Jev directory
CommunityJev Models & Reimplementations
Cascadia JEV-27B Decision Tree (OpenVINO int8)
Ready-made Cascadia decision tree that serves autotrust/JEV-27B from OpenVINO int8 weights split into two pipeline stages. Grafts JEV-27B's verbalizer LoRA onto Intel's int8 Qwen3.8-27B IR so /v1/decisions and /v1/systemone run without the exporter.
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- Two-stage tree of about 27 GB; each stage needs 13-25 GB of memory.
- Measured 82/82 parity against the fp32 reference, max |delta p| 0.0187, no flips.
- /v1/decisions and /v1/systemone returned identical results across all 82 questions.
- Weights are byte-identical to pawanpaudel93/jev27b-int8-ov-decision.
- Apache-2.0; redistributes converted weights only and adds no training.
Questions about this entry
Short answers drawn from the linked project or page.
Download the whole tree to the same path on both hosts, then start rank 1 on host B with cascadia worker --model ~/trees/cascadia-jev-27b/int8/stages-2 --engine qwen35 --device CPU --rank 1 --total 2 --listen 0.0.0.0:7801, and rank 0 with the API on host A using --rank 0 --total 2 --next HOST_B:7801 --api 127.0.0.1:8080.
It is built for a two-machine pipeline. Each stage needs 13-25 GB of memory, so a single 32 GB host cannot run both stages. The published measurements used two Intel Core Ultra 7 258V hosts with 32 GB each on Windows 11.
Use OpenVINO GenAI 2026.4.1. The tree has been measured on CPU only, with one rank per host, using --device CPU.
Against the streamed fp32 reference over 50 records and 82 questions, parity was 82/82 PASS (choice 33/33, predicate 24/24, score 25/25) with max |delta p| of 0.0187 and no flips.
It serves /v1/decisions in OpenAI Decisions format and /v1/systemone. In the measurements the two endpoints returned identical results across all 82 questions and all states.
Apache-2.0, matching all inputs: the autotrust/JEV-27B fine-tune, Intel's OpenVINO/Qwen3.8-27B-int8-ov IR, the Qwen base, and the Cascadia conversion. The full license text is in the repository's LICENSE file.
It was exported with tools/decision_export.py from the labscommunity/cascadia branch pawan/decisions-api, grafting JEV-27B's verbalizer LoRA (401 LoRA branches, 200 + 201) onto Intel's stock int8 Qwen3.8-27B IR, with the decision head reading verbalizer token logits.
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
v-jev 1.4 is a Jev-style decision model on Hugging Face in 0.8B and 4B sizes, built on Qwen3.5 backbones. It returns probabilities for typed questions in one forward pass and adds value extraction and image decisions.
Jev Models & Reimplementations#open-models#classification#calibration
Community
A 7.7M-parameter decision model on Hugging Face that reads text or speech and returns one probability per option for choice, score, and noul questions. Speech runs through femto-asr, giving a 10.1M-parameter spoken pipeline.
Jev Models & Reimplementations#open-models#classification#local
Community
Ready-made Cascadia decision tree for autotrust/JEV-9B in OpenVINO int8, packaged as a single stage of about 8.7 GB. Grafts the verbalizer LoRA onto Intel's int8 Qwen3.5-9B IR so one box can serve /v1/decisions and /v1/systemone.
Jev Models & Reimplementations#open-models#local#calibration
Community
arXiv paper presenting EDU-Jev, a family of single-pass System One decision models at 0.8B, 2B and 4B parameters that return a probability distribution over predefined education answer options without generating any text.
Jev Models & ReimplementationsArticle#arxiv#open-models#classification
Community
A 561M-parameter XLM-RoBERTa-large decision model given one extra training epoch on Jev's answers, with about 23% of examples black-box distilled from Jev's public API. Supports noul, choice, and score questions in 18 languages and loads through the bundled naluri.py.
Jev Models & Reimplementations#open-models#calibration#benchmark
Community
MLX builds of Jev-Style-2B-Decision-v4 for Apple silicon: bf16 (3.76 GB) and 8-bit (2.00 GB) folders for text and JSON states, run through one shared jev_style_decision_mlx.py runtime under Apache-2.0.
Jev Models & Reimplementations#open-models#calibration#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.