Jev directory
CommunityJev Models & Reimplementations
Kev-0.5B
The original 0.5B Kev prototype: a 9.3M-parameter LoRA and pointer head on Qwen2.5-0.5B that reproduced the architecture Archer Hume inferred for Jev. It scored 0.799 on a held-out split of its six training sources and is superseded by Kev-0.8B, 4B and 9B.
- Reading time
- 2 min
- Published by
- Community
- Author
- Jared Palmer
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- The original prototype: a 9.3M-parameter LoRA plus pointer head on Qwen2.5-0.5B, trained on a laptop in September 2026.
- Reproduces the architecture Archer Hume inferred for TypeSafe's Jev and serves the /v1/systemone contract.
- Held-out split of its six training sources: accuracy 0.799, ECE 0.065 raw and 0.031 after temperature T=1.47.
- Superseded by Kev-0.8B, Kev-4B and Kev-9B; kept on the Hub for reference and reproducibility.
- Serving allows 8,192 tokens per question branch, though training used at most 384 state tokens and 1,024 per branch.
Watch out
Card reviewed 2026-10-17; the 0.799 accuracy and ECE figures are self-reported on the card's own model index, and the card labels it a superseded prototype not intended for any production decision affecting people, because its knowledge is limited to a 0.5B backbone and its calibration is verified only on the training distributions. A transfer comparison number in the fetched text is garbled.
Questions about this entry
Short answers drawn from the linked project or page.
The original Kev prototype: a decision model that takes one state and typed questions and returns a probability distribution per question in one forward pass. It was built to show the mechanism works and to reproduce the architecture Archer Hume inferred for TypeSafe's Jev.
One packed sequence: the state prefix, then one isolated branch per question, with a block-causal attention mask so questions cannot see each other. A pointer head scores every option boundary token against the decide token and softmaxes over the options.
A 494M-parameter Qwen2.5-0.5B backbone with 9.3M trainable parameters (8.8M LoRA plus a 0.46M head), 1.9% of the backbone. Training used at most 384 state tokens and 1,024 per branch; serving allows 8,192 per branch.
Model-index accuracy 0.799 on a held-out split of the six training sources, ECE 0.065 across 10 bins and 0.031 after temperature scaling at T=1.47. The card does not present a JevBench or locked-test result for this prototype.
Run python -m kev.serve --run runs/kev and call POST /v1/systemone, or point typesafe-sdk at base_url http://127.0.0.1:8009. Weights ship as a GitHub release tarball (about 38 MB) with the adapter, head, tokenizer and eval.json.
Any production decision that affects people, including moderation, fraud, credit, hiring, medical or legal routing. The card states its knowledge is limited to a 0.5B backbone, its calibration is verified only on the training distributions and unfamiliar-task outputs were not measured.
Apache-2.0 for the adapter and head; the Qwen2.5-0.5B base is Apache-2.0, and datasets carry their own licences. It was developed by Jared Palmer with Devin (Cognition) and trained on 9,000 records and 13,500 questions from six public datasets.
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
Mapika's NVFP4 fine-tune of Gemma-4-31B-it answers typed Choice, Noul and Score questions with calibrated probabilities and no generated text, self-reporting a 64.27 chance-corrected public Decision Index and 39 ms median latency on a B300.
Jev Models & Reimplementations#community#open-models#local
Community
Standard Thinking's 8B LoRA adapter for the Standard One decision model: merge it with the supplied recipe and serve it, and it scores choice, noul and score labels over a scenario with calibrated probabilities via POST /v1/systemone, never generating text.
Jev Models & Reimplementations#community#open-models#local
Community
The smallest decider from Mapika: a 1.4 GB bf16 Qwen3.5-0.8B-Base fine-tune that returns calibrated probabilities for typed Choice, Score and Noul questions in one forward pass and runs 1.5x faster than decider-2b, within one to four points on routing and lookups.
Jev Models & Reimplementations#community#open-models#local
Community
A local server that reads next-token probabilities from a fine-tuned Spark model and returns Choice, Score, and Noul answers with zero generated tokens.
Jev Models & Reimplementations#community#python#open-models
Community
Rizzo-AI-Academy/rizzo-flowPython
Rizzo-AI-Academy/rizzo-flowPythonFrom 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.