Jev directory
CommunityJev Models & Reimplementations
Bekko System One v0 17M
The smallest Bekko System One v0 model: a 17M-parameter English encoder (3.90M excluding lookup-only embeddings, about 29 MB as browser ONNX) that answers choice, noul and score decisions without generating text.
- Reading time
- 2 min
- Published by
- Community
- Author
- hotchpotch
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Platforms
- Browser
- Capabilities
- Self-hostedRuns locally
Highlights
- 17M total parameters, 3.90M excluding lookup-only embeddings; the browser ONNX file is about 29 MB.
- Answers Choice, Noul and Score typed decisions in one encoder pass, returning candidate probabilities rather than generated text.
- Same v0 architecture as the rest of the family: Ettin reranker backbone, shared-prefix encoder and task-specific heads over supplied candidates.
- The card frames the v0 family as an experiment into whether ultra-small models can be capable System One decision models, in Jev's category.
- CPU or CUDA via the exported BekkoSentenceTransformer, with optional FlashAttention 2 and torch.compile(); the browser demo runs 17M and 68M.
Quickstart
bash
pip install 'torch>=2.10,<2.11' 'transformers==5.17.0' 'sentence-transformers==6.1.0' 'safetensors>=0.7' 'tqdm>=4.67'Watch out
Card reviewed 2026-10-17; the card's own note says the v0 models fall far behind Jev 1.13 on S1MB's generalization benchmarks and that training data includes dataset families represented in the evaluation, so strong results on those tasks do not establish broad generalization. No licence is stated, and parameter counts are rounded.
Questions about this entry
Short answers drawn from the linked project or page.
The 17M-parameter variant of Bekko System One v0 from hotchpotch: a compact English encoder for Choice, Noul and Score decisions. It has 16.80M total parameters and 3.90M excluding lookup-only embeddings, with a browser ONNX file of about 29 MB.
It targets the same category as TypeSafe AI's Jev — typed decisions from a supplied state, without generating text — but the card warns the v0 family falls far behind Jev 1.13 on S1MB benchmarks designed to measure generalization.
For a Choice decision it returns candidate probabilities and the selected option; for Noul it estimates the probability of an authored binary condition; for Score it returns the expected value of a numeric rubric. Instructions, state and candidates are supplied in one input object.
Instructions and state share one encoded prefix within a microbatch, and candidates are evaluated in parallel over task-specific heads, so computation is reused across the supplied alternatives rather than repeated per option.
Install torch, transformers, sentence-transformers, safetensors and tqdm, then load BekkoSentenceTransformer through get_class_from_dynamic_module with trust_remote_code=True on CPU or CUDA and call model.predict(inputs). Raw-text encode() is not the typed-decision interface.
The card links a browser demo offering the 17M and 68M variants with CPU/WebGPU execution. The browser export quantizes the token lookup table to row-wise INT8 while keeping transformer blocks and task heads FP32, so it is not fully quantized.
No licence is stated on the card. The encoder builds on cross-encoder/ettin-reranker-17m-v1, and the family shares Ettin reranker building blocks with a shared-prefix encoder and task-specific heads; training and inference code is on GitHub under hotchpotch/bekko-system-one.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Text-to-screenshot (Malay Vasa)
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
X: Google's MediaPipe Decision Maker, tested in the browser
More Jev resources like this
The 68M member of hotchpotch's Bekko System One v0 family: a compact English encoder for choice, noul and score decisions that returns candidate probabilities in parallel, with a ~196 MB browser ONNX export and no text generation.
Jev Models & Reimplementations#community#open-models#local
Community
FRIDA-Decisions answers Russian text with structured decisions in a single encoder pass — choice, ordinal score, yes/no or ranking — and scores 0.893 on the 735-item razvilka benchmark, effectively level with TypeSafe Jev's 0.897 (McNemar p = 0.84).
Jev Models & Reimplementations#community#open-models#russian
Community
Perplexity AI's 27B decision model fine-tuned from Qwen/Qwen3.8-27B, producing option probabilities through a custom head instead of generated text; the card reports 85.71% overall across 11 benchmarks against Jev's 84.51%.
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
The 4B model in vLLM Semantic Router's Decision 2.0 family: a 4.21B fine-tune of Qwen3.5-4B-Base that answers choice, yes/no and score questions in one forward pass, with a self-reported JevArena 63.6 and a 12.9 ms median per request.
Jev Models & Reimplementations#community#open-models#local
Community
The 0.6B model in vLLM Semantic Router's Decision 2.0 family: a 0.60B fine-tune of Qwen3-0.6B-Base with an 8,192-token context that returns probabilities for choice, yes/no and score questions from one forward pass.
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.