v34

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

OpenAI-compatible embeddings proxy. A 200 response contains one embedding per input; partial or truncated internal encode success is treated as a 500 INTERNAL_ERROR instead of returning a partial 200. Every error path returns the OpenAI {error:{message,type,param,code}} envelope (inner SIE-native encode failures are translated), so an openai-client error handler works unchanged.

post/v1/embeddings

Headers

X-SIE-MACHINE-PROFILEstring nullable

Preferred GPU or machine profile

X-SIE-Poolstring nullable

Explicit pool routing override

X-SIE-SDK-Versionstring nullable

Client SDK version for skew warnings

Request body

dimensionsinteger nullable

Accepted but ignored; the gateway returns the model's native dimension.

encoding_format'float' | 'base64'
modelstring required
userstring nullable

Accepted but ignored; kept for OpenAI SDK compatibility.

Response

OpenAI-compatible embeddings response

modelstring required
objectstring required