Jev directory
CommunityJev Models & Reimplementations
Saina Helm 0.8B on Replicate
run-saina's 0.8B decision model for workflow automation on Replicate: context, question, and answers become structured probabilities for yes/no, single-choice, multi-choice, and rating questions, with distribution or thresholded decision modes.
- Reading time
- 2 min
- Published by
- Community
- Author
- run-saina
- Last verified
- Oct 2026
- Platforms
- API
Highlights
- 0.8B purpose-built decision model for workflow automation, published on Replicate by run-saina, with the project site at saina.run.
- Question types: yes_no, single_choice (default), multi_choice, rating; exclusive choices use softmax, multi-label use independent sigmoid probabilities.
- Three required fields and no JSON: context, question, and at least two unique answers entered one per line.
- Modes: distribution returns scores, decision applies your threshold; request_json batches multiple questions but context, question, and answers remain required.
- Runs on Nvidia L40S hardware with pinned version c9dd1da66266d604acf37819ae24ea5ad372dca587cb9ef2e006bec391fe9c4f; the listing showed 1 public run.
Watch out
The readme calls this an experimental checkpoint whose training-data licence review remains unresolved, including educational/research-only and other source-specific terms, and says commercial-use clearance is not claimed. Probabilities are not calibrated confidence, results shift with wording, descriptions, and option order, and the page advises evaluating on held-out data first.
Questions about this entry
Short answers drawn from the linked project or page.
A purpose-built decision model for workflow automation, published on Replicate by run-saina. It turns context, a question, and your answer options into structured probabilities for classification, routing, or multi-tagging without a free-form response.
Yes/no for binary classification, single_choice for selecting among caller-defined options, multi_choice for tagging with zero, one, or several matches, and rating when answers are ordered from lowest to highest.
Labels travel with each request rather than being fixed to one taxonomy: fill in context, question, and at least two unique answers, one per line. For yes/no, enter exactly two answers with the yes answer first.
distribution returns scores for each answer; decision applies your configured threshold and returns a pick. By default Helm scores the answers and chooses one, with form results appearing under answers.answer.
Advanced API users can supply request_json for multiple questions. It overrides the form values, but context, question, and answers must still be supplied to satisfy the required input schema, and earlier pinned versions keep their original input format.
No. The readme says probabilities are not calibrated confidence, results can change with question wording, option descriptions, and option order, and the model should be evaluated on your own held-out data before you rely on its decisions.
Experimental. The page says training-data licence review remains unresolved, including educational/research-only and other source-specific terms, and commercial-use clearance is not claimed for these weights. The pinned version is c9dd1da66266d604acf37819ae24ea5ad372dca587cb9ef2e006bec391fe9c4f.
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
Together AI's experimental decision family on Ollama, fine-tuned from Qwen3.5 at 4B and 0.8B: 93K downloads, up to 64 questions per call, an 88.0% development accuracy on 1,000 decisions, and MIT-licensed dataset builders and training scripts.
Jev Models & Reimplementations#community#open-models#local
Community
A 27B decision model from torchcast-ai that answers choice, noul and score questions together in one forward pass over the TypeSafe-compatible /v1/systemone format; the card reports Decision Index 0.2.1 of 65.00, +7.09 over Jev 1.13, at 33 ms per yes/no question.
Jev Models & Reimplementations#community#open-models#local
Community
The Ollama build of iapp's OpenThai-SystemOne: a 0.8B Thai and English decision model whose v0.3 weights are fine-tuned for Ollama's /v1/systemone letter readout, reporting a 74.3 macro across 13 public subsets for Q8_0.
Jev Models & Reimplementations#community#open-models#local
Community
A 27B decision model from StartLux Labs that takes text, JSON or images up to 262,144 tokens and returns a probability per option through the TypeSafe /v1/systemone format, reporting Decision Index 0.2.1 63.88 and 208 of 231 on JevBench public.
Jev Models & Reimplementations#community#open-models#local
Community
The Q8_0 8-bit GGUF build of StartLux-Decision-v1-27B, 28.60 GB for llama.cpp, served through a decision wrapper that the card measures at 100% same decisions as the original and 209 of 231 on JevBench public.
Jev Models & Reimplementations#community#open-models#quantized
Community
The Q8_0 8-bit GGUF build of StartLux-Decision-v1-9B, 9.53 GB for llama.cpp, with the card reporting 100.0% same decisions as the original weights and 200 of 231 correct (86.6%) on JevBench public.
Jev Models & Reimplementations#community#open-models#quantized
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.