MakazhanAlpamys/Soup v0.4.3

v0.4.3 — Unsloth Backend (2-5x Faster Training)

on GitHub

What's New

Unsloth Backend — Train 2-5x faster with up to 80% less VRAM.

Features

• backend: unsloth config option — one line to enable
• Works with all training tasks: SFT, DPO, and GRPO
• Auto-detection: Soup suggests enabling unsloth when it's installed
• New extra: pip install 'soup-cli[fast]'
• soup doctor now checks for unsloth
• backend available as sweep shortcut

Usage

pip install 'soup-cli[fast]'

base: meta-llama/Llama-3.1-8B-Instruct
task: sft
backend: unsloth  # 2-5x faster, -80% VRAM

data:
  train: ./data/train.jsonl
  format: alpaca

training:
  quantization: 4bit
  lora:
    r: 64
    alpha: 16

Stats

• 33 new tests (404 total)
• ruff clean, CI green

Install / Upgrade

pip install --upgrade soup-cli
# With unsloth:
pip install --upgrade 'soup-cli[fast]'

Full Changelog: v0.4.2...v0.4.3