Jev directory
CommunityJev Models & Reimplementations
Laya Vision (201M checkpoint)
A fork of Laya that answers choice, yes/no and score questions about an image plus optional text in one forward pass. This 201M checkpoint is SmolVLM-256M cut to 20 of 30 language layers and trained for two hours, with no text generation.
- Reading time
- 2 min
- Published by
- Community
- Author
- thaitea
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceRuns locally
Highlights
- 201M parameters: SmolVLM-256M-Instruct cut to 20 of its 30 language layers, trained for 2 hours on question-answering data and game frames.
- Answers choice, noul and score questions about an image plus optional text in one forward pass, with per-type temperatures 3.86, 3.05 and 2.10.
- A-OKVQA 4-way choice 59.8% on 1,138 questions; ScienceQA image subset 82.4% on 2,097; VQAv2 yes/no 71.4% on 5,000.
- Games from pixels: ViZDoom basic 0.99, Atari Freeway 0.81, Acrobot 0.76, mean 0.35 where 0 is random play and 1 the expert.
- Same model as thaitea/laya-vision-201m; the previous 237M checkpoint stays at revision d1fbdc0 in this repo's history.
Quickstart
python
import laya
from PIL import Image
agent = laya.load_vlm("thaitea/laya-vision")
result = agent.predict(
{"image": Image.open("photo.jpg"), "note": "customer says it arrived broken"},
{"damage": {"type": "score", "instructions": "How much damage does the item show?",
"criteria": ["none", "cosmetic: scratches or dents", "functional: parts broken or missing", "destroyed"]},
"category": {"type": "choice", "instructions": "What kind of item is this?",
"criteria": ["electronics", "clothing", "furniture", "food", "other"]}},
)Watch out
Model card reviewed 2026-10-15; the fetched text was truncated inside the limitations section. The licence is cc-by-nc-sa-4.0, so commercial use is restricted; games are uneven, small eval sets dropped several points, and the VQAv2 re-split is not comparable to published numbers.
Questions about this entry
Short answers drawn from the linked project or page.
A fork of Laya that answers typed questions about an image plus optional text: choice with per-option probabilities, noul for yes/no, and score on a rubric you write, all in one forward pass with no text generation.
201M parameters: SmolVLM-256M-Instruct cut to 20 of its 30 language layers, with the 12-layer vision tower kept and frozen. It was found by the autoresearch loop and trained for 2 hours on one H100.
A-OKVQA 59.8%, ScienceQA image subset 82.4% and VQAv2 yes/no 71.4% on official validation splits, plus 69.1% accuracy over 59,427 questions across 34 validation sets with ECE 0.041.
Yes, from pixels with one forward pass per move: ViZDoom basic 0.99, Atari Freeway 0.81, Acrobot 0.76, mean 0.35 on a scale where 0 is random play. The previous checkpoint, never trained on games, scores -0.04.
git clone the laya-vision repository, pip install -e ./laya-vision torchvision, then laya.load_vlm("thaitea/laya-vision"). Pin a revision if you need the weights not to change under you.
cc-by-nc-sa-4.0, so non-commercial use with share-alike terms. The card also calls it experimental research and an independent fork of Laya, not affiliated with Convai Innovations.
Games are uneven, with LunarLander worse after training and 6x6 mazes rarely solved, and the smallest eval sets dropped several points. On human vote spreads it beats a predict-the-average baseline on only 3 of 9 sets.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Happy JevDay: images, box, longer context, and an opt-in to find jaggedness
X: First among models under 10B on the Decision Index v0.2.1. Liquid AI's d1-3B is built from LFM2.5-VL-3B and…
X: The classic hotdog classifier as a real-time vision decision. Fayaz Ahmed rebuilt the not hotdog app on…
X: Liquid d1 with vision
More Jev resources like this
An independent fork of Laya that replaces its text encoder with a small vision-language model, giving calibrated typed decisions over an image plus text in one forward pass from a 201M-parameter SmolVLM-256M backbone.
Jev Models & Reimplementations#community#catalog-import#cobanov
Community
r33drichards/laya-visionPython
r33drichards/laya-visionPythonGestalt Labs' Jeff is a LoRA adapter on Qwen3-4B-Instruct-2507 that answers Jev-style choice, noul and score questions with labels and probabilities instead of prose, runs locally, and ships a train-your-own guide; on 9,730 fact-checking examples it trails Jev 1.13.0 on accuracy (0.8183 vs 0.8283).
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
Zefan Cai's LoRA adapter and scalar decision head for Qwen3.5-9B score caller-supplied candidates for choice, yes/no and score answers without generating text, served over a local /v1/systemone endpoint.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
HAL-X AI's 322M non-autoregressive decision model returns calibrated answers in one forward pass of about 10 ms, with native number, excerpt and citation outputs beyond Jev's choice, noul and score types. Apache-2.0, post-trained in 18 languages.
Jev Models & Reimplementations#community#catalog-import#madewithjev
Community
An 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/OpenDecisionPythonJev-compatible decision models with vision input: LoRA fine-tunes of Qwen3.5 at 4B and 9B serving Choice, Score and Noul on the /v1/systemone wire format, with screenshots and photos placed beside text in the state. Runs on your own NVIDIA GPU; weights are non-commercial.
Jev Models & Reimplementations#community#vision#open-weights
Community
Xiaooolong/vevPython
Xiaooolong/vevPythonFrom 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.