Kev-0.5B is an open-source decision model with weights on GitHub

@jaredpalmerKev-0.5B: A tiny open source Jev-like decision model with a TypeSafe-compatible API based on Qwen2.5-0.5B that you can train and run on a MacBook Pro. Model card and weights are available on GitHub https://github.com/jaredpalmer/kev

@jaredpalmerKev is a LoRA adapter + a small readout head on top of Qwen2.5-0.5B. It reads a document once and answers many typed questions about it in parallel, in a single prefill pass with no decoding. The doc and every question are packed into one sequence. A block-causal mask lets each question see the document but never another question. A pointer head then scores each question's options against its decision token and applies softmax. Those probabilities are the returned output. The head is trained with cross-entropy against labelled outcomes, so the probabilities are learned rather than generated as text. Kev is trained on about 13,000 samples from Banking77, BoolQ, AG News, MNLI, SST-5, and Yelp Review Full[^1]. Kev is based on @4rcherhume’s blog post theorizing about Jev’s architecture[^2] 1: https://github.com/jaredpalmer/kev/blob/main/MODEL_CARD.md 2: https://archerhume.com/posts/jevs-architecture-unmasked/

@jaredpalmerHere's a little demo of Kev-0.5B running locally on my MacBook Pro M5 playing Chess against itself (h/t @nutlope for the demo idea)

@jaredpalmerKev-0.5B playing chess against itself entirely locally on my MacBook M5 (h/t @nutlope for demo idea)

@mathemagic1an@jaredpalmer @nutlope quite the matchup

@jaredpalmerKev-0.5B is now available on Hugging Face https://huggingface.co/jaredpalmer/kev-0.5b

Kev-0.5B is an open-source decision model with weights on GitHub

TLDR

Combined views

Kev-0.5B is an open-source decision model with weights on GitHub

6 Sources

Combined views

Sentiment

Summary

TLDR

Sentiment

Summary

Sentiment

Summary

6 Sources

Sentiment

Summary