diff --git a/test/integration/test_vllm.py b/test/integration/test_vllm.py index 7bb9a6defa..b82f4f9794 100644 --- a/test/integration/test_vllm.py +++ b/test/integration/test_vllm.py @@ -17,10 +17,10 @@ import torch from packaging import version -from torchao.utils import TORCH_VERSION_AT_LEAST_2_8 +from torchao.utils import TORCH_VERSION_AT_LEAST_2_7 -if not TORCH_VERSION_AT_LEAST_2_8: - pytest.skip("Requires PyTorch 2.8 or higher", allow_module_level=True) +if not TORCH_VERSION_AT_LEAST_2_7: + pytest.skip("Requires PyTorch 2.7 or higher", allow_module_level=True) VLLM_AVAILABLE = importlib.util.find_spec("vllm") is not None diff --git a/torchao/quantization/__init__.py b/torchao/quantization/__init__.py index 44fc6c8397..cde2c571a2 100644 --- a/torchao/quantization/__init__.py +++ b/torchao/quantization/__init__.py @@ -111,9 +111,6 @@ ) from .weight_only import WeightOnlyInt8QuantLinear -# TODO: remove after migration of APIs are done -AOPerModuleConfig = ModuleFqnToConfig - __all__ = [ # top level API - auto "autoquant", @@ -154,7 +151,6 @@ "IntxWeightOnlyConfig", "FPXWeightOnlyConfig", "GemliteUIntXWeightOnlyConfig", - "AOPerModuleConfig", "ModuleFqnToConfig", "FbgemmConfig", # smooth quant - subject to change