Jev directory
CommunityJev Models & Reimplementations
Jev-Style-2B-Decision-v4.2-MLX
MLX precision variants of Jev-Style-2B-Decision-v4.2, a 1.88B-parameter text-only decision model for Apple silicon. Ships bf16 and 8-bit weights with a custom runtime that returns a probability for every supplied option.
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- 1,881,825,088 parameters before quantization, built on Qwen3.5-2B-Base.
- Ships bf16 and 8-bit affine (group64) MLX variants; quantized size is 5.76 GB.
- Supports choice, noul and score questions; option order is preserved.
- Input budget of 32,768 tokens and at most 255 options; over-budget requests are rejected.
- The original PyTorch BF16 CUDA model scored 46.74 on the Decision Index public 0.3 suite.
Questions about this entry
Short answers drawn from the linked project or page.
Use Apple silicon and Python 3.12. Install the runtime with pip install "jev-style[mlx]", or download the weights with huggingface_hub, then run jev_style_decision_mlx.py with --precision bf16 or --precision 8bit and pass a state plus a question.
Two: bf16 weights that use FP32 activations as configured in release_config.json, and 8-bit affine group64 weights that use native BF16 activations. Keep release_config.json, readout_config.json and the precision folder's macjev_norms_fp32.safetensors alongside the weights.
choice, noul and score, with option order preserved. The readout dots verdict hidden states with the tied yes/no embedding-row difference, followed by softmax, using a global temperature of 0.8813575635811984 for all three question types.
The input budget is 32,768 tokens with at most 255 options, and over-budget requests are rejected. Text and JSON inputs are accepted. The main Hugging Face model should be used for images, since this release is text-only.
The weights are released under Apache-2.0, which does not relicense third-party data or teacher outputs. The model card directs users to read LICENSE, NOTICE and the parent model card for the full terms.
JevStyle is independently maintained by Chaoliang Yan, without TypeSafe affiliation or endorsement. The release follows the v4 lineage, including TypeSafe Jev 1.14 API soft-label distillation and generated or upstream training material.
No. The card says to use the custom decision runtime rather than stock chat generation, and the runtime pins and checks mlx-lm 0.31.3 internals. It also uses the trained block-attention protocol with shared state computation for multiple questions.
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 Hugging Face checkpoint that fine-tunes UltraBERT-380M into a Jev-style decision model: it reads a state, a question and a list of options and picks one in a single forward pass. Reports 72.6% mean accuracy over 13 datasets on the test split.
Jev Models & Reimplementations#open-models#classification#local
Community
LoRA adapter for Qwen3.5-4B that scores medical multiple-choice options in a single forward pass and returns a short JSON object with the chosen answer key and per-option probabilities, without explanations or chain-of-thought.
Jev Models & Reimplementations#open-models#python#classification
Community
Russian-first open typed-decision model on Hugging Face. Given a context and a set of candidate options it returns a calibrated probability and picks one, covering yes/no, choice, and ordinal scale decisions. Built on mmBERT-base with 307M parameters under an MIT license.
Jev Models & Reimplementations#open-models#calibration#classification
Community
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
A 1.7B-parameter SocialJev preview model on Hugging Face, fine-tuned from Qwen3-1.7B for typed classification in the social sciences. Runs through Transformers with trust_remote_code or the social-jev package with a vLLM backend, returning a choice and probabilities per query.
Jev Models & Reimplementations#open-models#classification#python
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.