cleanup: remove botched-filename trash from repo root

remove generic docs; keep run journal

remove docs/A100_KICKOFF_PROMPT.md

Add AdamW vs Muon optimizer ablation results (300M tokens/variant)

Fix RoPE dtype cast for bfloat16 inference

Add lm-eval results: HellaSwag 26.6%, ARC-easy 36.3%, PIQA 56.4%

Upload scripts including export_hf.py and ablate.py fixes

Upload artifacts/train.log with huggingface_hub

Upload artifacts/metrics.jsonl with huggingface_hub

Upload artifacts/ckpt_5639.pt with huggingface_hub

Add trained checkpoint: 3B tokens, loss=3.16, MFU=31.5%

add A100 kickoff prompts

second review fixes

runbook: A100 40GB rec + SDPA flash note

Muon optimizer + README

runbook: ablation section

add ablation harness

GPU-session fixes (RNG cpu, shard filter, cu124, 3090 config)

update runbook delivery section

Matilda-Mini phases 1-5 + runbook

initial commit