sakshamio/legal-financial-modernbert-150m
017
Remove smoke-test weights from main; checkpoints live on step-N branches
Remove smoke-test weights from main; checkpoints live on step-N branches
Training in progress, step 2000
Sync technical plan to launched numbers
Update: launched config (271k steps, 475 tok/param), 20% general mix, custom domain tokens, Go dedup, three silent bugs caught
Add full technical plan
Update: bandwidth-bound findings, FP8 verdict, 115GB corpus, 30-day budget, dedup + fertility measurements
Document checkpoint series (step-N branches)
Add model card: full training plan, design decisions, GB10 hardware findings
Add domain BPE tokenizer (50,368 vocab)
Training in progress, step 20
initial commit
