Add critical technical details: BF16 merge, fresh LoRA init, dataset mixing

Update README: Clarify no LoRA stacking, add dataset strategy

Update paper: Add FAQ, clarify no LoRA stacking, add dataset mixing strategy

Upload PAPER.md with huggingface_hub

Upload plm.py with huggingface_hub

Upload README.md with huggingface_hub

initial commit