CommunityJev Models & Reimplementations
Drex DLM
Nace.AI's 8B decision model on NVIDIA's Efficient-DLM diffusion backbone: a 256-dim pointer head scores every option in one pass, behind the same POST /v1/systemone API as its hosted service.
- Published by
- Community
- Author
- Nace.AI
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Attention is asymmetric by design: the shared state is bidirectional, question branches are causal, and branches cannot see each other.
- Scoring is pointer-head math — decision markers as queries, option-ending markers as keys — rather than generation.
- Its 52.31 on the Decision Index 0.2 leaderboard put it ahead of five named models in October 2026, per the card.
- Runs on Apple Silicon through the project's llama.cpp and Ollama forks, and on H100s through its Python server.
Watch out
The weights are CC BY-NC 4.0 and cannot be used commercially, though the inference code is Apache-2.0. Context defaults to 16,384 of the supported 32,768 and must be raised explicitly, and on Apple Silicon GGML_METAL_TENSOR_DISABLE=1 is required for long inputs.
Questions about this entry
Short answers drawn from the linked project or page.
The weights cannot: they ship under CC BY-NC 4.0 because of the training data, even though the inference code is Apache-2.0. That split is stated on the card and differs from most Jev-adjacent weights, which are Apache-2.0.
Yes. It serves POST /v1/systemone like the hosted Drex API, and the card says SDKs and servers written for the Jev or Kev format work unchanged. It runs through the project's Python server, its llama.cpp fork, or its Ollama fork.
Clone nace-ai/drex-decision-models, make a Python 3.12 venv, install requirements.txt, and download the weights. python serve.py --model weights/drex-dlm --port 8000 exposes POST /v1/systemone, and GET /health returns ok once loaded.
It is tested on Apple M5 Max (128 GiB), M5 Pro (48 GiB), and NVIDIA H100 80 GB. The BF16 weights take about 16 GB and memory grows with context length and batch size; CPU-only inference has not been tested.
It supports up to 32,768 tokens, but every runner defaults to 16,384, so the full window must be enabled explicitly. On Apple Silicon the llama.cpp runner needs GGML_METAL_TENSOR_DISABLE=1 for long inputs.
The shared state uses bidirectional attention, question branches are causal and cannot see each other, and a shared 256-dim pointer head projects decision markers as queries and option-ending markers as keys. Scaled dot products and a softmax give each question's probabilities.
52.31 on the Decision Index 0.2 leaderboard as of October 2026, ahead of the next five entries, including Decider chat Gemma-4-31B at 51.93 and Jev at 51.67, per the card.
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
Supersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Liquid AI's 3.12B decision model built on LFM2.5-VL: state plus typed questions in one forward pass with zero output tokens, 48.57 on Decision Index 0.2.1 and 8 ms per decision on an RTX 4090.
Jev Models & Reimplementations#community#open-models#local
Community
An untrained readout: stock Gemma-4-31B weights plus a decider config that reads option letters at the answer slot, scored #2 of 70 on Decision Index v0.2.1 at 57.33 against Jev 1.13.0's 57.91.
Jev Models & Reimplementations#community#open-models#local
Community
The vLLM Semantic Router project's 29.37B decision model on Qwen3.8-27B: a probability per supplied option with zero generated tokens, 74.0 on JevArena and 56.5 on Jev Decision Index 0.2.1.
Jev Models & Reimplementations#community#open-models#local
Community
A 4B scorer rebuilt from Qwen3-4B into a branching cross-encoder: one branch per option, no position bias, up to 256 options, and calibrated probabilities with no text to parse.
Jev Models & Reimplementations#community#open-models#local
Community
A 0.4B Jev-compatible decision model with measured order invariance: 960 random option shuffles moved no probability by more than 1e-5, and bev-decider serve exposes /v1/systemone.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
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
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.
Jev Models & Reimplementations#open-models#hugging-face#local
Read article
From the Jev community
Posts from builders shipping with Jev right now.