Jev directory
CommunityJev Models & Reimplementations
Kev-0.8B
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.
- 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 (11.3M parameters) plus a pointer head on a frozen Qwen3.5-0.8B-Base, served with the adapter merged.
- Runs on a laptop, an L4 or a 4 GB GPU; the card positions it where memory or cost rules out Kev-4B.
- Card-reported transfer-v4 accuracy 0.697 with Brier 0.397; documents-v1 0.851 and hard-v1 0.665.
- Calibration is one temperature, T = 2.35; states up to 65,536 tokens are served, with 8,192 validated.
- Apache-2.0 adapter and head; a fine-tuning starting point via kev.train --init_from jaredpalmer/kev-0.8b.
Watch out
Card reviewed 2026-10-17; all scores are self-reported on the card, which says out of domain it is markedly less accurate than Kev-4B and truncates its out-of-scope list mid-item on tool-call routing. The validated context is 8,192 tokens even though longer states are accepted by the server.
Questions about this entry
Short answers drawn from the linked project or page.
The smallest Kev decision model: it reads a state and typed questions and returns calibrated option probabilities in one forward pass. It is for typed decisions on hardware too small for Kev-4B, where the task resembles its training families.
It implements TypeSafe's public System One API (POST /v1/systemone) without generating text, intended as a self-hosted stand-in for a System One endpoint and a way to prototype against the API before moving to a larger Kev.
An 11.3M-parameter rank-16 LoRA on a frozen Qwen3.5-0.8B-Base (24 layers, 18 with Gated DeltaNet and 6 full attention) plus a pointer head. States up to 65,536 tokens are served and at least 8,192 per question, but only 8,192 is validated.
A laptop or a 4 GB GPU; the card mentions running on a laptop, an L4 or a 4 GB GPU. It serves through the same System One interface as the family, so a Kev-4B deployment can be swapped down without changing client code.
The card is blunt: out of domain it is markedly less accurate than Kev-4B, with transfer-v4 0.697 and breadth-v1 0.586 among its held-out reports. Use it where memory or cost rules out the 4B, and measure on your own data.
Tool-call routing is out of scope per the card, whose fetched text truncates that list mid-item, and the intended uses are limited to decisions close to its training families: document classification, routing and policy checks over stated rules.
Apache-2.0 adapter and head over an Apache-2.0 base, published 2026-09-24 in the Kev 1.0 line (revision 9a45d25e), with older checkpoints under the tags night2-du-release and v7-base. Fine-tune with kev.train --init_from jaredpalmer/kev-0.8b.
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
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.
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.