Update to LoRA Phase 2 merged weights (PPL 15.78)

Update model card: LoRA Phase 2 (PPL 15.78, 97.3% instruction following)

Add detailed base model info, pre-training datasets, and research context

Clarify: 20M is SFT tokens, base model pre-trained on 9.8B tokens

Upload README.md with huggingface_hub

Upload model.pt with huggingface_hub

Upload tokenizer.model with huggingface_hub

Upload README.md with huggingface_hub

initial commit