Jev directory
CommunityJev Models & Reimplementations
Decider-0.8B Q8 GGUF
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.
- Reading time
- 2 min
- Published by
- Community
- Author
- ma77145
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- Q8_0 GGUF of Mapika/decider-0.8b converted with llama.cpp's convert_hf_to_gguf.py and llama-quantize; no official 0.8b GGUF existed.
- Measured on a GTX 1050 Ti 4 GB via Vulkan: JevBench hard-public 111 accuracy 0.405 and a toy routing battery at 5/5 in about 0.7 s.
- A fitted temperature of 5.55 cuts NLL from 1.62 to 1.12 and ECE from 0.33 to 0.13; divide log-probs by 5.55 and softmax.
- Runs with llama-server at 920 MB resident, one forward pass per decision; Apache-2.0 inherited from the base model.
Quickstart
bash
llama-server -m decider-0.8b-q8.gguf -c 8192 --device Vulkan1 -ngl 99
curl localhost:8080/completion -H 'Content-Type: application/json' -d '{
"prompt": "Context: My card was charged twice.\n\nQuestion: Which team?\nOptions:\n(A) billing: charges, refunds\n(B) technical: bugs\nAnswer: (",
"temperature": 0, "n_predict": 1, "n_probs": 10}'Watch out
Card reviewed 2026-10-17; the 0.405 JevBench accuracy, 5/5 toy routing battery and 920 MB residency are self-reported on a single GTX 1050 Ti, the harness is a letter-readout prompt rather than the official serving stack, and the fitted temperature of 5.55 has no stated fitting set.
Questions about this entry
Short answers drawn from the linked project or page.
A community Q8_0 GGUF conversion of Mapika/decider-0.8b, the smallest typed-decision System One model. One forward pass returns calibrated Choice, Score and noul answers over caller-defined options.
The card states no official 0.8b GGUF existed, so this repository fills that gap. It was converted with llama.cpp's convert_hf_to_gguf.py and then quantized with llama-quantize to q8_0.
On a GTX 1050 Ti 4 GB via Vulkan with llama.cpp: 0.405 accuracy on JevBench hard-public 111 through a letter-readout harness, a 5/5 toy routing battery in about 0.7 s total, and 920 MB resident.
The author fitted a temperature of 5.55, cutting NLL from 1.62 to 1.12 and ECE from 0.33 to 0.13. To get calibrated confidence, divide the letter log-probs by 5.55 and softmax.
The card shows llama-server with -m on the gguf file, -c 8192, Vulkan device selection and -ngl 99, then a curl to the completion endpoint with n_predict 1 and n_probs 10 to read one letter.
Apache-2.0, inherited from Mapika/decider-0.8b. The conversion adds no separate terms; the card carries the same licence metadata as the base repository and its decision-model tags.
They are one author's GPU runs on a single older card, the benchmark is a letter-readout harness rather than the official Jev serving path, and the temperature was fitted without documenting the fitting data.
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
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
prithivMLmods' GGUF builds of autotrust/JEV-35B, the Qwen3.5-35B-A3B MoE decision model that returns calibrated choice, noul and score answers over text and images; files run from BF16 to Q3_K_L for llama.cpp.
Jev Models & Reimplementations#community#open-models#local
Community
A 4-bit AWQ plus GPTQ build of AutoTrust's JEV-35B from AI Outfitters: INT4 weights with FP8 and INT8 activations, calibrated on all 256 experts, and served by vLLM on a single 24 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
A 4-bit W4A8 AWQ plus GPTQ build of AutoTrust's JEV-27B-VL from AI Outfitters that keeps the rank-32 LoRA decision head and a bf16 vision tower, served by vLLM on a single 24 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
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.
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.