Jev directory
CommunityJev Models & Reimplementations
Jev Boost re7-bright-mmbright Embeddings
Hugging Face dataset of Jev-shaped retrieval and classification decisions built from MM-BRIGHT, holding Qwen3.6-35B-A3B-FP8 state, action, and hard-negative embeddings for post-training CLM heads.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
Highlights
- 4,716 train pairs and 542 validation pairs, each with one hard negative (k 1).
- Embeddings are 2048-d float16, produced with vLLM pooling, last token, L2-normalized.
- States render the benchmark request plus its instructions and never list the options.
- One-token actions and negatives are zeroed and resolved from a separate single-token embedding dataset.
- Total file size is 67.2 MB; license follows the source text.
Questions about this entry
Short answers drawn from the linked project or page.
It supplies state, action, and hard-negative embeddings for post-training CLM heads, consumed through clm_train.data.jev_boost as described in docs/jev-boost.md. Rows are shaped like the Jev Decision Index's Retrieval & Classification requests.
It is pulled with clm-train posttrain --source hf:boost-re7-bright-mmbright-qwen3.6. Files are per-split .pt tensors (state_embeddings, action_embeddings, neg_embeddings, neg_counts) plus single_tokens.npz and metadata.json.
Qwen/Qwen3.6-35B-A3B-FP8, embedded with vLLM pooling using the last token, L2-normalized, at 2048 dimensions in float16. The metadata records hidden_size 2,048 and max_model_len 16,384.
Train holds 4,716 pairs and validation holds 542 pairs, each with one hard negative. The dataset viewer shows 2 rows across train and validation splits, and the total file size is 67.2 MB.
vLLM's pooling of a one-token prompt is noise for this model, so those sequences are never embedded. single_tokens.npz points them at jackyk02/qwen3.6-35b-a3b-fp8-single-token-embeddings, which EmbeddingDir applies on load.
The source's license: MM-BRIGHT is CC BY 4.0 and the StackExchange posts are CC BY-SA. The vectors are a derivative of that text, so the same terms carry over to the embeddings.
Items whose state shares a paragraph with a Jev suite item are dropped, and at most 20,000 rows are kept via a hash sample. Hard negatives are the other options, capped at 8 per row, with near-copies of the gold dropped.
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
Hugging Face dataset that renders HoVer claim-verification items as Jev Decision Index retrieval and classification requests, with Qwen3.6-35B-A3B-FP8 state, action, and hard-negative embeddings for post-training CLM heads.
Jev Models & Reimplementations#classification#benchmark#open-models
Community
A family of 0.8B-9B ophthalmic decision models that score candidate answers without autoregressive decoding, trained on RetiJev-Data: 78,158 clinical-decision and diagnosis questions and 50,074 guideline-grading questions. RetiJev-9B answers a case in 158 ms on one GPU.
Jev Models & ReimplementationsArticle#arxiv#open-models#benchmark
Community
Apache-2.0 Hugging Face repository for RSI-JEV-27b, a 27B multimodal decision model based on Qwen3.8-27B that answers supplied multiple-choice questions over images, returning a selected option and probabilities for all options in one forward pass.
Jev Models & Reimplementations#open-models#benchmark#calibration
Community
Open-weight vision-language judge that takes a task instruction plus 10 frames of a robot manipulation attempt and predicts success or failure with calibrated probabilities in a single forward pass. Released with inference code and a 20,628-prediction evaluation set.
Jev Models & ReimplementationsArticle#arxiv#open-models#evaluation
Community
Hugging Face dataset that converts AVeriTeC claim-verification records into Jev Decision Index retrieval and classification requests, embedded with Qwen3.6-35B-A3B-FP8 into 2048-d vectors for post-training CLM heads.
Jev Models & Reimplementations#open-models#classification#benchmark
Community
Hugging Face dataset of Jev Decision Index-shaped state/action embeddings built from Re-DocRED. Holds 13,312 train and 1,517 validation pairs embedded with Qwen3.6-35B-A3B-FP8 at 2048 dimensions for post-training CLM heads.
Jev Models & Reimplementations#open-models#classification#benchmark
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.