Description
Fine-tuning an LLM means SSH into servers, endless config and out-of-memory errors. Soup turns fine-tuning and post-training into one YAML and one command, with layer streaming that trains an 8B model on a 4GB laptop GPU.
It supports SFT, DPO, LoRA and QLoRA, exports GGUF for Ollama, and has a web UI.
One config:YAML for everything.
Low VRAM:8B on 4GB via layer streaming.
Methods:SFT, DPO, LoRA and QLoRA.
GGUF export:Use it in Ollama.
It supports SFT, DPO, LoRA and QLoRA, exports GGUF for Ollama, and has a web UI.
Features
One config:YAML for everything.
Low VRAM:8B on 4GB via layer streaming.
Methods:SFT, DPO, LoRA and QLoRA.
GGUF export:Use it in Ollama.
