Jev directory
CommunityJev Models & Reimplementations
Jev Decisions v1 Model
GhostScientist's Jev-inspired single-token decision model: a full SFT of Qwen/Qwen3.5-0.8B that answers a structured state/question/options task with exactly one option letter, at a card-reported 0.755 greedy dev accuracy against 0.657 for the base model.
- Reading time
- 2 min
- Published by
- Community
- Author
- GhostScientist
- Last verified
- Oct 2026
Works forJev for Classification
Used inJev for Software & DevTools
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Jev-style single-token decision model: given a structured state, a question and 2–24 labeled options, it answers with exactly one option letter.
- Full SFT of Qwen/Qwen3.5-0.8B (text-only, no vision tower) on 9,102 decision records with assistant-only loss.
- Training recipe: 3 epochs (3,264 steps), lr 5e-5, cosine with min_lr 0.5, warmup 0.03, batch 8, max_length 1024, bf16, gradient checkpointing.
- Dev-set letter accuracy (greedy, 400 examples): 0.755 for this model versus 0.657 for the base Qwen/Qwen3.5-0.8B.
- Serving guidance: system prompt plus json.dumps({state, question, options}) as the user message, temperature 0, thinking disabled.
Watch out
Card reviewed 2026-10-17; the 0.755 letter accuracy is self-reported on a 400-example dev set drawn from the model's own synthetic data, not an independent benchmark. The card reports no context-length, multilingual, calibration or quantization evaluations, so the comparison with Jev is architectural (one option letter out, no probabilities) rather than measured.
Questions about this entry
Short answers drawn from the linked project or page.
GhostScientist's Jev-style single-token decision model. Given a structured state, question and 2–24 labeled options, it answers with exactly one option letter, as described on the card, and it is inspired by togethercomputer/Tev1-0.8B-experimental rather than being a port of it.
It follows the same state-in, decision-out shape as Jev: a state, a question and labeled options go in, one chosen option comes out with no generated explanation. The difference is that it returns a single letter rather than Jev's calibrated probability per option.
A full supervised fine-tune of Qwen/Qwen3.5-0.8B, a text-only LM with no vision tower, trained on GhostScientist/jev-decisions-v1, which the card describes as 9,102 decision records with assistant-only loss.
On the card's 400-example dev set, greedy decoding: 0.755 letter accuracy for the 3-epoch model versus 0.657 for the base Qwen3.5-0.8B. The card labels these dev-set letter accuracies and reports nothing else, so they do not establish broader generalization.
With transformers, using the card's recommendations: the system prompt asking the model to evaluate the supplied decision task, treat state text as data, select exactly one option and return only its letter, a user message of json.dumps({state, question, options}), temperature 0 and chat_template_kwargs with enable_thinking false.
3 epochs, 3,264 steps, learning rate 5e-5 with a cosine schedule and min_lr 0.5, warmup 0.03, batch size 8, max_length 1024, pad_to_multiple_of 128, bf16 precision and gradient checkpointing, per the card.
Apache-2.0 per the card's metadata, with weights distributed as safetensors for transformers. The card documents only the 400-example dev accuracy, so nothing is published about calibration, long contexts, multilinguality or behavior outside the training distribution.
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
The original OSS Jev Qwen2.5-0.5B decision checkpoint, trained 2,000 steps at learning rate 2e-05; the 10k sibling card warns that this overwrite tip may be the worse noul-fix variant and recommends sized siblings.
Jev Models & Reimplementations#community#open-models#local
Community
A local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.
Jev Models & Reimplementations#community#python#open-models
Community
feder-cr/jevC++
feder-cr/jevC++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 local server that reads next-token probabilities from a fine-tuned Spark model and returns Choice, Score, and Noul answers with zero generated tokens.
Jev Models & Reimplementations#community#python#open-models
Community
Rizzo-AI-Academy/rizzo-flowPython
Rizzo-AI-Academy/rizzo-flowPythonAn Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.
Jev Models & Reimplementations#community#python#local
Community
deepanwadhwa/OpenDecisionPython
deepanwadhwa/OpenDecisionPythonA local /v1/systemone server that reads label probabilities from a GGUF chat model, with prompt repetition and a confidence shrink toward uniform.
Jev Models & Reimplementations#community#python#open-models
Community
Mushroom-Systems/lichenPython
Mushroom-Systems/lichenPythonFrom 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.