Datasets:

LocalLLaMA
/

typed-decisions

#68 opened about 10 hours ago by

#67 opened about 13 hours ago by

#64 opened 2 days ago by

#62 opened 2 days ago by

#61 opened 2 days ago by

#60 opened 3 days ago by

#59 opened 3 days ago by

#57 opened 4 days ago by

#55 opened 4 days ago by

#52 opened 6 days ago by

#47 opened 11 days ago by

#46 opened 11 days ago by

#45 opened 12 days ago by

#44 opened 12 days ago by

#39 opened 17 days ago by

#38 opened 19 days ago by

#37 opened 19 days ago by

#36 opened 19 days ago by

#35 opened 20 days ago by

#34 opened 20 days ago by

#33 opened 21 days ago by

#28 opened 22 days ago by

#27 opened 22 days ago by

#25 opened 23 days ago by

#23 opened 23 days ago by

#22 opened 23 days ago by

#20 opened 25 days ago by

#19 opened 26 days ago by

#18 opened 27 days ago by

#17 opened 27 days ago by

#15 opened 27 days ago by

#14 opened 27 days ago by

#13 opened 28 days ago by

#12 opened about 1 month ago by

#11 opened about 1 month ago by

#10 opened about 1 month ago by

#8 opened about 1 month ago by

#7 opened about 1 month ago by

#6 opened about 1 month ago by

#5 opened about 1 month ago by

#4 opened about 1 month ago by

#3 opened about 1 month ago by

#2 opened about 2 months ago by

#1 opened about 2 months ago by

bench: save per-decision probabilities and add them to the eval-results PR

K-Decision-9B (ONYULTEA) — 0.818 accuracy (fitted)

Submission: ten Rev-gen1 models (general), 150M–25.6B

Interactive explorer for the leaderboard: ranking + Pareto charts, reads this card live

Submission: skamalj/signpost-d-4b (4.5B, general, zero-shot): 0.712 acc / 0.1615 KL / 0.0853 Brier / 0.0914 ECE

Submission: Qwen/Qwen3.8-27B base model (general, zero-shot): 0.661 acc / 0.590 KL / 0.214 Brier, plus FP8, NVFP4 and Swift 1.5

[Leaderboard Submission] nanoGentzen-v3 (4.93M) - 76.75% Accuracy (General) - New Version

Submission: TypeScript/FastBrain-9B (9.4B): 0.7625 acc / 0.1508 KL / 0.0795 Brier (zero-shot) & 0.8070 acc / 0.0816 KL (fitted)

Submission: gnumanth/nullspan v1.3.0 (395M 8K tree-isolated encoder, specialist): 0.7900 acc / 0.0760 KL / 0.0420 Brier / 0.1665 ECE

Add Sagicc/nanoGentzen-Jev-v3 (4.93M params) — 0.731 acc, 0.150 KL, 0.081 Brier, 0.087 ECE

Submission: rahulkumar0612/lazydecider-0.8b (0.8B, specialist): accuracy 0.770 / KL 0.094 / Brier 0.051 / ECE 0.145

Submission: apracv/pramana-2b (2B) and apracv/pramana-0.8b (0.8B), fine-tuned on train: 0.8005 / 0.7970 acc, KL 0.0755 / 0.0819

Submission: tasksource/tasksource-decider-nano (149M, general, train is 0.5% of its mix): 0.7515 acc / KL 0.113 / Brier 0.062

Submission: forward-metric/sysone-typed-decisions-0.4B (specialist): accuracy 0.8025

Submission: yotambr/safedecide-140m (general, then fine-tuned on train): accuracy 0.754

Hosted provider request: Decision Models (decisionmodels.io)

Benchmark results for 17 trending decision models (zero-shot + 2 fitted, HF Jobs run 2026-10-16)

Leaderboard submission: roudone/Decider-4B (fine-tuned on train)

Submission: linne-oak 1.0 (31B, general, zero-shot)

Leaderboard submission: Goddragon/Arbiter-4B

Submission: FINAL-Bench/Darwin-27B-ZTC-v5 (general, then fine-tuned on train): accuracy 0.812

Add jev-vision-27b (general, zero-shot): acc 0.714, KL 0.318

Submission: rony/gemma3-270m-typed-decisions (Specialist)

Leaderboard question: the current #1 was trained on test-split rows (Phocinae-Largha-150M-v1)

Submission: ProKope-421M-8K (Alan Canto) — 0.7185 acc / 0.385 Brier / 8K context (specialist)

Qwen3.5-4B LoRA fine-tune: results submitted, model missing from the Hub leaderboard

Add ProKope-421M (Alan Canto) to Table 2 (fine-tuned models)

Nirnaya fine-tuned specialist evaluation (observability results)

Submission: Phocinae/Phocinae-Largha-150M-v1 — specialist (fitted on the train split); accuracy 0.906 (en) / 0.848 (zh)

Submission: FINAL-Bench/Darwin-27B-ZTC (general, zero-shot), accuracy 0.743

Phocinae-Largha-150M-v1 — test scores (specialist: fitted on train split) + reproduction

Aplomb 1 (general, trained on train): 0.737 accuracy, KL 0.123, Brier 0.065, ECE 0.137

Seb-9B (9B, fitted on train, one question per request): 0.754 acc / KL 0.349 / Brier 0.131 / ECE 0.031 on the official test split

MAV (specialist, 421M, Laya-RLCD reproduction on one RTX 4090): 0.7800 acc / KL 0.086 / Brier 0.050 on the official test split

ProKope-421M (Alan Canto): 0.714 acc / 0.071 Brier / 0.254 Score MAE on official test split

Call for results: add your decision model to the leaderboard

Leaderboard submission: OpenDecider (open weights, Apache-2.0), one zero-shot model and four fine-tuned on `train`

Bongard-mini: general, zero-shot, open weights (accuracy 0.594, KL 0.256, Brier 0.132)

Japanese translation: GeneLab/typed-decisions-ja

prima-ratio + 12B (generalist, zero-shot, default calibration): 0.702 acc / KL 0.564 / Brier 0.234 / ECE 0.146

od1-typed-decisions (specialist, 4B): 0.7965 acc / KL 0.082 / Brier 0.045 on the official test split

soft-decider-421m (specialist): 0.774 acc / 0.141 ECE on the official test split

Laya benchmark results

[bot] Conversion to Parquet