sync local anonymous/model/ contents

docs: drop verbatim YAML/JSON dumps; point to source files

infer: accept HF Hub repo id for --adapter-dir

docs: complete inference snippet + embed rubric prompt templates

Strip base-model identity from README and training_args.json (kept only in adapter_config.json where PEFT requires it)

Initial anonymous release: Qwen3.5-9B judge LoRA (subq+human, ckpt-294)

initial commit