Files
laya/pyproject.toml
NandakishorandClaude Opus 5 3152a82b97 feat: Router.attach to share an already-built checkpoint (v0.3.3)
A process that already holds a checkpoint -- the demo Space builds
convaiinnovations/laya at start-up -- had no way to give it to the router, so
routing loaded a second copy: a duplicate 421M parameters, and a multi-second
wait the user sees on their first click.

router.attach(name, agent) registers it, counts it as resident, and raises
max_loaded to hold it. preload() now skips anything already attached.

Measured on a T4: a language flip at max_loaded=1 costs a 10.3 s median reload.
Preloading removes that from the request path entirely.

Co-Authored-By: Claude Opus 5 <[email protected]>
2026-09-19 14:57:50 +05:30

39 lines
1.2 KiB
TOML

[build-system]
requires = ["setuptools>=61.0"]
build-backend = "setuptools.build_meta"
[project]
name = "laya"
version = "0.3.3"
description = "Fast, non-autoregressive System 1 decision engine with calibrated probabilities"
readme = "README.md"
requires-python = ">=3.8"
license = { text = "Apache-2.0" }
authors = [
{ name = "Convai Innovations" }
]
keywords = ["decision-model", "rlcd", "calibration", "system-one", "routing", "guardrails", "moderation", "triage"]
classifiers = [
"Development Status :: 4 - Beta",
"Intended Audience :: Developers",
"License :: OSI Approved :: Apache Software License",
"Programming Language :: Python :: 3",
"Programming Language :: Python :: 3.8",
"Programming Language :: Python :: 3.9",
"Programming Language :: Python :: 3.10",
"Programming Language :: Python :: 3.11",
"Programming Language :: Python :: 3.12",
"Topic :: Scientific/Engineering :: Artificial Intelligence",
]
dependencies = [
"torch>=2.0.0",
"transformers>=4.45.0",
"safetensors>=0.4.0",
"huggingface_hub>=0.20.0",
"numpy>=1.20.0",
]
[project.urls]
Homepage = "https://huggingface.co/convaiinnovations/laya"
Demo = "https://huggingface.co/spaces/convaiinnovations/laya-demo"