119 lines
2.6 KiB
Markdown
119 lines
2.6 KiB
Markdown
# Setup Local — DictIA
|
|
|
|
Guide pour deployer DictIA localement avec GPU NVIDIA ou CPU.
|
|
|
|
## Profil local-gpu
|
|
|
|
### Prerequis
|
|
|
|
- NVIDIA GPU avec support CUDA
|
|
- [nvidia-container-toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html)
|
|
- Docker + Docker Compose V2
|
|
- 8GB+ RAM (16GB recommande)
|
|
- Token HuggingFace (pour la diarisation)
|
|
|
|
### Installation nvidia-container-toolkit
|
|
|
|
```bash
|
|
# Ubuntu/Debian
|
|
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
|
|
sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
|
|
curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
|
|
sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
|
|
sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
|
|
sudo apt-get update
|
|
sudo apt-get install -y nvidia-container-toolkit
|
|
sudo nvidia-ctk runtime configure --runtime=docker
|
|
sudo systemctl restart docker
|
|
|
|
# Verifier
|
|
docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi
|
|
```
|
|
|
|
### Setup DictIA
|
|
|
|
```bash
|
|
cd dictia
|
|
bash deployment/setup.sh --profile local-gpu
|
|
```
|
|
|
|
Le setup va verifier:
|
|
- nvidia-container-toolkit installe
|
|
- GPU accessible depuis Docker
|
|
- Assez de RAM disponible
|
|
|
|
### Configuration du modele
|
|
|
|
Par defaut, WhisperX utilise `large-v3`. Pour changer:
|
|
|
|
```bash
|
|
# Editer .env
|
|
ASR_MODEL=large-v3 # Meilleure qualite
|
|
# ASR_MODEL=medium # Plus rapide, qualite correcte
|
|
# ASR_MODEL=small # Tres rapide, qualite reduite
|
|
```
|
|
|
|
---
|
|
|
|
## Profil local-cpu
|
|
|
|
### Prerequis
|
|
|
|
- Docker + Docker Compose V2
|
|
- 18GB+ RAM (WhisperX CPU est gourmand)
|
|
- Patience (transcription ~10x temps reel)
|
|
|
|
### Setup
|
|
|
|
```bash
|
|
cd dictia
|
|
bash deployment/setup.sh --profile local-cpu
|
|
```
|
|
|
|
### Limitations
|
|
|
|
- Transcription lente: 1h d'audio prend ~10h
|
|
- Utilise float32 (pas de GPU acceleration)
|
|
- Limite memoire a 18GB par defaut
|
|
- Recommande pour: tests, petits fichiers, demos
|
|
|
|
Pour reduire l'utilisation memoire, utiliser un modele plus petit:
|
|
|
|
```bash
|
|
# Editer .env
|
|
ASR_MODEL=small # ou medium, base, tiny
|
|
```
|
|
|
|
---
|
|
|
|
## Verification
|
|
|
|
```bash
|
|
# Health check
|
|
bash deployment/tools/health-check.sh
|
|
|
|
# Test rapide: ouvrir le navigateur
|
|
open http://localhost:8899
|
|
|
|
# Verifier WhisperX
|
|
curl http://localhost:9000/health
|
|
```
|
|
|
|
## Gestion des containers
|
|
|
|
```bash
|
|
COMPOSE_FILE=deployment/docker/docker-compose.local-gpu.yml # ou local-cpu
|
|
|
|
# Logs
|
|
docker compose -f $COMPOSE_FILE logs -f
|
|
|
|
# Redemarrer
|
|
docker compose -f $COMPOSE_FILE restart
|
|
|
|
# Arreter
|
|
docker compose -f $COMPOSE_FILE down
|
|
|
|
# Voir l'utilisation GPU
|
|
nvidia-smi # (profil GPU seulement)
|
|
```
|