Jev directory
CommunityJev Models & Reimplementations
JEV-27B-VL on ModelScope
AutoTrust's vision-language decision model on ModelScope: an adapter on Qwen3.8-27B, Apache-2.0, 1,735 downloads, with lora, safetensors, and pytorch assets, trained on the SargeDev/jev-distill-corpus-v3 corpus and English metadata.
- Reading time
- 2 min
- Published by
- Community
- Author
- AutoTrust
- Last verified
- Oct 2026
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Listed on ModelScope as an adapter on Qwen/Qwen3.8-27B under the Qwen3_5ForConditionalGeneration architecture.
- Apache-2.0 licence, 1,735 downloads, and lora, safetensors, and pytorch assets with English as the listed language.
- Trained on SargeDev/jev-distill-corpus-v3, the distillation corpus shared with its JEV-9B sibling.
- The readme capture shows only the vision chat template; the model card and benchmarks were not available, so performance is not established here.
- A vision-language member of the JEV family whose sibling card names autotrust/JEV-27B as the successor to JEV-9B.
Watch out
Only the platform JSON metadata is visible: no accuracy, context window, or hardware guidance appears in the capture, and the readme was cut before any prose. Apache-2.0 is the metadata field, and because this is an adapter on Qwen3.8-27B rather than a merged model, the base model's terms also apply and this page does not spell them out.
Questions about this entry
Short answers drawn from the linked project or page.
An AutoTrust repository listed as a vision-language adapter on Qwen/Qwen3.8-27B, using the Qwen3_5ForConditionalGeneration architecture and trained with the SargeDev/jev-distill-corpus-v3 corpus.
Licence apache-2.0, 1,735 downloads, English language, lora, safetensors, and pytorch libraries, a pytorch framework entry, and an upload from GitHub into the autotrust organisation.
Qwen/Qwen3.8-27B, recorded in the BaseModel field with the relation adapter. Because it is an adapter rather than a full fine-tune, serving it means combining it with that base model.
No benchmark values are visible. The Metrics array is empty in the captured page and the readme was truncated before any model card prose, so accuracy and speed claims cannot be checked from this source alone.
The capture shows no file list for this repository, only library and framework metadata, so the adapter size and memory requirements are not established by this page.
Both live under the autotrust organisation on ModelScope and both train on SargeDev/jev-distill-corpus-v3. The JEV-9B card names autotrust/JEV-27B as its successor, and this repository is the vision-language variant in that family line.
The ModelScope metadata records apache-2.0. The page carries no licence name or link text, and the readme capture includes no licence discussion, so the metadata field is the only licence evidence here.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker 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 NewsX: Happy JevDay: images, box, longer context, and an opt-in to find jaggedness
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
More Jev resources like this
AutoTrust's JEV-35B page on ModelScope: metadata records a fine-tune of Qwen/Qwen3.5-35B-A3B under the Qwen3_5MoeForConditionalGeneration architecture, apache-2.0, 521 downloads, English only, with safetensors and PyTorch files.
Jev Models & Reimplementations#community#open-models#local
Community
Bespoke Labs' 9B decision model on Ollama, fine-tuned from Qwen3.5-9B: Apache-2.0, up to 64 typed questions per call, answers in under 100 ms on a MacBook Pro with an M5 Max, and 75.7% mean accuracy across 13 human-labelled datasets.
Jev Models & Reimplementations#community#open-models#local
Community
The typed-decisions checkpoint of the Laya family on ModelScope: a 421M ModernBERT-large fine-tune reporting 0.766 accuracy over 2,000 decisions, above TypeSafe Jev 1.13.0's published 0.727 and the 0.735 teacher ceiling, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
barakplasma's Replicate deployment of Jev-Omni: ask a typed yes/no or multiple-choice question about an image and get a calibrated probability per option from one forward pass, $0.00098 per run on L40S, Apache-2.0, and #1 on Image JevBench v0.1.
Jev Models & Reimplementations#community#open-models#local
Community
Docker's verified ai/openjev image: an 18.9 GB artifact converting the OpenJev weights from huggingface.co/openjev/openjev, served with llama serve through the /v1/systemone API and released under CC BY-NC 4.0 for non-commercial use only.
Jev Models & Reimplementations#community#open-models#local
Community
A 3.74B language-diffusion decision model that answers 184 of 231 matching adaptation questions (79.65%) against Jev 1.13.0's public 86.58%, matching it on the easy subset at 33 ms p50 on one RTX 5090.
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.