Jev directory
CommunityJev Models & Reimplementations
Jev NLI Qwen3.6-35B-A3B-FP8 Embeddings (256k)
Hugging Face dataset of Qwen3.6-35B-A3B-FP8 state/action embeddings for contrastive post-training on the Jev Decision Index's entailment and verification train splits: ANLI, ContractNLI, NLI4CT, HoVer and RAGTruth.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
Highlights
- 188,120 train and 20,901 validation (state, action) pairs at 2048 dimensions
- Two hard negatives per row, zero-padded, with valid counts in neg_counts.pt
- Covers the ANLI, ContractNLI, NLI4CT, HoVer and RAGTruth train splits
- Every item is a closed-label decision whose state is the index's request rendered as text
- One-token actions and negatives are taken from a separate single-token embedding table
Questions about this entry
Short answers drawn from the linked project or page.
The card lists python scripts/download.py data jev-nli-qwen3.6 --rows 20000 --blocks 8 and a clm-train posttrain --source hf:jev-nli-qwen3.6 command. It is also readable through the Datasets library and the Hub dataset viewer.
Five Jev Decision Index benchmarks: ANLI, ContractNLI and NLI4CT from Language Understanding, HoVer from Retrieval, and RAGTruth from Language Understanding. Each contributes its train split.
188,120 train pairs and 20,901 validation pairs, each with up to 2 hard negatives (k 2). The train split has 62,895 one-token actions or negatives and 86,375 in validation.
Qwen/Qwen3.6-35B-A3B-FP8 via a vLLM pooling runner with last-token pooling and L2 normalization, giving 2048-dimensional vectors with max_model_len 16,384.
The dataset is licensed per-source. The vectors are a derivative of the source text and carry its terms, which the dataset documentation lists.
The index scores ANLI, ContractNLI and NLI4CT test splits, so their dev splits are training data here. HoVer's dev split is the scored one and is not read. RAGTruth keeps both splits in one file, and the reader skips lines whose split is not train.
It is split by task: int(task_id[:8], 16) % 10 == 0 sends a task to validation. Row i of every tensor and samples[i] in metadata.json describe one pair.
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 of Qwen3.6-35B-A3B-FP8 state/action embeddings for post-training contrastive heads on the Jev Decision Index's Knowledge & Reasoning train splits: CommonsenseQA, MMLU auxiliary_train and ChessBench.
Jev Models & Reimplementations#open-models#calibration#research
Community
PostHog's reasoning Jev-style classifier with a diffusion drafter, reporting higher accuracy than Kev-9B and hosted Jev on held-out and JevBench tiers.
Jev Models & Reimplementations#community#evaluation#open-models
Community
PostHog/jeevesPython
PostHog/jeevesPythonHugging Face dataset of (state, action) embedding pairs from Qwen3.6-35B-A3B-FP8, rendered from the ESCI, ToolRet and API-Bank train splits for post-training contrastive retrieval heads. Ships float16 tensors, hard negatives and per-row metadata.
Jev Models & Reimplementations#open-models#research#evaluation
Community
An independent, from-scratch reconstruction of a Jev-style decision model that never generates text: a state plus typed questions returns a probability for every option. Trained on Qwen3-0.6B and 1.7B backbones with LoRA and served through a system_one-shaped API.
Jev Models & Reimplementations#open-models#python#research
CommunityES7/JEV-from-Scratch
SecJev is a family of security-specialized System One decision models from 0.8B to 9B parameters, built on Kev's single-pass candidate scorer. It spans 14 tasks and eight sources, and SecJev-0.8B beats general Kev-9B by 20.51 points in task-macro accuracy.
Jev Models & ReimplementationsArticle#community#research#arxiv
Community
Shows that open instruction-tuned models already read decisions from answer-letter logits without training, and that LoRA fine-tuning with a decision head lifts them past Jev: Gemma 4 31B goes from 79.6 to 84.4 on the new 2,352-question OpenDecisions set and scores 81.5 on JevBench v1.7.
Jev Models & ReimplementationsArticle#community#research#arxiv
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.