Yesterday
-
typo: close italic
-
README: v2 family rows + dense-supervision finding in headline
-
Upload REPORT.md with huggingface_hub
-
prose catch-up: what-remains tested/untested, reproduce stages 3-5, dense dynamics
-
Upload REPORT.md with huggingface_hub
-
prose refresh: TL;DR, 9-family tables, marginal-returns conclusions, v2 findings
2 days ago
-
Upload REPORT.md with huggingface_hub
-
Upload results/results_table.csv with huggingface_hub
-
Upload plots/training_curves.png with huggingface_hub
-
Upload plots/perk_zoom.png with huggingface_hub
-
Upload plots/acc_vs_k.png with huggingface_hub
-
Upload plots/budget_map.png with huggingface_hub
-
Upload results/rec1m4dense-seed2_eval.json with huggingface_hub
-
Upload results/rec1m4dense-seed2_history.json with huggingface_hub
-
Upload ckpts/rec1m4dense-seed2.pt with huggingface_hub
-
Upload results/rec1m4dense-seed1_eval.json with huggingface_hub
-
Upload results/rec1m4dense-seed1_history.json with huggingface_hub
-
Upload ckpts/rec1m4dense-seed1.pt with huggingface_hub
-
Upload results/rec1m4dense-seed0_eval.json with huggingface_hub
-
Upload results/rec1m4dense-seed0_history.json with huggingface_hub
-
Upload ckpts/rec1m4dense-seed0.pt with huggingface_hub
-
Upload results/rec1m4-seed2_eval.json with huggingface_hub
-
Upload results/rec1m4-seed2_history.json with huggingface_hub
-
Upload ckpts/rec1m4-seed2.pt with huggingface_hub
-
stage5 resume support
-
Upload results/rec1m4-seed1_eval.json with huggingface_hub
-
Upload results/rec1m4-seed1_history.json with huggingface_hub
-
Upload ckpts/rec1m4-seed1.pt with huggingface_hub
-
Upload results/rec1m4-seed0_eval.json with huggingface_hub
-
Upload results/rec1m4-seed0_history.json with huggingface_hub
-
Upload ckpts/rec1m4-seed0.pt with huggingface_hub
-
§4.8 v2 variants: registered predictions, computed verdicts, families registered everywhere
-
margin + stem accounting
-
stem blocks, per-unroll logits, step_emb 80
-
aligned4 mode, dense hold-target supervision, stem
-
stage5: margin-only and margin+stem+dense variants
-
margin-aware R=k+4 readout, stem loading
-
Upload REPORT.md with huggingface_hub
-
correct 100k verdict: climb-then-plateau + credit-assignment explanation
-
Upload REPORT.md with huggingface_hub
-
fix §4.4 interpretation to match data; aligned in §6
-
Upload REPORT.md with huggingface_hub
-
Upload results/results_table.csv with huggingface_hub
-
Upload plots/training_curves.png with huggingface_hub
-
Upload plots/acc_vs_iters.png with huggingface_hub
-
Upload plots/perk_zoom.png with huggingface_hub
-
Upload plots/acc_vs_k.png with huggingface_hub
-
Upload plots/budget_map.png with huggingface_hub
-
SHORT dict: add rec1aligned
-
budget map: aligned offsets + marker