modelscope/ms-swift — Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 600+ LLMs (Qwen3.6, DeepSeek-R1, GLM-5.1, InternLM3,…

models​cope/ms‑swift is an open‑source toolkit that streamlines parameter‑efficient fine‑tuning (PEFT) and full‑parameter training for a wide range of large lan