Datasets:
Tasks:
Text Classification
Formats:
parquet
Languages:
English
Size:
1K - 10K
Tags:
structured-decisions
calibration
probabilistic-classification
system-one
workflow-evaluation
Synthetic
License:
Commit History
Yesterday
-
bench: save per-decision probabilities and add them to the eval-results PR
Verified signaturescore_with_jobs.py already scores every decision but keeps only the summary. This PR: - writes `predictions.jsonl` next to `results.json`: one line per decision with case id, workflow, question id, type and the normalized probabilities over the gold options (the vector `score_one` scores, after the yes/no to true/false mapping); - has `submit_results.py` add that file to the model-repo PR as `eval/typed-decisions/predictions.jsonl` when it is present next to the results file (`--predictions` points elsewhere). Scoring is unchanged. Replaying a published Jev run (OpenDecider's whole-case file, which their benchmarks README lists as Jev 1.13 through TypeSafe's API) through both versions with vLLM, HTTP and `datasets` mocked gives a byte-identical `results.json` (accuracy 0.7305), and re-scoring `predictions.jsonl` gives the same accuracy, KL, Brier and ECE. The tie rule (first-listed option vs split credit) is left for a separate change; with the probabilities saved it can be re-appli…0f4bce1 -
Move nanoGentzen to v3 (#59) in the fitted table (#66)
- Move nanoGentzen to v3 (#59) in the fitted table (cb693435d4d2c34e5b1d288caf682d92f9691873) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>c9e8c41
2 days ago
-
Add ten Rev-gen1 models (#64) to the fitted table (#65)
- Add ten Rev-gen1 models (#64) to the fitted table (a2a246788bc67951bc594664643057e7bba986e6) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>fa01e86 -
Add signpost-d-4b (#61), Qwen3.8-27B base (#60) and FastBrain-9B fitted (#57) (#63)
- Add signpost-d-4b (#61), Qwen3.8-27B base (#60) and FastBrain-9B fitted (#57) (42617aa87edde7b463bfba4da3be7b01c134b5bf) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>992ce44
4 days ago
-
Add FastBrain-9B (#57) to the zero-shot table (#58)
- Add FastBrain-9B (#57) to the zero-shot table (5de28598f08fd2d7d8b525288fe91796ea4cf89d) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>94993b8 -
Update Decider-4B (#36) and add nullspan v1.3.0 (#55) to the fitted table (#56)
- Update Decider-4B (#36) and add nullspan v1.3.0 (#55) to the fitted table (597f3ebfa55872c2463d299f634040750d49afb1) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>1c9479d
5 days ago
-
Add d1-2B and d1-2B-OptiQ-4bit (zero-shot) (#54)
- Add d1-2B and d1-2B-OptiQ-4bit (zero-shot) (4c7d40c65f219e10fe1dc43e6f35044c4079c27d) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>d501540
6 days ago
8 days ago
-
Leaderboard: add serin-0.8b to the fitted table (trained on train) (#51)
- Leaderboard: add serin-0.8b to the fitted table (trained on train) (17ec45292d2fe0e14d182f422e24ec0b8ba319cf) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>7f479ec -
Leaderboard: add SafeDecide-400M (#39), Seb-9B (#13) and SAP RPT 1.6 (#49) to the fitted table (#50)
- Leaderboard: add SafeDecide-400M (#39), Seb-9B (#13) and SAP RPT 1.6 (#49) to the fitted table (7811b037a0c3e4e2d0666e9e078130824ad9c261) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>9a83c37
11 days ago
-
Leaderboard: add lazydecider-0.8b (#47) to the fitted table (#48)
- Leaderboard: add lazydecider-0.8b (#47) to the fitted table (278073c2be7d7d28a5b79b7aa8a5c661d9078d74) Co-authored-by: Sandeep Das <sandeep85@users.noreply.huggingface.co>9ee6942 -
Add pramana-2b and pramana-0.8b to the fitted table (#46)
Verified signature80cb93a
13 days ago
-
sysone-typed-decisions-0.4B (#44): add KL, ECE and latency
Verified signature4772cfe -
Add sysone-typed-decisions-0.4B (#44) and tasksource-decider-nano (#45) to the fine-tuned table
Verified signaturefbae038 -
Add Baseten Mammoth 1 and Elephant 1 (scored by us via OpenRouter); regenerate chart
Verified signature6c7cf4b -
Card: condense row notes to who scored each row
Verified signaturedbeb1c3 -
Chart: reverse KL and latency axes so top right is best
Verified signature9c67df6 -
Card: remove footnotes and row markers
Verified signature3e63d4f -
Chart: Decider 1 drawn like the other hosted APIs
Verified signaturee7ddf84 -
Card: drop the request-shape caveat and the ♠ note
Verified signature528ef9f -
Add Decision Models rows (#38): quyet-1-large, s1-pro, s1-fast; regenerate chart
Verified signature6644cea
16 days ago
-
Table 2: add rank column
Verified signatured9eda7f -
