tya-math-ling-fmt-t06
rita-cohere/tya-math-ling-hyp-5327 (math-ling hyp ckpt) packaged for IOL-AI 2026 with the decode stack from linguini-test-iolai.ipynb:
- FORMAT IS CRITICAL user prompt + per-task FORMAT/DIRECTION hints
(translation, fill_blanks, match_letters, num↔text) /think+reasoning_options.enabled- think 2048 @ T=0.6; force-close
<|END_THINKING|><|START_RESPONSE|>if needed - answer 512 @ T=0.6
- soft
P(END_THINKING)≥0.55early exit + loop break - marker-gated
FINAL ANSWERS:parser (no think-trace dump); strip**bold** - up to 3 quality resamples on empty/bad format
script.py reads /tmp/data/test.csv and writes submission.csv (id,pred).
- Downloads last month
- 41
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for rita-cohere/tya-math-ling-fmt-t06
Base model
rita-cohere/tya-math-ling-hyp-5327