Scoring
      summary: Score with a cross-encoder model
      operationId: score
      responses:
        '200':
          description: Score
  /v1/rerank:
    post:
      tags:
      - Scoring
      summary: Rerank results
      operationId: rerank
      responses:
        '200':
          description: Rerank results
  /classify:
    post:
      tags:
      - Scoring
      summary: Classify input
      operationId: classify
      responses:
        '200':
          description: Classification
  /pooling:
    post:
      tags:
      - Scoring
      summary: Pooling operation
      operationId: pooling
      responses:
        '200':
          description: Pooling result
  /generative_scoring:
    post:
      tags:
      - Scoring
      summary: Score items using a generative model
      operationId: generativeScoring
      responses:
        '200':
          description: Scoring result
components:
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      description: API key supplied at server startup via --api-key