Jev directory
CommunityJev Models & Reimplementations
OSS Jev Qwen05 Decision
The original OSS Jev Qwen2.5-0.5B decision checkpoint, trained 2,000 steps at learning rate 2e-05; the 10k sibling card warns that this overwrite tip may be the worse noul-fix variant and recommends sized siblings.
- Reading time
- 2 min
- Published by
- Community
- Author
- gyanbhambhani
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- The first tip of the OSS Jev Qwen2.5-0.5B line, later superseded by the noul-aug, noul-bin and noul-v2 chain.
- 2,000 steps at learning rate 2e-05, batch size 2, seed 42, AdamW fused, finishing at epoch 4.4843.
- Final training loss 0.1651; evaluation loss 0.4809.
- The 10k sibling card calls this possible noul-fix tip worse and prefers sized siblings for Choice workloads.
Watch out
Card reviewed 2026-10-17; the card is auto-generated and reports only a 0.4809 evaluation loss, with no decision-quality numbers of its own, and the 10k sibling card warns that this overwrite tip may be the worse noul-fix variant.
Questions about this entry
Short answers drawn from the linked project or page.
The original checkpoint of the OSS Jev Qwen2.5-0.5B family: Qwen2.5-0.5B-Instruct fine-tuned to answer Choice, Score and Noul questions by one constrained next token over answer codes, released under Apache-2.0.
2,000 steps at learning rate 2e-05 with batch size 2 for training and 8 for evaluation, seed 42, AdamW fused with betas 0.9 and 0.999, a linear schedule, finishing at epoch 4.4843 with training loss 0.1651.
Its own card lists only a 0.4809 evaluation loss. The caution comes from the 10k sibling card, which says this overwrite tip may be the worse noul-fix variant and advises preferring the sized siblings for Choice workloads.
The 10k sibling card labels it an older overwrite tip and suggests it may be the noul-fix variant, which performed worse; later checks used new hub ids (10k, noul-aug, noul-bin, noul-v2) specifically so they would not overwrite this and other early entries.
It belongs to the OSS Jev effort to reproduce TypeSafe Jev's structured judgments (Choice, Noul, Score) locally with a 0.5B model, but this early checkpoint's own card does not document its behaviour on the family's frozen 98-row eval.
Only the loss figures on this card are grounded here; the family's decision numbers live in the sibling cards (10k, noul-bin, noul-v2), all self-reported, and this checkpoint has no accuracy or calibration figures of its own.
Apache-2.0 on the weights, with the base Qwen2.5-0.5B-Instruct licence also applying. In the family, labels are silver distilled from Jev 1.13.0 via the oss-jev-distill dataset, and the 10k card flags the suite as synthetic.
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 small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
Community
prithivMLmods' GGUF builds of autotrust/JEV-35B, the Qwen3.5-35B-A3B MoE decision model that returns calibrated choice, noul and score answers over text and images; files run from BF16 to Q3_K_L for llama.cpp.
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 community Q8_0 GGUF of Mapika/decider-0.8b, the smallest System One typed-decision model: one forward pass returns calibrated Choice, Score and noul answers, reporting 0.405 on JevBench hard-public at 920 MB resident.
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.