Leaderboard: add 18 zero-shot and 2 fitted rows scored by SaylorTwift (#37), linne-oak 1.0 (#35), Arbiter-4B (#34), Decider-4B (#36), SafeDecide-140M (#39)
Verified signature93eda01 -
Add HF Jobs how-to for benchmarking a model (#42)
- Add HF Jobs how-to for benchmarking a model (4cca1526864ab7819b5e298f1e5111249d7511c6) - Add bench/ scripts referenced by the README section (1ce7b5db35b4d7f6d5cb6c38d49cd775e301132c) Co-authored-by: Nathan Habib <SaylorTwift@users.noreply.huggingface.co>e49e13d -
Add HF Jobs how-to for benchmarking a model (#40)
- Add HF Jobs how-to for benchmarking a model (ffabd9610489d71ca05bef053191a42e031690ff) - Add bench/score_with_jobs.{sh,py} referenced by the README section (b1cf7e94404c41ccd5118ebb1337777789cee6af) Co-authored-by: Nathan Habib <SaylorTwift@users.noreply.huggingface.co>e742f0a
21 days ago
22 days ago
-
Leaderboard chart: add Amos 1 medium and large
Verified signaturef8f6651 -
Add rony/ministral3-3b-reasoning-typed-decisions (fine-tuned on train, 0.808) from #30
Verified signature4cad9f4 -
Add Amos 1 medium and large (general, zero-shot) and the large's fitted row (#32)
- Add Amos 1 medium and large (general, zero-shot) and the large's fitted row (1a9732640b9c9a40fea6a9804093580b0589ef9a) Co-authored-by: Zaid Mukaddam <mukaddamzaid@users.noreply.huggingface.co>b0084ae -
Leaderboard chart: add jev-vision-27b (zero-shot #5); generated by mini-jev/typed_decisions/make_leaderboard_chart.py
Verified signaturec806f6c -
Zero-shot table: add jev-vision-27b at #5 (0.714 acc, self-reported, #28)
Verified signaturee4502d0 -
Table 2: add Qwen3.5-4B LoRA (#22), gemma3-270m specialist and general (#26, #27), ProKope-421M-8K (#23/#24)
Verified signature2e58551 -
Table 2: list Phocinae-Largha-150M-v1 at its clean re-score (0.791), drop the test-exposed 0.906 (discussion #25)
Verified signature1d513eb
25 days ago
-
Add Phocinae-Largha-150M-v1 (specialist, fitted on train) to Table 2 (#21)
- Add Phocinae-Largha-150M-v1 (specialist, fitted on train) to Table 2 (18d0e1477d4d3ba38af86cea3a4382b73ff47ec6) Co-authored-by: Perry Link <perrylink@users.noreply.huggingface.co>6a7927f
27 days ago
-
Add Bekko System One v0 (17M, 68M, 400M) to the fine-tuned table, scored by us
Verified signatured0e2f0c -
Leaderboard: zero-shot models only; add self-reported entries; second table for models fitted on train
Verified signatureda062fc
-
Add eval.yaml to register as a Hub benchmark (#9)
- Add eval.yaml to register as a Hub benchmark (dc021733b120def5221fdce28d01049bb0216947) Co-authored-by: Nathan Habib <SaylorTwift@users.noreply.huggingface.co>561333a -
Restore the teacher description (4B-class); keep the request-shape note
Verified signatureReverts the wording change in 1111d14: the teacher is a roughly 4B-class model, as the card said before. The request-shape paragraph stays. No scores changed.e135720 -
Correct the teacher description; state the request shape
Verified signatureThe card called the gold teacher '4B-class', but the labeller only records the method (a routing proxy, 3 samples, temperature 0.7), not which model answered. Say that, and say closed models cannot be ruled out. Also state that every row sends the whole case in one request, and cite the third-party finding that request shape moves Jev's yes/no accuracy. No scores changed.1111d14
-
Add Liquid AI d1 (0.742)
Verified signatureScored 2026-09-30 through api.liquid.ai (d1:free) with the same client code as the Jev row: 400 cases, 2,000 decisions, zero errors. Accuracy 0.742, KL 0.475, Brier 0.155, ECE 0.124, p50 525 ms.d51d993 -
Streamline the card: one leaderboard table, a chart, shorter notes
Verified signatureMerges the Leaderboard and Baseline results tables into one, adds a two-panel chart (accuracy vs KL, accuracy vs latency with price), and trims the prose from about 3,000 to 1,400 words. No numbers changed.f2491dd -
Tag meraGPT links with UTM parameters so card-driven visits can be attributed
Verified signaturebd9eee1 -
Add Jeff (0.8B, 2B, Gemma4-E2B) to the leaderboard, scored 2026-09-29 through the Jev client path
Verified signature2386726
-
Leaderboard: add Featherless Simple Jev (Qwen3.6-35B-A3B classifier), 0.716 acc, measured on the 400-case test split
Verified signaturef7a2487
-
Latency footnote: say exactly how it was measured
Verified signaturec76749e -
Leaderboard: Decider 1 beats Jev on accuracy, latency and price
Verified signature811ea6c -
Leaderboard: meraGPT Decider 1 is state of the art (0.768 accuracy, KL 0.096)
Verified signaturee9e70d4