- Category
- Developer Tools
- Rank
- No. 1120Tools index
- Pricing
- Open Source
- Platform
- cli
- Type
- TOOL
- GitHub
- 1.8k stars
- Latest release
- v0.73.2
- Added
- Aug 16, 2026
About
Soup is a CLI for fine-tuning and post-training LLMs from a single YAML config, using layer streaming to keep the frozen base model out of VRAM so an 8B model can train on a 4GB laptop GPU. It also includes a 'soup ship' regression gate that benchmarks model checkpoints against suites like tool-calling, MMLU, and refusal-rate to catch regressions before deployment.
What it can do
Fine-tune and post-train LLMs from a YAML config
YAML config file → Fine-tuned LLM
Stream model layers to keep frozen base model out of VRAM during training
Base LLM weights → Reduced VRAM usage during training
Benchmark model checkpoints against regression test suites (tool-calling, MMLU, refusal-rate) before deployment
Model checkpoint → Regression benchmark report
Why it made the leaderboard
Fine-tuning normally means renting a GPU box and hand-writing training config; this reduces it to one config file and one command, and the layer-streaming path makes an 8B run feasible on consumer hardware you already own.
Tags
Tech Stack
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.