Jev directory
CommunityJev Models & Reimplementations
Bekko System One v0 400M
The largest of Bekko's v0 encoder models for Choice, Noul, and Score decisions: 395M parameters built on an Ettin reranker, with shared-prefix encoding and a 1.43 GB browser ONNX export.
- Reading time
- 2 min
- Published by
- Community
- Author
- hotchpotch
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- 395M total parameters, of which 343M are active excluding lookup-only token embeddings.
- Shared-prefix encoding computes instructions and state once per unique prefix and reuses them across candidate options.
- Browser export is 1.43 GB for the 400M variant, 196 MB for 68M, and 29 MB for 17M, with token lookup in row-wise INT8.
- Family spans 17M, 68M, and 400M parameters, all evaluated in parallel over supplied candidates with no text generation.
Watch out
An experimental v0: the card says these models fall far behind Jev 1.13 on S1MB's generalization benchmarks, and their training data includes dataset families represented in the evaluation, so strong results on those tasks do not establish broad generalization. Loading needs custom dynamic-module code rather than a stock pipeline.
Questions about this entry
Short answers drawn from the linked project or page.
The card says these models can perform very well on some tasks but fall far behind Jev 1.13 on S1MB's benchmarks designed to measure generalization; training data also includes dataset families represented in the evaluation.
Use get_class_from_dynamic_module('inference_v0.BekkoSentenceTransformer', repo)(repo, trust_remote_code=True, device='cpu'), then call model.predict(inputs); pass the same revision hash to both calls for reproducibility.
No. It builds on Ettin rerankers with a shared-prefix encoder and task-specific heads, evaluating supplied alternatives in parallel; predict() returns candidate probabilities, a selected option, or a numeric rubric score.
Choice picks among named alternatives with descriptions and returns selected_id and probabilities; Noul returns probability_yes; Score takes a rubric with at least two distinct numeric values and returns score and normalized_score.
Yes. The repo ships onnx_browser/model.onnx, 1.43 GB for the 400M variant. The export quantizes the token lookup table to row-wise INT8 while transformer blocks and task heads stay FP32, so it is not a fully INT8 model.
torch>=2.10,<2.11, transformers==5.17.0, sentence-transformers==6.1.0, safetensors>=0.7, and tqdm>=4.67; FlashAttention 2 and torch.compile() are optional, per the card.
No; the family is ranked on an S1MB leaderboard space, and the card notes that browser demo examples illustrate selected tasks and should not be used as evidence of broad generalization.
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
An unofficial cross-encoder fine-tune of ModernBERT-ja-310m for Jev Local: it reads a question-plus-state and one candidate per sequence and hits 92.62% on the JGLUE JNLI test, with CC BY-SA 4.0 weights.
Jev Models & Reimplementations#community#open-models#local
Community
A 4B decision model taking text, JSON, or images up to 262,144 tokens: 52.75 on Decision Index 0.2.1, 204 of 231 on JevBench public, and 26.0 ms for a three-question request on one H200.
Jev Models & Reimplementations#community#open-models#local
Community
Supersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
A small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
Community
Perplexity's 27B decision model rebuilt on Qwen3.8-27B with a joint request format that answers a whole request's questions in one sequence, reporting 68.09 on Decision Index 0.3 and 77.43 on vision.
Jev Models & Reimplementations#community#open-models#multimodal
Community
The 9B entry in vLLM Semantic Router's Decision 2.0 family: a 7.94B decision model with a 16,384-token context that answers choice, yes/no and score questions in one pass, reporting 68.1 JevArena and 18.4 ms median single-question latency.
Jev Models & Reimplementations#community#open-models#local
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.