Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
# Run from the repository root:
# docker compose -f compose.yaml -f compose.example.yml run --build --rm laya
# For NVIDIA, append -f compose.cuda.yaml before "run".
# This override inherits the image build and persistent cache from compose.yaml.
services:
laya:
environment:
LAYA_DEVICE: "${LAYA_DEVICE:-cpu}"
LAYA_MODEL: "${LAYA_MODEL:-auto}"
LAYA_REQUEST_FILE: /inputs/request.json
OMP_NUM_THREADS: "${OMP_NUM_THREADS:-4}"
HF_HUB_OFFLINE: "${HF_HUB_OFFLINE:-0}"
# Local checkpoint: leave LAYA_MODEL at the auto default above. The entry
# script reads a name here as an explicit alias and refuses an alias and a
# path together, so this file cannot preselect one and still show the mount.
# Enable both this setting and the checkpoint mount below. The directory
# must be writable by UID 10001 for tokenizer compatibility updates.
# LAYA_MODEL_PATH: /models/custom
# Optional token: enable this setting and its read-only mount below.
# The token file must be readable by UID 10001. Never put its value here.
# HF_TOKEN_FILE: /run/secrets/hf_token
volumes:
# Replace the source path with your own request file.
- ./examples/docker/request.json:/inputs/request.json:ro
# - ${LAYA_CHECKPOINT_PATH:?Set LAYA_CHECKPOINT_PATH}:/models/custom
# - ${HF_TOKEN_PATH:?Set HF_TOKEN_PATH}:/run/secrets/hf_token:ro