selfhostyourtech/reflector
2
0
Fork 0
mirror of https://github.com/Monadical-SAS/reflector.git synced 2026-04-13 08:46:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: 3-mode selfhosted refactoring (--gpu, --cpu, --hosted) + audio token auth fallback (#896)

Browse Source 
* fix: local processing instead of http server for cpu

* add fallback token if service worker doesnt work

* chore: rename processors to keep processor pattern up to date and allow other processors to be createed and used with env vars
This commit is contained in:
Juan Diego García
2026-03-04 16:31:08 -05:00
committed by GitHub
parent 4235ab4293
commit a682846645
34 changed files with 2640 additions and 172 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
server/.env.selfhosted.example
View File

@@ -32,26 +32,46 @@ AUTH_BACKEND=none
# =======================================================
# Specialized Models (Transcription, Diarization, Translation)
# These run in the gpu/cpu container — NOT an LLM.
# The "modal" backend means "HTTP API client" — it talks to
# the self-hosted container, not Modal.com cloud.
# These do NOT use an LLM. Configured per mode by the setup script:
#
# --gpu mode: modal backends → GPU container (http://transcription:8000)
# --cpu mode: whisper/pyannote/marian/pyav → in-process ML on server/worker
# --hosted mode: modal backends → user-provided remote GPU service URL
# =======================================================
# --- --gpu mode (default) ---
TRANSCRIPT_BACKEND=modal
TRANSCRIPT_URL=http://transcription:8000
TRANSCRIPT_MODAL_API_KEY=selfhosted
DIARIZATION_ENABLED=true
DIARIZATION_BACKEND=modal
DIARIZATION_URL=http://transcription:8000
TRANSLATION_BACKEND=modal
TRANSLATE_URL=http://transcription:8000
PADDING_BACKEND=modal
PADDING_URL=http://transcription:8000
# HuggingFace token — optional, for gated models (e.g. pyannote).
# Falls back to public S3 model bundle if not set.
# --- --cpu mode (set by setup script) ---
# TRANSCRIPT_BACKEND=whisper
# DIARIZATION_BACKEND=pyannote
# TRANSLATION_BACKEND=marian
# PADDING_BACKEND=pyav
# --- --hosted mode (set by setup script) ---
# TRANSCRIPT_BACKEND=modal
# TRANSCRIPT_URL=https://your-gpu-service.example.com
# DIARIZATION_BACKEND=modal
# DIARIZATION_URL=https://your-gpu-service.example.com
# ... (all URLs point to one remote service)
# Whisper model sizes for local transcription (--cpu mode)
# Options: "tiny", "base", "small", "medium", "large-v2"
# WHISPER_CHUNK_MODEL=tiny
# WHISPER_FILE_MODEL=tiny
# HuggingFace token — for gated models (e.g. pyannote diarization).
# Required for --gpu and --cpu modes; falls back to public S3 bundle if not set.
# Not needed for --hosted mode (remote service handles its own auth).
# HF_TOKEN=hf_xxxxx
# =======================================================