Jev directory
CommunityJev Models & Reimplementations
Train-Jev 27B LoRA
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.
- Reading time
- 2 min
- Published by
- Community
- Author
- shibing624
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- Qwen3.8-27B with a rank-16 LoRA on every attention, DeltaNet and MLP projection plus a 24-slot fp32 head.
- Kev transfer-v4 accuracy 0.851 vs JEV-27B-VL's 0.852, with much lower calibration error (ECE 0.053 vs 0.097).
- Decision Index 54.6 vs 54.1 for the official model; 13-dataset PK accuracy 0.678 with ECE 0.103.
- Needs an 80 GB GPU; single-request p50 latency is 82 ms on an H20.
- Trained 5,123 steps at batch 128 on SargeDev/jev-distill-corpus-v3 with KL plus 0.5 x RPS on ratings.
Quickstart
bash
git clone https://github.com/shibing624/train-jev.git && cd train-jev
pip install -r requirements.txt
python -m trainjev.serve --base Qwen/Qwen3.8-27B --adapter shibing624/train-jev-27b-lora --port 8020Watch out
Card reviewed 2026-10-17; the whole benchmark table, including the ECE and Decision Index claims, is self-reported by the author's train-jev harness with no third-party reproduction, calibration holds only in distribution, only the first 16 choice labels (A-P) are trained, and the results come from a single seed.
Questions about this entry
Short answers drawn from the linked project or page.
A JEV-style System 1 decision model: Qwen3.8-27B plus a LoRA adapter, a 24-slot decision head and per-type temperatures. It returns a calibrated probability for every option in one forward pass and does not generate text, so confident answers can act and the rest route to a person.
The card reports matching accuracy on the same questions, Kev transfer-v4 0.851 vs 0.852 and Decision Index 54.6 vs 54.1, with a much lower calibration error (ECE 0.053 vs 0.097, NLL 0.411 vs 0.460). Every figure is the author's own evaluation.
The card says it needs an 80 GB GPU; single-request p50 latency is 82 ms on an H20. For a smaller, faster alternative the same project publishes train-jev-4b-lora, which runs on any 24 GB GPU at 43 ms.
Clone github.com/shibing624/train-jev, install requirements and run python -m trainjev.serve with the Qwen/Qwen3.8-27B base and the shibing624/train-jev-27b-lora adapter on port 8020. The server speaks TypeSafe's System One API at /v1/systemone.
The LoRA adapter, the decision head (head.pt) and the calibration temperatures (calibration.json); the 27B backbone downloads on first start. Load it through train-jev, because plain PEFT or Transformers loading applies the adapter but not the head or temperatures.
A rank-16 LoRA on every attention, DeltaNet and MLP projection with a 24-slot fp32 head initialised from the backbone's lm_head, trained 5,123 steps at batch 128 on SargeDev/jev-distill-corpus-v3 using KL to the teacher plus 0.5 x RPS on rating questions.
Calibration holds only in distribution, so check a probability threshold on your own data. Only the first 16 choice labels A-P are trained, so choices with more than 16 options use untrained labels, and the model was trained for decisions only on a single seed.
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 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
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.