Jev directory
CommunityJev Models & Reimplementations
Spark-S1 on Ollama
Open Spark Jev's Qwen3.5-4B decision model on Ollama: one forward pass returns a calibrated probability per option for allow, ask, or deny tool gating, with 4b-v6 and 4b-v8 tags carrying fitted temperatures of 1.481 and 2.5 under Apache-2.0.
- Reading time
- 3 min
- Published by
- Community
- Author
- Open Spark Jev
- Last verified
- Oct 2026
- Platforms
- CLI
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Qwen3.5-4B fine-tunes in Q8_0 format under Apache-2.0, with 4.5 GB checkpoints and a 256K context window.
- Returns a calibrated probability for every option from one forward pass without generating text; keep menus to 20 options or fewer.
- Two tags: 4b-v6 for best general accuracy at fitted temperature 1.481, and 4b-v8 for agent-harness decisions at 2.5.
- On v6 the Q8_0 build agreed with the full-precision model on 99.5% of 200 held-out decisions; the v8 build has not been checked this way.
- Not a chat model: ollama run prints one letter, so requests need raw: true and logprobs, or the Open Spark Jev gateway on port 8400.
Watch out
It is a decision aid, not an authorization system: the page says to keep deterministic policy checks, least privilege, and human approval in front of anything with side effects, and warns that thresholds and calibration may not transfer to your domain. The project is not affiliated with Jev or TypeSafe AI, the v8 build's calibration is unverified, and the 99.5% figure has no accuracy table.
Questions about this entry
Short answers drawn from the linked project or page.
A calibrated System 1 decision model from Open Spark Jev, built on Qwen3.5-4B. Give it a state and a question with fixed options and it returns a probability for every option from one forward pass, without generating text.
Agent decisions: approve, ask about, or deny a proposed tool call; route a request; check whether an answer is sufficient; or screen tool output for prompt injection. The page presents these as the model's typical uses.
Run one of the tags such as abhishekrai085/spark-s1:4b-v6, send the fully rendered prompt with raw: true and logprobs, then read the first token's top_logprobs, keep the option letters, divide by the tag temperature, and apply softmax. Ollama 0.12 or later is required.
The page tags v6 as best general accuracy, stronger on hard long-document reasoning, fitted at temperature 1.481; v8 targets agent-harness decisions such as next tool or finish, command risk, and tool-output screening at temperature 2.5 with far fewer confidently wrong answers.
Both tags are Qwen3.5-4B fine-tunes in Q8_0 format released under Apache-2.0, with 4.5 GB downloads and a 256K context window. GGUF builds of v6 and v8 are also published on Hugging Face.
It is a decision model, not a chat model, so it scores options instead of writing prose. The page says to use the API with raw: true and logprobs, keeping menus at 20 options or fewer because Ollama returns at most 20 logprobs.
No. The page calls it a decision aid, not an authorization system, and says to put deterministic policy checks, least privilege, and human approval in front of anything with side effects and to validate thresholds and calibration on your own data first.
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 2B decision model from Strands Agents for agentic workflows: a LoRA adapter on Qwen/Qwen3.5-2B-Base with a readout head that answers noul, choice and score questions with calibrated confidence, reporting 0.7619 on JevBench at 4096 context.
Jev Models & Reimplementations#community#open-models#local
Community
Cloudflare's 9B multimodal decision model on Ollama: text, JSON, or images in one non-autoregressive forward pass, Apache-2.0, and the lowest measured latency of any decision model in Cloudflare's Decision Index runs at 38.8 ms median.
Jev Models & Reimplementations#community#open-models#local
Community
Bespoke Labs' 9B decision model on Ollama, fine-tuned from Qwen3.5-9B: Apache-2.0, up to 64 typed questions per call, answers in under 100 ms on a MacBook Pro with an M5 Max, and 75.7% mean accuracy across 13 human-labelled datasets.
Jev Models & Reimplementations#community#open-models#local
Community
ActBro's 2B calibrated System 1 decision model on Ollama, built on OpenBMB's MiniCPM5-2B: 78.79% on JevBench among sub-2B open-weight models, 86.41% on the 10-source decision-v7 set, and one output token per decision via /v1/systemone.
Jev Models & Reimplementations#community#open-models#local
Community
A pointer-network decision model from Keyush Nisar that reads one agent state document plus typed questions and returns calibrated tool-routing probabilities in a single forward pass, reporting 89.00% zero-shot tool routing.
Jev Models & Reimplementations#community#open-models#local
Community
The first v1 checkpoint of the Strands decider line from the Strands Agents team: a LoRA adapter and readout head on Qwen3.5-2B-Base that answers typed decisions with calibrated confidence, reporting 0.7792 JevBench public accuracy (180/231).
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.