Jev directory
CommunityJev Models & Reimplementations
Jev-Style-0.8B-Decision-v3 MLX 8-bit
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.
- Reading time
- 2 min
- Published by
- Community
- Author
- SirSahOl
- Last verified
- Oct 2026
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- 8-bit MLX conversion averaging 8.25 bits per weight; output size 782.1 MB, converted with mlx-lm 0.31.3 in 4.76 seconds.
- Active VRAM footprint of about 990 MB with 8 GB unified memory recommended; 262,144-token context length.
- Card reports 52.23 tokens per second, 19.15 ms TTFT and 919.3 MB peak memory on an Apple M1 with 8 GB, five runs.
- 4-bit (about 450 MB) and 16-bit (about 1620 MB) MLX siblings are linked, and an Ollama Modelfile with stop tokens is documented.
- Apache-2.0; Qwen3_5ForCausalLM architecture at 0.8B parameters; conversion dated 2026-10-13.
Quickstart
bash
pip install mlx-lm
mlx_lm.chat --model SirSahOl/Jev-Style-0.8B-Decision-v3-mlx-8bitWatch out
Card reviewed 2026-10-17; the performance table (52.23 tokens per second, 19.15 ms TTFT and 919.3 MB peak on an Apple M1) and the quick-start examples measure chat generation rather than the decision readout, and the card reports no decision-accuracy comparison between this 8-bit file and the 4-bit or 16-bit MLX siblings.
Questions about this entry
Short answers drawn from the linked project or page.
An 8-bit MLX conversion by SirSahOl of chaoliangUNSW/Jev-Style-0.8B-Decision-v3, formatted for Apple Silicon native GPU inference. The source is a Jev-style decision model built on Qwen3.5; this repository only converts and quantizes it.
Apple Silicon: the card puts the active VRAM footprint at about 990 MB with 8 GB of unified memory as the minimum recommendation, targeting M1, M2, M3 and M4 machines.
Install mlx-lm, then use mlx_lm.chat or mlx_lm.generate with the model name, or load it in Python through mlx_lm.load and apply the chat template. The card also documents an Ollama Modelfile route with stop tokens for im_start, im_end and endoftext.
On an Apple M1 with 8 GB of unified memory, five runs with 256 max tokens: 52.23 tokens per second, 19.15 ms TTFT and 919.3 MB peak memory. These measure chat generation, and the card gives no decision-accuracy numbers for the quantization.
The card links 4-bit (about 450 MB, recommended for 8-16 GB machines), this 8-bit file (about 855 MB on disk) and 16-bit (about 1620 MB) MLX versions, and its repository table recommends the closest match to your memory budget.
SirSahOl converted chaoliangUNSW/Jev-Style-0.8B-Decision-v3 with mlx-lm 0.31.3 on 2026-10-13, producing a 782.1 MB output. The card declares Apache-2.0.
No. Jev-Style-0.8B-Decision-v3 is an independent Jev-style decision model on a Qwen3.5 architecture. The conversion card's usage examples are generic chat prompts, so consult the source model's own card for the typed decision readout.
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 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
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
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
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
A local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.
Jev Models & Reimplementations#community#python#open-models
Community
feder-cr/jevC++
feder-cr/jevC++Serves Shanghai AI Lab's Intern-Decision-0.8B vision decision model on Apple Silicon behind a /v1/systemone-shaped endpoint: screenshots plus typed questions in, calibrated Choice, Score and Noul probabilities out, for thresholding and escalation.
Jev Models & Reimplementations#community#apple-silicon#local
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.