# Laya, a 322M-parameter decision engine, gains 26,639 GitHub stars in nine days

Digest AI · Research · published 2026-09-27T23:55:59Z

Canonical: https://digestai.news/story/laya-a-322m-parameter-decision-engine-gains-26-639-github-stars-in-nin

## Summary

Laya, an open-source decision engine released on September 18, 2026, has rapidly gained traction with 26,639 GitHub stars and 2,326 forks within nine days. Developed by NandhaKishorM and licensed under Apache-2.0, the tool offers a non-autoregressive alternative to using large language models for simple classification tasks. Instead of generating text tokens, Laya performs a single forward pass to answer typed questions, such as multiple-choice labels, intensity scores, or yes/no probabilities, reporting zero output tokens.

The project ships three checkpoints on Hugging Face, including a 421M-parameter English model and a 322M-parameter multilingual model supporting over 100 languages. The repository claims a latency of 33ms per question on a T4 GPU, dropping to 7.2ms when batched. A hands-on test on a CPU-only VM confirmed that the system runs locally without API keys or GPU requirements, though CPU inference is slower at roughly 0.7 seconds per question. The tool includes a built-in abstention mechanism that flags low-confidence answers, allowing developers to route uncertain cases to human review or larger models.

While the speed and cost efficiency are significant, the documentation warns that the shipped checkpoints are over-confident and require users to fit temperature parameters on their own data before relying on the probability scores. Laya is positioned for high-volume, bounded decisions like ticket triage and moderation, where deterministic, fast responses are preferred over the nuanced reasoning of frontier models.

## Key points

- Laya gained 26,639 GitHub stars and 2,326 forks in nine days since its September 18, 2026 release.
- The 322M-parameter multilingual model answers typed questions in one forward pass with zero output tokens.
- Docs warn checkpoints are over-confident; users must fit temperatures on held-out data to trust probabilities.

## Why it matters

Laya offers a cost-effective, low-latency alternative to LLMs for high-volume classification tasks. By eliminating token generation and API costs, it enables scalable, offline decision-making for support triage and moderation, reducing reliance on expensive frontier models for simple judgments.

## Sources

1. [Laya: replace LLM-as-a-judge with a 322M-parameter decision engine (26,639 stars in 9 days, hands-on test)](https://aifrontierpost.com/articles/laya-typed-decisions-tutorial) (aifrontierpost.com, 2026-09-27)

## Cite

Digest AI, "Laya, a 322M-parameter decision engine, gains 26,639 GitHub stars in nine days", 27 September 2026, https://digestai.news/story/laya-a-322m-parameter-decision-engine-gains-26-639-github-stars-in-nin

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/laya-a-322m-parameter-decision-engine-gains-26-639-github-stars-in-nin.json