88 lines
1.8 KiB
Nix

{
lib,
buildPythonPackage,
fetchFromGitHub,
# build-system
setuptools,
# dependencies
jinja2,
mlx,
numpy,
protobuf,
pyyaml,
transformers,
# tests
lm-eval,
sentencepiece,
pytestCheckHook,
writableTmpDirAsHomeHook,
}:
buildPythonPackage rec {
pname = "mlx-lm";
version = "0.26.3";
pyproject = true;
src = fetchFromGitHub {
owner = "ml-explore";
repo = "mlx-lm";
tag = "v${version}";
hash = "sha256-O4wW7wvIqSeBv01LoUCHm0/CgcRc5RfFHjvwyccp6UM=";
};
build-system = [
setuptools
];
dependencies = [
jinja2
mlx
numpy
protobuf
pyyaml
transformers
];
nativeCheckInputs = [
lm-eval
pytestCheckHook
sentencepiece
writableTmpDirAsHomeHook
];
pythonImportsCheck = [
"mlx_lm"
];
disabledTestPaths = [
# Requires network access to huggingface.co
"tests/test_datsets.py"
"tests/test_generate.py"
"tests/test_server.py"
"tests/test_tokenizers.py"
"tests/test_utils.py::TestUtils::test_convert"
"tests/test_utils.py::TestUtils::test_load"
"tests/test_utils_load_model.py"
"tests/test_prompt_cache.py::TestPromptCache::test_cache_to_quantized"
"tests/test_prompt_cache.py::TestPromptCache::test_cache_with_generate"
"tests/test_prompt_cache.py::TestPromptCache::test_trim_cache_with_generate"
# RuntimeError: [metal_kernel] No GPU back-end.
"tests/test_losses.py"
"tests/test_models.py::TestModels::test_bitnet"
];
meta = {
description = "Run LLMs with MLX";
homepage = "https://github.com/ml-explore/mlx-lm";
changelog = "https://github.com/ml-explore/mlx-lm/releases/tag/v${version}";
license = lib.licenses.mit;
platforms = [
"aarch64-darwin"
];
maintainers = with lib.maintainers; [ ferrine ];
};
}