Jev directory
CommunityJev Models & Reimplementations
OpenThai-SystemOne Ollama
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.
- Reading time
- 3 min
- Published by
- Community
- Author
- iapp
- Last verified
- Oct 2026
- Languages
- Shell
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Ollama build of iapp/OpenThai-SystemOne, a 0.8B Thai and English decision model that returns choice, noul and score probabilities without generating text.
- Served through Ollama's System One API (POST /v1/systemone, Ollama 0.35 or later), the same API Ollama serves Nimble and Tev1 with.
- These v0.3 weights are fine-tuned for Ollama's rendering: 3,000 steps (192k questions), cross-entropy over the candidate letters A–Z.
- Ollama allows 2–26 options per question (the main API supports 255), runs one prompt per question with the prefix cached, and has no abstain mode.
- Reported macro over 13 public subsets: 74.3 for Q8_0 and 74.0 for Q4_K_M, against Tev1 0.8B 64.3, Tev1 4B 74.9 and Nimble 9B 74.8.
Quickstart
bash
ollama pull iapp/openthai-systemone
ollama pull hf.co/iapp/OpenThai-SystemOne-Ollama:Q8_0Watch out
Card reviewed 2026-10-17; the evaluation tables were run by the authors through Ollama's /v1/systemone on the first 800 records of each set and keep only records with up to 26 options, and the card's Thai-sets table is truncated in the copy reviewed. This Ollama path reads next-token letter probabilities rather than the original 256-slot decision head, changing option limits and dropping the order-invariant and abstain modes.
Questions about this entry
Short answers drawn from the linked project or page.
iapp's Ollama build of OpenThai-SystemOne, an open Thai and English System One decision model at 0.8B under Apache-2.0. It takes a state (text or JSON) and typed questions and returns probabilities for choice, noul and score answers, without generating text.
It accepts a state and typed questions and returns probabilities instead of text, served through Ollama's System One API at POST /v1/systemone with Ollama 0.35 or later — the same API Ollama serves Nimble and Tev1 with. Everything runs locally with no API key.
choice between named options, noul (yes/no) and score on an ordered scale. The card's example routes a billing complaint, rates frustration on a three-level scale and answers whether a refund was requested, returning probabilities and confidence for each.
ollama pull iapp/openthai-systemone (ollama.com lists 0.8b as q8_0, plus q4_K_M and bf16), or ollama pull hf.co/iapp/OpenThai-SystemOne-Ollama:Q8_0 to fetch the same files from this repository, then POST to localhost:11434/v1/systemone.
Ollama does not run OpenThai-SystemOne's 256-slot decision head; it renders one chat prompt per question and reads the next-token probabilities of letters A–Z, so these v0.3 weights were fine-tuned for that prompt, with a fitted temperature folded into the final norm.
Through Ollama 0.35 on the first 800 records of each set with at most 26 options: a 74.3 macro across 13 public subsets for Q8_0 and 74.0 for Q4_K_M, compared with Tev1 0.8B 64.3, Tev1 4B 74.9 and Nimble 9B 74.8.
Apache-2.0, the same licence as OpenThai-SystemOne. The model is a 0.8B Qwen3.5-based checkpoint, the GGUF files are plain text models with tied embeddings, and the Modelfile sets the training system prompt with num_ctx 8192.
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 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
A local daemon that pulls open decision models and serves them on TypeSafe's /v1/systemone, so a Jev client switches hosts with TYPESAFE_BASE_URL.
Jev Models & Reimplementations#community#rust#open-models
Community
ollaya-dev/ollayaRust
ollaya-dev/ollayaRustA 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
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
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.