Phase 2 prep: copy Phase 1 ckpt to repo root with attnres model_type + GPT2TokenizerFast; fixes Phase 2 from_pretrained 404

README.md v2: clarify trust_remote_code=False caveat for external users; replace local-style AttnResLM.from_pretrained snippet with model.model.layers[i].attn_res.proj probing path that works through the HF wrapper

Patch README: replace local-style AttnResLM snippet with HF AutoModel API path + AttnRes-aware LoRA target modules

Phase 1 training complete

initial commit