Spaces:
Running
Running
Commit History
-
Make tombench-en public — MIT license permits redistribution Running
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>6fbcefc -
Keep tombench-en private (third-party benchmark data)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>0f5e2a3 -
Swap GSM8K for ARC-Challenge (paper's 2x2 design); all artifacts now public
ARC-Challenge bin scores aggregated from per-doc TrackStar scores and published to dolma3-influence-heatmaps; reproduces Figure 2C exactly. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>cff3bb4 -
Use real WebOrganizer taxonomy and live influence data from the Hub
Replace synthetic np.random heatmaps with bin-level TrackStar scores loaded from HCAI-Lab/dolma3-influence-heatmaps, z-scored per benchmark (paper section 3.3), RdBu color capped at |z|=2.5 as in Figures 2/12. Reproduces the paper's Literature x Customer Support cell exactly (+16.0 / -7.31 / -5.75). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>2285938 -
Add Data Artifacts tab mapping Hub repos to main-paper figures
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>a330464
-
fix personal site
280e0d9
-
Deploy Capabilibara Space under HCAI-Lab
Verified signature5673c43 -
Deploy Capabilibara Space under HCAI-Lab
Verified signature1fa3133 -
initial commit
Verified signatured45e6a0
