train: higher LR=1e-5, more SFT=20, lower temp=1.0->0.7

fix: C-4 reward clamp, C-6 HF_TOKEN, W-2 citation floor, W-9 dup import

fix: force torch 2.5.1+cu121 after unsloth to prevent colab-new downgrade

fix: upgrade PyTorch to 2.5.1 to fix unsloth_zoo AttributeError on torch._inductor.config

Upload hf_training/train.py with huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload hf_training