Soup
Soup is a command line tool that handles the fine-tuning (fine-tune) and post-training processes of large language models over a single YAML file. With layer flow (layer streaming), it allows to train a model with 8 billion parameters on laptop graphics cards with a capacity of 4 GB VRAM. It eliminates infrastructure complexity by offering command line and Web interface support.
Soup
80MakazhanAlpamys · Local AI / Self-Hosted LLM
Common commercial products in this category: ChatGPT, Claude, Gemini
- You can fine-tune large language models with a single YAML configuration file and single command.
- By enabling the layer flow feature, you can run a laptop GPU 8B model on the 4 GB VRAM border.
- Using QLoRA on local hardware, you can perform model training without the need for cloud servers.
- ↑ PositiveHacker News
Beautiful Soup
“Beautiful Soup is an incredibly robust, and powerful tool. However, it can be sometimes be an intimidating tool for beginners (which ".find_x_y_z" method should I use again?). To that end you should check out gazpacho (with just a single "find" method):”
- ↓ NegativeGitHub Issue
MLXSFTTrainerWrapper: add end-to-end training-loop integration test on Apple Silicon
“## Problem v0.25.0 shipped the MLX backend (`backend: mlx`) with a full SFT trainer wrapper (`MLXSFTTrainerWrapper`) that delegates to `mlx_lm.tuner.trainer.train()`. However, the **training loop itself is not integration-tested** against a real MLX runtime. Current test…”