Convai Innovations Releases Laya, An Open-Source Judgment-Only AI Model
Laya tests whether a closed, fee-charging judgment model like Jev can be matched in the open, and the source's own benchmark puts the tradeoff at speed against accuracy above 50 options.
Reporting from 1 source: GameBusiness.jp.
Convai Innovations has released Laya, an open-source decision-making AI that makes judgments instead of generating text. Laya reads input data with prepared typed questions and predicts the optimal option in a single computation, around 33 milliseconds on GPU. It is provided under Apache 2.0. The article states Laya runs about 7.8 times faster than Jev, though Jev is said to hold higher accuracy when more than 50 options are processed.
Laya is the work of the Indian AI company Convai Innovations, and it is distributed under Apache 2.0. Instead of producing prose, it reads a user's email body or JSON alongside preconfigured typed questions (multiple-choice, score-based, Yes/No) and predicts which option is optimal, returning a probability for each.
The source puts a single judgment at about 33 milliseconds in a GPU environment. It also supports automatic detection of more than 100 languages, routes them to the optimal model, and handles long texts of up to 8192 tokens in the multilingual version. Fine-tuning with proprietary data and integration with frameworks such as LangChain are supported.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.
Sources
- GameBusiness.jp 新種の“判断だけAI”が無料で使える「Laya」がオープンソースで登場。Jevの約7.8倍高速（生成AIクローズアップ）