Independent Jev-like model scores text options in one pass
Reverse-engineered Jev-like model
A developer has released jevlike, an open-source model that mimics TypeSafe's commercial Jev by scoring a list of text options in a single forward pass instead of generating answers word by word. It includes a byte encoder, an optional frozen Hugging Face encoder, and examples for Doom and chess. On Wikispeedia next-click prediction, a small from-scratch model reached 29% accuracy, and one pass was about 100 times faster than a small decoder writing 400 tokens.
We did not show equal quality with Jev or reproduce TypeSafe's private training method.
- vrc
Out of curiosity and semi unrelated — why do so many of these projects with customized encoder-decoder setups use earlier Qwen versions like 2.5 and 3 and not the smallest 3.5? Purely the few 100m params, or something else in the latter’s arch or pretraining?
- steeve
https://x.com/harshagundal/status/2100044305536889015?s=20
> They were building in stealth for 2 years, I was building in stealth for 2 hours…
> Happy to open source Qwen-2.5-1B-RLCD, 5x faster on-device inference for JSON workloads that need to be type-safe.
- mmastrac
Any diffusion model is potentially a Jev in disguise: https://github.com/vllm-project/vllm/pull/57250
Runs ~0.2s per decision on my DGX Spark.
10/10 programming language detection
9/10 human language detection
10/12 unit magnitude comparison
All incorrect answers are marked with low-P.
It (DiffusionGemma with the Jev mode) can also solve an ASCII maze.