Jev directory
CommunityJev Models & Reimplementations
Decider-31b
Mapika's NVFP4 fine-tune of Gemma-4-31B-it answers typed Choice, Noul and Score questions with calibrated probabilities and no generated text, self-reporting a 64.27 chance-corrected public Decision Index and 39 ms median latency on a B300.
- Reading time
- 2 min
- Published by
- Community
- Author
- Mapika
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- Gemma-4-31B-it fine-tune merged into the attention and MLP projections, quantized to NVFP4 at about 31 GB: MLP in NVFP4, attention bf16, FP8 KV cache.
- Self-reported public index 64.27 (raw 73.24), coverage 1.0 and ECE_bw 0.039 over all 140,178 requests of the 0.3 suite.
- Questions below 0.7 top probability get a second reading with reversed options; that touched 28% of public-suite questions.
- Median latency 39 ms and p95 531 ms on one B300; the card's RTX PRO 6000 estimate stays under one second for mean and p80.
- Needs an NVFP4 GPU (B200/B300, RTX PRO 6000, RTX 50 series) and decider-ai 1.9.0; no reasoning trace is ever generated.
Quickstart
bash
python -m venv decider-vllm && . decider-vllm/bin/activate
pip install vllm==0.29.0 fastapi "uvicorn[standard]" jinja2 huggingface_hub
pip install --no-deps "decider-ai>=1.9.0"
DECIDER_MODEL=Mapika/decider-31b uvicorn decider.serve_vllm:app --host 127.0.0.1 --port 8000Watch out
Card reviewed 2026-10-17; the card states its Decision Index run was not verified by the maintainers and is not on the board, temperatures were fitted on the author's own rows, training used public train splits of benchmarks the index tests (test rows were matched and dropped), and results live in a gated dataset. The repository declares Apache-2.0 while the Gemma-4-31B-it base carries Google's Gemma terms.
Questions about this entry
Short answers drawn from the linked project or page.
Mapika's decision model on Gemma-4-31B-it: it answers typed Choice, Noul and Score questions with a probability distribution over the options, generates no text and never writes a reasoning trace.
Each question is rendered in the chat template with thinking off, and the log-probabilities of the option letters are read at the answer slot; a softmax at T1(n) = 2.042 + 0.003 ln n over the option count gives the probabilities, and nothing needs parsing.
When a question's top probability is below 0.7, it is read again with the options reversed, the two log-softmaxes are averaged and a softmax at T2(n) = 1.953 + 0.010 ln n gives the final answer. On the public suite this touched 28% of questions.
The card reports a chance-corrected public index of 64.27 (raw 73.24), ECE_bw 0.039 and coverage 1.0 over all 140,178 requests. Against board values of 2026-10-06 it sits above Perplexity Decider v1.1 (62.25), Quyet-1.0-Large (60.79), Fastino GLiDE (59.06) and Jev (57.96).
One NVFP4-capable GPU: B200/B300, RTX PRO 6000 or an RTX 50 series card with enough memory. The card serves it with vLLM 0.29.0 in a separate environment and decider.serve_vllm from the decider-ai package.
Mapika published the fine-tune and the decider-ai serving package; this repository declares Apache-2.0, while the google/gemma-4-31B-it base carries Google's Gemma licence terms.
It speaks the same System One wire format (POST /v1/systemone) with a state plus typed questions, returning probabilities for every option instead of generated text, and the card places its public index alongside Jev's 57.96.
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 LoRA plus 24-slot decision head over Qwen3.8-27B from Ming Xu's train-jev project: it serves TypeSafe's System One API with 0.851 Kev transfer-v4 accuracy and ECE 0.053, and needs an 80 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
A 24 GB-GPU alternative from the train-jev project: Qwen3.5-4B plus a rank-16 LoRA and 24-slot decision head scoring 0.647 PK accuracy, 40.4 Decision Index and 43 ms p50 latency, ahead of Kev-4B at the same size.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune for coding decisions: roughly 9k rows from SWE-Gym teach keep/drop file relevance, priority scoring and injection detection, reporting 0.833 accuracy on 1,150 held-out rows at 126 ms p50.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune from Convai Innovations on the 1,200-case typed-decisions benchmark: it reports 0.777 accuracy on the official 400-case test set, above TypeSafe Jev 1.13.0 at 0.727 and the teacher ceiling at 0.735.
Jev Models & Reimplementations#community#open-models#local
Community
A 9B bilingual Turkish and English Jev-class decision model from Berkay Hasret: a 173 MB LoRA adapter on Qwen/Qwen3.5-9B that answers up to 16 typed questions about one text in a single forward pass and reports 0.783 macro accuracy on full JevBench.
Jev Models & Reimplementations#community#open-models#local
Community
A bilingual Turkish and English 4B Jev-class decision model from Berkay Hasret: a roughly 130 MB LoRA adapter on Qwen/Qwen3.5-4B that answers up to 16 typed questions per text in one pass and reports 0.774 JevBench macro accuracy.
Jev Models & Reimplementations#community#open-models#local
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.