Jev directory
CommunityJev Models & Reimplementations
Kev-4B Artemis GGUF
TurinTech's independent GGUF conversion of Jared Palmer's Kev-4B decision model; the default Q4_K_M file is 2.54 GB and returns the same answer as the BF16 original on 93.0% of 2,000 test decisions with no confident-answer changes.
- Reading time
- 2 min
- Published by
- Community
- Author
- TurinTech
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Shell
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Independent TurinTech GGUF conversion of jaredpalmer/kev-4b, a 4B decision model that answers typed questions with probabilities.
- Default Q4_K_M is 2.54 GB: 93.0% same answers as BF16 on 2,000 test decisions, 0.0 accuracy change, no changed confident answers.
- IQ3_M (2.17 GB) keeps 90.3% same answers at plus 0.4 accuracy; Q3_K_M (2.04 GB) falls to 88.6% and minus 2.2.
- Measured 7.6 GiB peak GPU memory for the default file; tested on an Apple M3 Max with llama.cpp Metal.
- Needs llama.cpp from 9 October 2026 or newer and keeps -ub 4096 -b 4096 because batch size moves probabilities.
Quickstart
bash
llama-server -hf turintech/Kev-4B-Artemis-GGUF:Q4_K_M --port 8080 -c 16384 -ub 4096 -b 4096Watch out
Card reviewed 2026-10-17; the same-answer percentages and accuracy deltas are TurinTech's own measurements on the card's selection panel of 2,000 test decisions, which the card says do not establish performance on unseen inputs. It is an independent conversion not affiliated with or endorsed by the Kev authors, and batch size shifts the probabilities slightly.
Questions about this entry
Short answers drawn from the linked project or page.
An independent TurinTech conversion of jaredpalmer/kev-4b into standard GGUF files for llama.cpp, with a default Q4_K_M file at 2.54 GB that the card recommends starting with.
Kev-4B accepts text with a set of typed questions and returns a probability for every allowed answer in a single pass instead of generating text, served through llama-server's /v1/systemone endpoint.
Q4_K_M 2.54 GB (93.0% same answers, 0.0 accuracy change, default), IQ3_M 2.17 GB (90.3%, plus 0.4), Q3_K_M 2.04 GB (88.6%, minus 2.2); no confident answers changed in any variant. Public ggml-org Q8_0 is 4.48 GB at 98.4% same answers and minus 0.3 accuracy.
Use llama-server with -c 16384 -ub 4096 -b 4096 after installing llama.cpp from 9 October 2026 or newer; the card measured 7.6 GiB peak GPU memory for this setup and tested it on an Apple M3 Max.
The card measured probabilities moving with batch size: on an RTX 5090 the default 512 gave 0.803 and 4096 gave 0.815 on an 854-token input, so the published numbers were taken with 4096.
They were measured on the card's selection panel of 2,000 test decisions and, as the card says, do not establish performance on unseen inputs. It is an independent conversion not affiliated with or endorsed by the Kev authors.
Yes, with an optional configuration of -np 8 -c 32768 --cache-ram 0 (eight slots with 4,096 tokens each and the host prompt cache disabled), which used about 0.6 GiB more GPU memory in the card's RTX 5090 test.
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
TurinTech's independent GGUF conversion of the 27B Kev decision model for llama.cpp: the default IQ4_XS file is 13.91 GB and keeps 97.8% of the BF16 original's answers over 2,000 test decisions, with zero changed confident answers.
Jev Models & Reimplementations#community#open-models#local
Community
TurinTech's independent GGUF conversion of Jared Palmer's Kev-9B decision model; the default IQ3_M file is 4.78 GB and matches the BF16 original on 92.2% of 2,000 test decisions with a minus 1.2-point accuracy change.
Jev Models & Reimplementations#community#open-models#quantized
Community
A community Q8_0 GGUF of Mapika/decider-0.8b, the smallest System One typed-decision model: one forward pass returns calibrated Choice, Score and noul answers, reporting 0.405 on JevBench hard-public at 920 MB resident.
Jev Models & Reimplementations#community#open-models#local
Community
Fluid Inference's fixed-shape Core ML conversion of the classification path of Fastino's GLiNER2.5-Decide for Apple silicon: the DeBERTa-v3-large encoder and shared label classifier score up to four decision heads per call, in fp16, 8-bit or 6-bit packages.
Jev Models & Reimplementations#community#open-models#local
Community
An 8-bit MLX conversion by SirSahOl of chaoliangUNSW/Jev-Style-0.8B-Decision-v3 built for Apple Silicon native GPU inference: about 990 MB of active VRAM, a 262,144-token context and an average of 8.25 bits per weight.
Jev Models & Reimplementations#community#open-models#local
Community
A Q8_0 GGUF of knowlet's Gemma-4-12B-Unified-System-One with a native System One decision runtime and an independently fitted GGUF temperature, shipped with an F16 mmproj for the model's native image and audio projections.
Jev Models & Reimplementations#community#open-models#quantized
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.