Jev directory
CommunityJev Models & Reimplementations
Kev-4B
Jared Palmer's Kev-4B: a 33.8M-parameter rank-16 LoRA and pointer head on Qwen3.5-4B-Base that returns calibrated probabilities for typed choice, score and noul questions in one forward pass, sized for one 24 GB GPU or a 32 GB Apple Silicon Mac.
- Reading time
- 2 min
- Published by
- Community
- Author
- Jared Palmer
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Platforms
- API
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- LoRA rank 16 (33.8M parameters) plus pointer head on a frozen Qwen3.5-4B-Base; the adapter is merged at load time.
- Fits one 24 GB GPU or a 32 GB Apple Silicon Mac; the card calls it a self-hosted System One replacement.
- Card-reported transfer-v4 accuracy 0.838 with Brier 0.224; hard-v1 0.803 and devtools-v1 0.756.
- Calibration is one temperature, T = 2.41; states up to 65,536 tokens are served, with 8,192 validated.
- Apache-2.0 adapter and head, an Apache-2.0 base, and a fine-tuning start via kev.train --init_from jaredpalmer/kev-4b.
Watch out
Card reviewed 2026-10-17; the scores are self-reported on the card and the fetched text truncates its out-of-scope list mid-sentence. Training states were at most 7,552 tokens and the validated context is 8,192, far below the 65,536-token states the server accepts.
Questions about this entry
Short answers drawn from the linked project or page.
A decision model that reads one document (the state) and typed questions and returns a calibrated probability distribution over the supplied options in a single forward pass, without generating text, for classification, routing, triage and document checks.
It implements TypeSafe's public System One API (POST /v1/systemone), so the TypeSafe SDK works against it unchanged, and the card positions it as a self-hosted drop-in replacement. It is an independent open-weight model, not TypeSafe's Jev.
Choice questions with 1 to 255 named options, score questions with 1 to 255 ordered levels and noul yes/no questions. Each question answers as its own row from the shared cached state, so questions cannot influence one another.
The card says one 24 GB GPU or a 32 GB Apple Silicon Mac, and that it is a good starting point for fine-tuning on your own labels with kev.train --init_from jaredpalmer/kev-4b.
Held-out results: transfer-v4 accuracy 0.838 with Brier 0.224, hard-v1 0.803 with Brier 0.278, devtools-v1 0.756 with Brier 0.342, breadth-v1 0.690 and tasksource-heldout-v1 development 0.677. All are self-reported.
One temperature, T = 2.41, is stored in head.pt and applied at load time. The model was trained with bf16 autocast over fp32 master weights, served in bf16 with the adapter merged, and evaluated in fp32.
Text generation, chat, summarisation and open-ended question answering are out of scope, since it only scores supplied options, and the card's out-of-scope list continues into fully automated decisions with legal or similar consequences before the fetched text cuts off.
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
Jared Palmer's 27B Kev decision model: a full-weight fine-tune of Qwen3.8-27B averaged with v1, plus a pointer head that returns calibrated probabilities for typed choice, score and noul questions in one forward pass. Card-reported transfer-v4 accuracy is 88.87%.
Jev Models & Reimplementations#community#open-models#local
Community
Jared Palmer's 9B Kev decision model: a rank-16 LoRA and pointer head on a frozen Qwen3.5-9B-Base that answers typed choice, score and noul questions with calibrated probabilities in one forward pass, fitting one 24 GB-class GPU.
Jev Models & Reimplementations#community#open-models#local
Community
The smallest Kev: an 11.3M-parameter LoRA and pointer head on Qwen3.5-0.8B-Base that answers typed choice, score and noul questions with calibrated probabilities in one forward pass, running on a laptop or a 4 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
Standard Thinking's 3B decision model with a joint schema head: it scores every option of every question in a scenario in one forward pass through POST /v1/systemone, and v3 self-scores 51.53 on the Decision Index 0.3 public index, up from v2's 42.90.
Jev Models & Reimplementations#community#open-models#local
Community
Standard Thinking's 8B decision model with a joint schema head that reads the scenario once and scores every option of every question; v4 self-scores 57.40 on the Decision Index 0.3 public index, up from v3's 48.34, with a 70.6M-parameter head.
Jev Models & Reimplementations#community#open-models#local
Community
The smallest decider from Mapika: a 1.4 GB bf16 Qwen3.5-0.8B-Base fine-tune that returns calibrated probabilities for typed Choice, Score and Noul questions in one forward pass and runs 1.5x faster than decider-2b, within one to four points on routing and lookups.
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.