@desplega.ai/laya-server
v0.1.1
Published
HTTP server for @desplega.ai/laya, mirroring laya-serve's /v1/systemone contract (Hono on Node).
Readme
@desplega.ai/laya-server
A Hono HTTP server for @desplega.ai/laya with the same POST /v1/systemone contract as upstream's laya-serve.
npm install @desplega.ai/laya-server
LAYA_MODEL_DIR=./bundles LAYA_MODELS=multilingual node node_modules/@desplega.ai/laya-server/dist/main.jsNode 22+. It serves from a CPU onnxruntime-node session and needs about 3 GB of RAM for one fp32 checkpoint. Checkpoints come from LAYA_MODEL_DIR (export them with tools/export); a listed checkpoint that is missing there is downloaded from the public Hugging Face repo desplega/laya-onnx and verified against pinned SHA-256s. HF_TOKEN is optional (rate limits, or a private mirror).
Most deployments use the container image instead of this package: see the Dockerfile and deployment guide. Environment variables and the request format are documented there and in docs/usage.md. Apache-2.0; see NOTICE.
