CommunityJev Models & Reimplementations
NeoHorse-Jev-4B GGUF
The GGUF edition of TokenRhythm's NeoHorse-Jev-4B: BF16 through Q2_K quantizations that each bundle the language backbone, visual weights, decision head, and tokenizer, plus a shared runtime for text and single-image requests.
- Published by
- Community
- Author
- TokenRhythm
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Five quantizations: BF16 at 9.11 GB down to Q2_K at 2.60 GB, with Q4_K_M at 3.39 GB about 63% smaller than BF16.
- Q4_K_M posts the best three-benchmark average, 83.77% across Nimble, VitaminC, and MASSIVE.
- Each GGUF bundles the language backbone, visual weights, decision head, and tokenizer; no separate mmproj or original weights are needed.
- Ships a shared runtime/ directory whose Python runtime handles text requests and single-image-plus-text requests.
- Q8_0 leads Nimble at 87.59%, while Q3_K_M records the best VitaminC (77.96%) and MASSIVE (87.43%).
Watch out
Apache-2.0 quantized weights of the NeoHorse-Jev-4B checkpoint, with llama.cpp's own licence file included for runtime components. Accuracy falls at the small end: Q2_K averages 81.02% against Q4_K_M's 83.77%. The three-benchmark numbers are the card's own accuracy runs on 282, 599, and 350 examples, and the decision demos shown are of the original model.
Questions about this entry
Short answers drawn from the linked project or page.
The card recommends Q4_K_M: 3.39 GB, about 63% smaller than BF16, with the best three-benchmark average of 83.77%, ahead of Q8_0's 83.46% and BF16's 83.36%.
Q2_K is 2.60 GB and averages 81.02% across Nimble, VitaminC, and MASSIVE (84.04, 74.46, 84.57), the lowest of the five files; Q3_K_M at 2.95 GB averages 83.62%.
No. Each GGUF contains the language backbone, visual weights, decision head, and tokenizer configuration, and the card says to download one GGUF plus the shared runtime/ directory, with no separate mmproj or original model weights.
Yes for a single image. The supplied runtime supports text requests and requests combining one image with text, and the visual weights are inside each quantized file rather than split out.
The quantized weights are Apache-2.0, matching the original model. The repo also ships LICENSE.llama.cpp for the llama.cpp-derived runtime components, and a SHA256SUMS file for the downloads.
Accuracy on Nimble (282 examples), VitaminC (599), and MASSIVE (350), with AVG the unweighted mean of the three and file sizes in decimal GB. Bold marks the best score per column and underline the second-best.
A shared runtime/ directory with runtime.py, example.py, build.py, hidden_bridge.cpp, and the neohorse_decision support package, paired with any one of the five GGUF files. The card's download step uses pip install huggingface_hub and hf download.
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
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
GGUF builds of the JevK5 family for llama.cpp: Q8_0 is effectively lossless (229 of 231 answers match the 4B v0.3 bf16), and every published file clears the card's 95% agreement gate.
Jev Models & Reimplementations#community#open-models#local
Community
The v3 Jev-Style decision model at 2B in GGUF: F16 3.78 GB and Q8_0 2.01 GB match PyTorch FP32 on 100% and 99.7% of parity rows, 73.6% on JevBench v1.4.1 public items.
Jev Models & Reimplementations#community#open-models#local
Community
TokenRhythm's 4B prefill-only decision model built on NeoHorse-1-4B: given a state and application-defined questions it returns Choice, Noul, and Score probabilities without generating text, reporting a 77.70 six-group aggregate under Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
GGUF builds of the Jev-Style v3 decision model on Qwen3.5-0.8B: a 0.53 GB Q4_K_M file the card says is decision-identical to FP32 on 240/240 parity rows, served through a pinned llama.cpp scorer.
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.
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
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
From the Jev community
Posts from builders shipping with Jev right now.