CommunityJev Models & Reimplementations
JEV-27B-VL GGUF
Vision-capable GGUF port of autotrust/JEV-27B-VL: BF16 through Q3_K_M with a separate mmproj file, running the calibrated System 1 typed-decision interface and a 27B System 2 through llama.cpp.
- Published by
- Community
- Author
- prithivMLmods
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- GGUF port of autotrust/JEV-27B-VL, a vision extension of JEV-27B pairing a calibrated System 1 decision head with Qwen3.8-27B as System 2.
- Five files: BF16 53.8 GB, Q5_K_M 19.2 GB, Q4_K_M 16.5 GB, Q3_K_M 13.3 GB, plus a 931 MB mmproj-bf16 multimodal projection.
- The reference model accepts image and multimodal prompts up to 256K tokens and extends choice questions to 256 options without retraining.
- Reported image results: 75% on MuJoCo robot-arm pick-and-place and 95% on 60 multi-step browser computer-use tasks, best or tied-best in the JEV family.
- Also reported: VL-RewardBench 78.3%, Plan-RewardBench 73.2%, and AUC 0.727 for zero-shot short-video recommendation from thumbnails.
Watch out
Apache-2.0 quantization; the card describes autotrust/JEV-27B-VL and this repo re-packages it as GGUF. Benchmarks are the card's own zero-shot runs, image-decision calibration is not yet systematically measured, and llama.cpp coverage points only to upstream v0.6.0. The reference service is a patched vLLM script requiring --max-num-seqs 8 for correct multimodal LoRA behavior.
Questions about this entry
Short answers drawn from the linked project or page.
prithivMLmods' GGUF quantization of autotrust/JEV-27B-VL, a vision-capable extension of JEV-27B that keeps the calibrated System 1 typed-decision interface, a yes/no, pick-one, or rating choice returned as a distribution in one forward pass, and pairs it with Qwen3.8-27B as System 2.
Five GGUF files: BF16 at 53.8 GB, Q5_K_M 19.2 GB, Q4_K_M 16.5 GB, Q3_K_M 13.3 GB, and a 931 MB mmproj-bf16 multimodal projection. The card calls Q4_K_M the default and Q5_K_M high quality.
The card points to llama.cpp, release v0.6.0, for GGUF inference. The reference model is otherwise served with a patched vLLM script, serve_decide.py, which requires --max-num-seqs 8 for correct multimodal LoRA behavior; text decisions match the original JEV-27B with a mean probability difference of 0.010 over 1,000 checks.
Yes/no questions, a pick from 2 to 256 options, or a 0-5 rating, each returned as a probability distribution in a single forward pass. Two-letter labels beyond the trained 16 extend choice to 256 options with no retraining.
75% success on a MuJoCo robot-arm pick-and-place task, 95% on 60 multi-step browser computer-use tasks, AUC 0.727 for zero-shot short-video recommendation from thumbnail covers alone, 73.2% on Plan-RewardBench, and 78.3% on VL-RewardBench.
The card reports text-only decisions match the original JEV-27B almost exactly, a mean probability difference of 0.010 across 1,000 checks, and states image-decision calibration is not yet systematically measured.
Apache-2.0, built from unchanged Qwen3.8-27B weights plus the JEV System 1 adapter and decision head. The benchmark figures are self-reported zero-shot runs, and this repo only re-packages the model as GGUF rather than training anything.
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 NewsMore Jev resources like this
The NVFP4 build of AutoTrust's JEV-35B: routed experts quantized to 4-bit cut the download to 25.6 GB and GPU weights to 23.3 GiB, keeping 96.6% of bf16's Decision Index answers.
Jev Models & Reimplementations#community#open-models#local
Community
akhilaaa3's merged multimodal decision classifier on Gemma 4 12B IT: probabilities for text, image, audio or video options, 87.57% on DecisionBench Medium and MMAU 63.10%, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Reza2kn's llama.cpp Q4_K_M build of Jev-Omni: a 6.87 GiB multimodal decision backbone plus projector and FP32 head, measured at 8.98 GiB VRAM on an RTX 5080 Laptop GPU.
Jev Models & Reimplementations#community#open-models#local
Community
AutoTrust's bf16 System 1 decision model on Qwen3.5-35B-A3B with about 3B active parameters: Decision Index 0.3 public 59.73, ERQA 51.3% in one pass with zero robot training, 66.5 GiB of weights.
Jev Models & Reimplementations#community#open-models#local
Community
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
Ruiruiz30's Apple Silicon MLX conversion of Jev-Omni: 4-bit affine language weights on a Mac mini M4 16GB, about 7 GB peak Metal memory and roughly 963 ms text decisions, Apache-2.0.
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
Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.
Jev Models & Reimplementations#multimodal#open-models#decision-model
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
From the Jev community
Posts from builders shipping with Jev right now.