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
"""Soup Cans — shareable .can artifact format (v0.26.0 Part E).
A ``.can`` file is a tar.gz containing:
manifest.yaml — format version, name, author, etc.
config.yaml — full SoupConfig
data_ref.yaml — how to fetch training data (hash + URL/HF id)
recipe.md — human-readable description (optional)
"""
from __future__ import annotations
from soup_cli.cans.schema import CAN_FORMAT_VERSION, DataRef, Manifest
__all__ = ["CAN_FORMAT_VERSION", "DataRef", "Manifest"]