y-agent/modular-addition-feature-learning

Add precomputed results for p=45

Add precomputed results for p=35

Add precomputed results for p=19

Add precomputed results for p=3

Add precomputed results for p=67

Add arXiv link (2602.16849)

Add precomputed results for p=5

Add precomputed results for p=7

Add precomputed results for p=37

Increase on-demand max modulo from 97 to 99

Add precomputed results for p=33

Set default modulo to largest available and stream child logs immediately

Add precomputed results for p=27

Improve HF result commit auth and increase training log heartbeat

Add precomputed results for p=25

Add precomputed results for p=21

Add precomputed results for p=9

Add blog link

Revert Tab 6 figures to full-width layout

Clean README for GitHub, use deploy script for HF metadata

Fix caption visibility and Generate button layout

Add training progress logs, suppress expected grokking warnings

Improve captions, expand Tab 4, horizontal layouts for Tabs 3/6

Split DFT heatmap into side-by-side W_E/W_L, simplify Tab 1

Update Tab 7 descriptions for consistent timepoints

Use same DFT timepoints [0, 1000, 5000] for both Quad and ReLU

Fix Tab 7: 10K epochs, better neuron selection, fixed DFT timepoints

Fix Tab 7: 10K epochs, better neuron selection, fixed timepoints

Add app.load() to populate tabs on initial page load

Add startup diagnostics for debugging

Disable SSR mode to fix rendering

Add HF Space config frontmatter to README

Deploy app with precomputed results for p=15,23,29,31

initial commit