CommunityJev Models & Reimplementations
Tiny-Jev
A 0.6B non-autoregressive decision model on Qwen3-0.6B that returns calibrated Choice, Score and Noul distributions in one forward pass, scoring 95.8% on its 22,110-item held-out test split.
- Published by
- Community
- Author
- lostargon
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- One safetensors file holds a Qwen3-0.6B fine-tune with LoRA merged in plus the decision head; deps are torch and transformers.
- Held-out test split of 22,110 items: 95.8% accuracy, ECE 0.004 and 99.8% accuracy on the 90% of items at confidence 0.9+.
- OOD template and dialogue families 93.5% over 48,021 items; hand-written business domains drop to 71.1% on 152 items.
- Choice, Score and Noul come out of one forward pass over option marker tokens; no text is generated or parsed.
- The card states it is not affiliated with TypeSafe AI and shares no code or weights with their models.
Watch out
Synthetic rule-labelled training data: on 152 hand-written business-domain items the card reports 71.1% accuracy and ECE 0.191, far below its in-distribution numbers, so calibration should be re-checked on your own distribution. The card does not state a weight licence, and the larger Tiny-Jev-1.7B sibling is the card's more accurate option.
Questions about this entry
Short answers drawn from the linked project or page.
No. The card states it is an independent open-weights model inspired by the System One idea and the Jev decision API, shares no code or weights with TypeSafe's models, and uses Jev in the name only for the interface style it reproduces.
Choice returns the best option plus a distribution over all supplied options; Score maps an ordered scale to an expected value that can land between levels; Noul returns a probability that a statement about the state is true, implemented as a two-option choice over no and yes.
A call takes a few milliseconds on a consumer GPU and roughly 20-50 ms on an Apple M-series laptop, because the whole model runs in one forward pass and returns a distribution instead of generating tokens. A decide call fans out several questions over one state in a single batch.
Synthetic datasets of typed decisions: roughly a hundred thousand states, each paired with several questions whose labels came from explicit rules or code. Families cover support routing, guardrails for other models and agents, moderation, voice and dialogue, structured reading, and robustness sets.
On 22,110 held-out test items the card reports 95.8% accuracy with ECE 0.004, and 99.8% accuracy on the 90% of items at confidence 0.9 or higher. OOD families score 93.5% over 48,021 items, but 152 hand-written business-domain items drop to 71.1% with ECE 0.191.
Yes. The card points to Tiny-Jev-1.7B as a larger, more accurate sibling, while this 0.6B model keeps the whole decoder stack and decision head in a single safetensors file loaded through trust_remote_code=True with no extra dependencies.
The card does not state a weight licence, so terms are unspecified until the repository itself is checked. The metadata tags describe the model as calibrated, non-autoregressive and routing-oriented but carry no license tag.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsRelated terms
Glossary definitions related to this entry.
More Jev resources like this
AutoTrust's combined System 1 and System 2 bundle on Gemma-4-26B-A4B-it: Decision Index 0.2.1 of 58.05, a 24-slot head with noul, choice and score, and the base chat model left bit-identical.
Jev Models & Reimplementations#community#open-models#local
Community
chenz53's standalone decision model derived from Google Gemma 4 26B A4B IT: full merged BF16 weights, a fitted temperature of 1.736507, and 79.50% equal-type accuracy over 1,100 final questions.
Jev Models & Reimplementations#community#open-models#local
Community
The fine-tuned Laya checkpoint for four typed-decision workflows: 0.766 accuracy on 2,000 held-out decisions, above Jev 1.13.0's published 0.727 and the teacher self-agreement ceiling it trained past.
Jev Models & Reimplementations#community#open-models#local
Community
alibiserikbay's Apache-2.0 Jev alternative on Qwen3.5-4B with a merged rank-16 LoRA: one forward pass returns a probability per option, 0.784 on JevBench's public hard tier at about 13 ms per decision.
Jev Models & Reimplementations#community#open-models#local
Community
The 9B JevK5: the same v0.3 data on Qwen3.5-9B, retrained for 1.5 epochs by a pre-registered sweep, reaching index proxy 0.794 and a 0.859 hand-written hard set, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Mithril's open, Jev-shaped typed-decision model on DeBERTa-v3-large: choice up to 255 options, score and noul from one forward pass, 0.854 in-domain with ECE 0.022, trained on public gold labels only.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Training recipes, looped architectures, and language ports: what the open side of the typed-decision class shipped this month, and how to pick by runtime and calibration instead of the leaderboard.
Jev Models & Reimplementations#open-models#training#local
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
From the Jev community
Posts from builders shipping with Jev right now.