Commit History
12 days ago
-
tokenizer ablation: digit-split pre-tok = 0.2% -> 99.7% on TinyStories-1M; smallest arithmetic LLM now 3.75M
Verified signaturef117e6b -
final verdict: smallest arithmetic LLM = pythia-14m 14.1M (99.8%); smallest from-scratch = 805,632 (100%); corrected param counts; full stage A/B records
Verified signature9feb5f5
13 days ago
-
add YAML frontmatter to model card
Verified signature9c8cabb -
1M-param 4-digit addition model + full results (phase diagram, extrapolation, bake-off)
Verified signature4a6c57b