Jev directory
CommunityJev Models & Reimplementations
Bespoke-Nimble-9B-v3
Bespoke Labs' CC BY-NC 4.0 LoRA adapter for Qwen/Qwen3.5-9B: given a context and typed questions, a choice among options or yes/no, it returns a probability for each option, scoring 56.88 on Decision Index 0.2.1.
- Reading time
- 2 min
- Published by
- Community
- Author
- Bespoke Labs
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Capabilities
- Open source
Highlights
- LoRA adapter for Qwen/Qwen3.5-9B that answers a choice among options or a yes/no question with a probability for each option.
- Scores 56.88 on Decision Index 0.2.1; the full run, inference code and instructions ship in a companion Hugging Face dataset.
- Released under CC BY-NC 4.0, so it may not be used commercially; the Qwen3.5-9B base is licensed separately under Apache 2.0.
- Card cites five authors, Shreyas Pimpalgaonkar, Sid Devic, Mrinal Kumar, Pratik Dalal and Mahesh Sathiamoorthy, and shows 440 downloads and 12 likes.
Watch out
CC BY-NC 4.0 forbids commercial use. The card is otherwise thin: it gives a licence, base model, Decision Index number and citation, but publishes no parameter counts, training data, latency or per-task results, leaving the full run and inference code to the companion bespokelabs/bespoke-nimble-9b-v3-decision-index dataset.
Questions about this entry
Short answers drawn from the linked project or page.
A LoRA adapter for Qwen/Qwen3.5-9B, made by Bespoke Labs. Given a context and typed questions it returns a probability for each option rather than generating text.
Two types: a choice among options, and a yes/no question. The card describes it as returning a probability for each option given the context and the typed questions.
No. It is released under CC BY-NC 4.0, so commercial use is not allowed; the base model, Qwen/Qwen3.5-9B, is licensed separately under Apache 2.0.
The card reports Decision Index 0.2.1: 56.88 but does not explain the index on this page; the full run, code and instructions live in the bespokelabs/bespoke-nimble-9b-v3-decision-index dataset.
In the companion dataset bespokelabs/bespoke-nimble-9b-v3-decision-index, which the card links for the full run and inference instructions; the adapter itself is on Hugging Face under bespokelabs/Bespoke-Nimble-9B-v3.
Bespoke Labs. The card's citation names Shreyas Pimpalgaonkar, Sid Devic, Mrinal Kumar, Pratik Dalal and Mahesh Sathiamoorthy for the 2026 v3 release.
Parameter counts, training data, latency, per-task results and how the Decision Index is computed are not on the card; only the licence, base model, one score and the citation are. The companion dataset holds the fuller record.
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 small, fast decision model — a system one model — for agentic workflows from Strands Labs: it picks between options or rates on a scale faster than an LLM and returns a calibrated confidence with every decision.
Jev Models & Reimplementations#community#github#python
Community
strands-labs/strands-deciderPython
strands-labs/strands-deciderPythonA 144.3M-parameter bilingual typed decision model fine-tuned from jhu-clsp/mmBERT-small: it answers noul, choice, and score questions in one forward pass with calibrated confidence and zero output tokens, and it runs on CPU.
Jev Models & Reimplementations#community#catalog-import#cobanov
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
The original OSS Jev Qwen2.5-0.5B decision checkpoint, trained 2,000 steps at learning rate 2e-05; the 10k sibling card warns that this overwrite tip may be the worse noul-fix variant and recommends sized siblings.
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
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.