Jev directory
CommunityJev Models & Reimplementations
Jev Arts & Human Taste Embeddings (Qwen3.6-35B-A3B-FP8)
Hugging Face dataset of (state, action) embedding pairs from Qwen/Qwen3.6-35B-A3B-FP8 for post-training contrastive heads on the Jev Decision Index's Arts & Human Taste benchmarks: Humicroedit, cfcolor and the Habermas Machine.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
Highlights
- 23,205 train pairs and 2,286 validation pairs as 2048-d float16 vectors
- Hard negatives at k=3 per row, zero-padded, with valid counts in neg_counts.pt
- Sources: Humicroedit subtask-2 train/dev/FunLines, cfcolor probe_vec and Habermas Machine TRAINING_DATA_V1-V5
- No one-token actions or negatives in this store (0 / 0)
- Metadata records max_model_len 4,096 and state_head_tokens 2,048
Questions about this entry
Short answers drawn from the linked project or page.
Each split holds state_embeddings.pt, action_embeddings.pt, neg_embeddings.pt, neg_counts.pt, state_options_embeddings.pt, has_options.pt, single_tokens.npz and metadata.json. Tensors are written with torch.save and can be read with torch.load(..., mmap=True); row i of every tensor matches samples[i] in metadata.json.
Qwen/Qwen3.6-35B-A3B-FP8, run through a vLLM pooling runner with last-token pooling and L2 normalization, producing 2048-d vectors. The metadata records max_model_len 4,096 and state_head_tokens 2,048.
Humicroedit subtask-2 train, dev and FunLines pairs (12,203), cfcolor probe_vec ratings paired per user with train_vec history (11,980), and initial-round panels of the Habermas Machine TRAINING_DATA_V1-V5 cohorts (1,308).
The split is by task, not by row: a task goes to validation when int(task_id[:8], 16) % 10 == 0. That yields 2,286 validation pairs across 748 tasks, of which 1,993 pairs carry hard negatives.
20,561 of the 23,205 train pairs carry hard negatives at k=3, and 1,993 of the 2,286 validation pairs do. Negatives are zero-padded and neg_counts.pt gives the valid count per row.
The sources' licenses, listed in the dataset documentation. The vectors are a derivative of the source text and carry its terms, so downstream use inherits the terms of the underlying releases.
No. This store has zero one-token actions and zero one-token negatives in both splits, so the separate single-token table used by the language store is not required here.
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 (state, action) embedding pairs from Qwen/Qwen3.6-35B-A3B-FP8 for post-training contrastive heads on the Jev Decision Index's Language Understanding benchmarks: iSarcasmEval, VAST, ACOS and New Yorker caption matching.
Jev Models & Reimplementations#open-models#research#benchmark
Community
Hugging 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
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/jeevesPythonOpen Jev-style decision model built on LiquidAI's LFM2.5-350M backbone, a clone of kev. Serves the System One API contract so the TypeSafe SDK works against a local server, returning probabilities for typed questions from one forward pass.
Jev Models & Reimplementations#open-models#python#local
Communityfranckverrot/lev
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
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.
Jev Models & Reimplementations#open-models#calibration#research
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.