v34

latestOpenAPI 3.1.0Apache-2.0raw.githubusercontent.com2026-08-0332109227.7 KB

inference

`/v1/completions` — legacy OpenAI Completions (single-candidate, raw-prompt).

OpenAI-compatible legacy Completions. Strict allow-list parser: unknown top-level fields reject with 400 unsupported_field. stream: true is supported (SSE text_completion). Known-rejected fields: echo, suffix, logprobs, best_of, n > 1, batched array prompt — each rejects with 400 unsupported_field. The response body no longer carries the always-null logprobs field.

post/v1/completions

Request body

frequency_penaltynumber float nullable

In [-2.0, 2.0]; out-of-range or non-numeric values yield 400.

max_tokensinteger nullable

Positive integer; defaults to 16 (OpenAI's documented default for completions) when absent.

modelstring required

ninteger nullable

Single-candidate only: integer 1 (or absent) accepted; n > 1 rejects with 400 unsupported_field (use chat for multi-candidate). n == 0 rejects with 400 invalid_request.

presence_penaltynumber float nullable

In [-2.0, 2.0]; out-of-range or non-numeric values yield 400.

promptstring required

Single prompt string. Batched array prompts reject with 400 unsupported_field.

seedinteger nullable

Optional signed 64-bit per-request sampling seed. Reproducibility is best effort, not guaranteed, and depends on the active generation backend and deployment configuration. Non-integer or out-of-range values reject with 400 invalid_request.

streamboolean nullable

SSE streaming. When true, the response is a stream of text_completion events terminated by data: [DONE]. Non-boolean values reject with 400 invalid_request.

temperaturenumber float nullable

Sampling temperature. Finite number >= 0.

top_pnumber float nullable

Nucleus sampling. Finite number in (0, 1].

Response

Text completion response