PEFT
Safetensors
RL-sft-adapter / tokenizer.json

Commit History

  1. SFT v2: Qwen2.5-3B QLoRA, eval_loss=0.983, acc=73.9%

    Verified signature
    8fabe65

  1. v2: sft quality improvements - loss masking, neftune, system prompt, oversampling

    Verified signature
    75af566

  1. Upload folder using huggingface_hub

    Verified signature
    b0e8d85