CommunityJev Models & Reimplementations
OpenThai-SystemOne
An open Thai + English System One decision model at 0.8B, Apache-2.0: choice with up to 255 options, fractional score, and noul answers in one forward pass over a Qwen3.5-0.8B text tower.
- Published by
- Community
- Author
- iapp
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Choice accepts up to 255 options, score takes 2–10 ordered levels, and noul returns p(yes), all from one forward pass.
- Backbone: Qwen3.5-0.8B-Base's text tower, continued-pretrained on about 5B Thai tokens after the vision encoder was removed.
- A 256-way slot head replaces the LM head; slot 255 is abstain, and slots beyond the option count are masked out.
- Trained on about 2–3M decision examples with option-order shuffling and abstain examples, then temperature-calibrated with Brier loss.
- Ships an Ollama build and GGUF that serve the same contract at localhost:11434/v1/systemone with no API key.
Watch out
Apache-2.0. The request and response contract mirrors TypeSafe's POST /v1/systemone so TypeSafe SDK code can point at it unchanged. Through Ollama the 256-slot head does not run: each question is asked as a chat prompt and read from A–Z letter probabilities, which caps questions at 26 options and drops the abstain answer and order-invariant mode.
Questions about this entry
Short answers drawn from the linked project or page.
Up to 255 through this repository's API, using <|ts_opt_0|> through <|ts_opt_254|> control tokens; slot 255 is reserved for abstain, meaning none of the options fit.
Ollama does not run the 256-slot head: it asks each question as a chat prompt and reads the probabilities of answer letters A–Z, so questions take 2–26 options and there is no abstain answer or order-invariant mode. Accuracy matches this repo's API.
The text tower of Qwen3.5-0.8B-Base, 24 layers with hybrid Gated-DeltaNet and attention and a 262k context, continued-pretrained on about 5B Thai tokens including accessibility trees and JSON.
The hidden state at each <|ts_answer|> token is projected to 256 logits, slots beyond the option count are masked, and a softmax gives the distribution; score answers are probability-weighted and fractional.
No key for the Ollama path. ollama pull iapp/openthai-systemone runs it locally, ollama pull hf.co/iapp/OpenThai-SystemOne-Ollama:Q8_0 pulls the GGUF build, and the card says nothing leaves your machine.
pip install openthai-systemone, then OPENTHAI_SYSTEMONE_MODEL=iapp/OpenThai-SystemOne uvicorn openthai_systemone.server:app --port 8000, which exposes POST /v1/systemone. Plain transformers also works with trust_remote_code=True.
TypeSafe's POST /v1/systemone, so code written for the TypeSafe SDK can be pointed at this model unchanged. Responses carry choice, probabilities, and confidence per question.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsMore Jev resources like this
An unofficial cross-encoder fine-tune of ModernBERT-ja-310m for Jev Local: it reads a question-plus-state and one candidate per sequence and hits 92.62% on the JGLUE JNLI test, with CC BY-SA 4.0 weights.
Jev Models & Reimplementations#community#open-models#local
Community
The 322M mmBERT-base build of Laya for 100+ languages: state plus typed questions return typed answers with probabilities in one forward pass, with up to 8,192 tokens for long documents.
Jev Models & Reimplementations#community#open-models#local
Community
An Open Classification Foundation smoke checkpoint: a 164M-parameter ModernBERT-base with a cross-entropy decision head, 0.751 accuracy and 0.044 ECE on 2,000 held-out typed decisions.
Jev Models & Reimplementations#community#open-models#local
Community
A 4B decision model taking text, JSON, or images up to 262,144 tokens: 52.75 on Decision Index 0.2.1, 204 of 231 on JevBench public, and 26.0 ms for a three-question request on one H200.
Jev Models & Reimplementations#community#open-models#local
Community
A decision model that scores its own clarifying questions by value of information: at AUC .799 against an oracle .797, it asks only when asking is expected to change the answer, then decides or hands off.
Jev Models & Reimplementations#community#open-models#local
Community
The v3 Jev-Style decision model at 2B in GGUF: F16 3.78 GB and Q8_0 2.01 GB match PyTorch FP32 on 100% and 99.7% of parity rows, 73.6% on JevBench v1.4.1 public items.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Training recipes, looped architectures, and language ports: what the open side of the typed-decision class shipped this month, and how to pick by runtime and calibration instead of the leaderboard.
Jev Models & Reimplementations#open-models#training#local
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
From the Jev community
Posts from builders shipping with Jev right now.