Back to directory

laya

A multilingual, non-autoregressive decision engine that returns typed choice, score and yes/no judgments over text in a single forward pass, across more than 100 languages, with automatic routing to the right checkpoint.

laya

laya is a multilingual, non-autoregressive System 1 decision engine. It returns typed decisions, such as a choice from a set of options, a score, or a yes/no answer, over text in more than 100 languages in a single forward pass, taking about 33 milliseconds, and was trained with reinforcement learning against strictly proper scoring rules.

A router automatically picks the right checkpoint for each request, including a long-context, multilingual checkpoint that can read up to 8,192 tokens instead of the default 1,024-token limit; response time follows the input's real length rather than the configured limit. The router also detects the script and language of non-English text and routes it to the multilingual checkpoint automatically.

It installs with pip or uv, requires Python 3.10 or newer, and has optional extras for an HTTP server, an MCP server, LangChain and LangGraph, LlamaIndex, CrewAI, ONNX Runtime and a GPU fast path; a separate TypeScript/Node.js/browser package, laya-ts, is also published. It is licensed Apache-2.0.

More in Models

See all 405
  • mini-AGI

    github.com

    A continual learning language model that trains from scratch on a single 8 GB VRAM GPU, assembling its own architecture and adapting its size to the hardware and data available rather than shipping as a fixed pretrained model.

  • An open-source contrastive-embedding pipeline that maps technical CVE vulnerability descriptions to formal cybersecurity compliance controls in NIST SP 800-53 and CMMC, published with its model generation, fine-tuning and evaluation code.

  • peekaboolean

    github.com

    A small vision-language model, fine-tuned with LoRA on a 500-million-parameter backbone, that answers typed multiple-choice, ordinal-score and yes/no questions about an image with calibrated probabilities, in about 400 milliseconds on a laptop.