CommunityJev Models & Reimplementations
JevEmbed Qwen3 Embedding 0.6B
A Qwen3-Embedding-0.6B fine-tune merged from LoRA that raises JevEmbed-Data test accuracy from 33.79% to 82.30%, producing 1,024-dimensional embeddings for Choice, Score and Noul decisions.
- Published by
- Community
- Author
- HIT-TMG
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- LoRA fine-tune of Qwen3-Embedding-0.6B over all 1,601,157 JevEmbed-Data training questions, then merged for standalone use.
- Test-set hard-label accuracy 82.30% over 64,110 questions, up 48.51 points from the base's 33.79%.
- Noul binary accuracy 94.38% and Choice 84.53%; Score level accuracy 69.55% with MAE down from 1.0107 to 0.4313.
- One epoch on 16 GPUs: LoRA rank 64, alpha 32, effective batch 512, ending at adapter step 3,128.
- 1,024-dimensional embeddings; jevembed.yaml applies a 1,024-token limit, Choice/Score temperature 0.1 and Noul slope 10.
Watch out
Apache-2.0 over Qwen3-Embedding-0.6B. The numbers are on the JevEmbed-Data test split with hard labels, BF16, fixed prompts and 1,024-token truncation, and the card notes training-data source licences vary with a source report in the dataset card. Some examples come from game cases such as Doom and Snake.
Questions about this entry
Short answers drawn from the linked project or page.
A Qwen3-Embedding-0.6B checkpoint fine-tuned with LoRA on all 1,601,157 training questions in JevEmbed-Data and then merged. It produces 1,024-dimensional embeddings that JevEmbed turns into Choice, Score and Noul decisions using the prompts and scoring in jevembed.yaml.
On the same 66,482-question test split: overall hard-label accuracy climbs from 33.79% to 82.30%, Choice from 33.02% to 84.53%, Score level from 28.56% to 69.55%, Noul from 40.05% to 94.38%, and Score MAE drops from 1.0107 to 0.4313.
Install JevEmbed and run python -m jevembed --config jevembed.yaml --input request.json from the model directory, or build the client from ModelConfig.load with model_name_or_path pointed at the resolved directory. The release was tested with transformers 4.51.0 and sentence-transformers 5.3.0.
A 1,024-token limit with truncation, Choice/Score temperature 0.1 and Noul slope 10. For raw embeddings, load the merged model directly with SentenceTransformer(model_dir) instead of going through the decision prompts and scoring.
One epoch on 1,601,157 questions across 16 GPUs on four nodes: LoRA rank 64, alpha 32, dropout 0.05 on Q/K/V targets, per-GPU batch 16 with accumulation 2 (effective batch 512), BF16, learning rate 2e-4 and 10% warmup, ending at adapter step 3,128.
Yes, within 6.9e-7 maximum absolute difference on five inputs, per the MERGE_MANIFEST.json. The adapter remains available in lora/ for anyone who wants to load it separately over the Qwen3-Embedding-0.6B base.
The release is Apache-2.0 over Qwen3-Embedding-0.6B. The card notes training-data source licences vary and points to the JevEmbed-Data processing report for the source-by-source terms; the game examples in the card come from the dataset's Doom, Snake and related cases.
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
58GitHub stars
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/OpenDecisionPythonAutoTrust's combined System 1 and System 2 bundle on Gemma-4-26B-A4B-it: Decision Index 0.2.1 of 58.05, a 24-slot head with noul, choice and score, and the base chat model left bit-identical.
Jev Models & Reimplementations#community#open-models#local
Community
Ruiruiz30's Apple Silicon MLX conversion of Jev-Omni: 4-bit affine language weights on a Mac mini M4 16GB, about 7 GB peak Metal memory and roughly 963 ms text decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
alibiserikbay's Apache-2.0 Jev alternative on Qwen3.5-4B with a merged rank-16 LoRA: one forward pass returns a probability per option, 0.784 on JevBench's public hard tier at about 13 ms per decision.
Jev Models & Reimplementations#community#open-models#local
Community
The 9B JevK5: the same v0.3 data on Qwen3.5-9B, retrained for 1.5 epochs by a pre-registered sweep, reaching index proxy 0.794 and a 0.859 hand-written hard set, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
Mithril's open, Jev-shaped typed-decision model on DeBERTa-v3-large: choice up to 255 options, score and noul from one forward pass, 0.854 in-domain with ECE 0.022, trained on public gold labels only.
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.