Adding GPU support for transcription

This commit is contained in:
2025-10-05 12:01:22 +02:00
parent 51624bd066
commit a450ab5c70
5 changed files with 82 additions and 14 deletions

View File

@@ -35,6 +35,26 @@ OPENAI_API_KEY=
# OPENAI_TRANSCRIBE_MODEL=whisper-1
# OPENAI_TRANSCRIBE_TIMEOUT=600
# Local Whisper settings
# Choose CPU explicitly unless you have a working GPU runtime in Docker
WHISPER_DEVICE=cpu
# Model and precision (large-v3 int8 is accurate but heavy; consider medium/small for speed)
WHISPER_MODEL=large-v3
WHISPER_PRECISION=int8
# Threads for CPU inference
WHISPER_CPU_THREADS=4
# --- GPU (CUDA) optional setup ---
# To enable NVIDIA GPU acceleration:
# 1) Install NVIDIA driver on the host and the NVIDIA Container Toolkit
# 2) Set the Docker runtime to NVIDIA for the worker containers
# DOCKER_GPU_RUNTIME=nvidia
# 3) Ensure GPU visibility (default is all)
# NVIDIA_VISIBLE_DEVICES=all
# 4) Use GPU-friendly precision and device
# WHISPER_DEVICE=cuda
# WHISPER_PRECISION=float16
# Docker volumes paths
LIBRARY_HOST_DIR=/mnt/nfs/library
TRANSCRIPTS_HOST_DIR=/mnt/nfs/transcripts