MakazhanAlpamys/Soup
★ 139 stars
PythonSoup turns the pain of LLM fine-tuning into a simple workflow. One config, one command, done.
Transcript
Fine-tuning an LLM usually means SSH and config hell. Soup streams the frozen 8B base one decoder layer at a time onto a 4 GB card. One YAML config and soup train run QLoRA with auto quantization — 119.6 tokens per second, 3.32 GB peak. Layer streaming is still opt-in beta, but fine-tuning now happens on your own GPU, no cloud.