Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.
Python cliconsumer-gpudpofine-tuningggufhuggingfacellmllmopslocal-ailocal-llmloralow-vrammachine-learningollamapeftpythonpytorchqlorasfttransformers
services:
soup:
image: ghcr.io/makazhanalpamys/soup:latest
build:
context: .
dockerfile: Dockerfile
args:
# A local build installs this soup-cli release from PyPI, e.g.
# SOUP_VERSION=0.77.0 docker compose build
# Unset, the Dockerfile stops the build. Pulling the published image needs none.
- SOUP_VERSION
volumes:
- .:/workspace
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]
ipc: host # Recommended for PyTorch multi-processing