You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Register previously missing OpenAI-compatible routes so they are
proxied to backends instead of being rejected with 404:
- /v1/audio/transcriptions and /v1/audio/translations (multipart/form-data)
- /v1/audio/speech (JSON)
- /v1/moderations
- /tokenize and /detokenize (vLLM extension)
Add BackendModeAudio and a dedicated handleAudioInference handler that
parses the model field from multipart form data rather than a JSON body.
vLLM (Linux and Metal) passes audio requests through natively; llama.cpp
returns a descriptive error directing users to use chat completions with
an input_audio content part instead.
Signed-off-by: Eric Curtin <eric.curtin@docker.com>
0 commit comments