serve/pyproject.toml

[project]
name = "amos-serve"
version = "1.0.0"
description = "Serving code for the Amos 1 decision models: one HTTP server, POST /v1/systemone, answers typed questions from one forward pass"
readme = "README.md"
requires-python = ">=3.12"
license = "Apache-2.0"
# pinned to the versions the published numbers were measured with: with these three the package answers byte for
# byte like our own servers; transformers 5.19 or the pure-PyTorch linear attention move probabilities by up to 0.02
dependencies = [
    "torch==2.14.0",
    "transformers==5.17.0",
    "flash-linear-attention==0.5.2",
    "safetensors>=0.8.0",
    "tokenizers>=0.23.2",
    "pillow>=12.3.0",
    "huggingface-hub>=1.33.0",
]

[build-system]
requires = ["setuptools>=75"]
build-backend = "setuptools.build_meta"

[tool.setuptools.packages.find]
include = ["amos*"]