Jev directory
CommunityJev Models & Reimplementations
Decider 2B
Mapika's open decision model is a 1.9B Qwen3.5-2B fine-tune that returns a probability for every option of every typed question in one forward pass, with no generated text. v11 adds a rank-64 LoRA stage and reports 0.577 on the JevBench public hard tier, up from v10's 0.459.
- Reading time
- 3 min
- Published by
- Community
- Author
- Mapika
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Capabilities
- Open source
Highlights
- Reads a state plus typed questions with explicit option lists and returns a distribution per question in one forward pass, with no decoding or parsing.
- v11 is v10 plus a rank-64 LoRA trained for 2 epochs on 42,749 rows, merged into the weights, with one temperature per answer type.
- On the same rows v11 beats v10 on hard decisions: JevBench public hard tier 0.577 vs 0.459 and held-out document questions 0.753 vs 0.646.
- The card also lists losses: human-labelled public sets -2.2 points, knowledge guard -1.6, greedy bag-draw play -10.9, sampled browser play -2.8.
- Apache-2.0, 1.88B BF16 parameters, English; the card calls it an open reproduction of the TypeSafe Jev System One model class.
Watch out
Copy reviewed against the model card fetched on 2026-10-15; the fetch was truncated at 12,000 characters dominated by the embedded chat template, so the usage, speed and reproduction sections were cut off. All benchmark numbers are the authors' own, and the card itself says v11 is worse than v10 on some everyday and game rows.
Questions about this entry
Short answers drawn from the linked project or page.
An open decision model built on Qwen3.5-2B-Base (1.9B parameters) that does not generate text: it reads a state and one or more typed questions with explicit option lists and returns a probability distribution per question in one forward pass.
The card calls it an open reproduction of the System One model class (TypeSafe AI's Jev). The weights live in the Mapika/decider project, which has its own separate entry in this catalog; this entry is the Hugging Face model page.
v11 is v10 plus one supervised LoRA stage of rank 64, trained for two epochs on 42,749 rows of harder decisions and replay with replay rows trained toward v10's own answers, merged into the weights, plus one temperature per answer type. v10 and v8 remain available under their Hub tags.
On the same rows: JevBench public hard tier 0.577 versus 0.459, and held-out document questions 0.753 versus 0.646. Losses are listed too: human-labelled public sets -2.2 points, knowledge guard -1.6, greedy bag-draw play -10.9, sampled slippery-grid play -4.3.
The card says it is better calibrated on both hard sets than v10 (0.156 versus 0.226 and 0.175 versus 0.307) while still overconfident there, so the calibration improvement is real but not a solved problem.
Apache-2.0, English, 1.88B BF16 parameters in safetensors, tagged text-classification. The Hugging Face repository holds the inference subset of the Mapika/decider package, whose code, data registry and training scripts live on GitHub.
Supervised stages v1 to v8 fine-tuned the base on about 95 public decision datasets, agent trajectories, web element choice, game states and teacher-written custom questions; v10 added 384 steps of calibration-aware RL whose rewards are task outcomes and the exact probability laws of games.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A LoRA plus 24-slot decision head over Qwen3.8-27B from Ming Xu's train-jev project: it serves TypeSafe's System One API with 0.851 Kev transfer-v4 accuracy and ECE 0.053, and needs an 80 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
A 24 GB-GPU alternative from the train-jev project: Qwen3.5-4B plus a rank-16 LoRA and 24-slot decision head scoring 0.647 PK accuracy, 40.4 Decision Index and 43 ms p50 latency, ahead of Kev-4B at the same size.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune for coding decisions: roughly 9k rows from SWE-Gym teach keep/drop file relevance, priority scoring and injection detection, reporting 0.833 accuracy on 1,150 held-out rows at 126 ms p50.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune from Convai Innovations on the 1,200-case typed-decisions benchmark: it reports 0.777 accuracy on the official 400-case test set, above TypeSafe Jev 1.13.0 at 0.727 and the teacher ceiling at 0.735.
Jev Models & Reimplementations#community#open-models#local
Community
A 9B bilingual Turkish and English Jev-class decision model from Berkay Hasret: a 173 MB LoRA adapter on Qwen/Qwen3.5-9B that answers up to 16 typed questions about one text in a single forward pass and reports 0.783 macro accuracy on full JevBench.
Jev Models & Reimplementations#community#open-models#local
Community
A bilingual Turkish and English 4B Jev-class decision model from Berkay Hasret: a roughly 130 MB LoRA adapter on Qwen/Qwen3.5-4B that answers up to 16 typed questions per text in one pass and reports 0.774 JevBench macro accuracy.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.