Complete the LLaMA-2-13B base block: TR-3 HellaSwag 10-shot (77.04)

Add TR-3 and ASVD on LLaMA-2-13B base; 13B base block now has all four arms

Add base-model few-shot/generative results: SVD-LLM, SVD-LLM V2, ASVD on 7B and 13B base (24 cells)

Add ASVD LLaMA-2-13B base results (zero-shot, BoolQ, three-corpus perplexity)

Refresh benchmark tables: base-model arms complete (SVD-LLM / V2 / ASVD at 7B, SVD-LLM / V2 at 13B)

Add full benchmark tables (perplexity / zero-shot / few-shot) for all methods and models

Add asvd_llm family: ASVD baselines @ matched budgets (heal 60M + unheal)

Add tt3 family and tr3/heal/qwen2.5-7b under <method>/<stage>/<model> convention

Reorganize repo: <method>/<stage>/<model> (tr3, svd_llm, svd_llm2 x heal/unheal)

Add LLaMA-2-13B SVD-LLM / SVD-LLM V2 baselines (healed 60M and unhealed)

Add healed LLaMA-2-13B TR-3 and healed weights for both TT-3 controls

Point every loading recipe at the renamed checkpoint

Index the two unhealed TT-3 topology controls

Add TR-3 Qwen2.5-7B (31.89% reduction, healed on 60M tokens)

Add matched-budget SVD-LLM / SVD-LLM V2 KD-healed baselines

Restructure into per-model subdirectories; add TR-3 Qwen2.5-3B

Update README.md

Upload README.md with huggingface_hub