Fix Pyre problems in executorch (pytorch#3955)

Summary:
Pull Request resolved: pytorch#3955

Current problems:

https://www.internalfb.com/phabricator/paste/view/P1410619602

We will eventually migrate from pyre to mypy at some point. But in the meantime, here's a fix.

Reviewed By: tarun292

Differential Revision: D58473224

fbshipit-source-id: a2e789b6c324c98d64c26f5ad5103de60dce513d

backends/xnnpack/test/models/inception_v3.py

@@ -13,7 +13,6 @@
 
 
 class TestInceptionV3(unittest.TestCase):
-    # pyre-ignore
     ic3 = models.inception_v3(weights="IMAGENET1K_V1").eval()  # noqa
     model_inputs = (torch.randn(1, 3, 224, 224),)
 

examples/models/inception_v3/model.py

@@ -8,7 +8,6 @@
 
 import torch
 
-# pyre-ignore
 from torchvision.models import inception_v3  # @manual
 
 from ..model_base import EagerModelBase
@@ -20,7 +19,6 @@ def __init__(self):
 
     def get_eager_model(self) -> torch.nn.Module:
         logging.info("Loading torchvision inception_v3 model")
-        # pyre-ignore
         inception_v3_model = inception_v3(weights="IMAGENET1K_V1")
         logging.info("Loaded torchvision inception_v3 model")
         return inception_v3_model

examples/models/llama2/lib/partitioner_lib.py

@@ -57,8 +57,12 @@ def get_coreml_partitioner(args):
     ), "CoreML backend currently only supports static shape and use_kv_cache=True is the only way to support it at the moment"
     try:
         import coremltools as ct
-        from executorch.backends.apple.coreml.compiler import CoreMLBackend
-        from executorch.backends.apple.coreml.partition import CoreMLPartitioner
+        from executorch.backends.apple.coreml.compiler import (  # pyre-ignore
+            CoreMLBackend,
+        )
+        from executorch.backends.apple.coreml.partition import (  # pyre-ignore
+            CoreMLPartitioner,
+        )
     except ImportError:
         raise ImportError(
             "Please install the CoreML backend follwing https://pytorch.org/executorch/main/build-run-coreml.html"

examples/models/llama2/source_transformation/quantize.py

@@ -17,13 +17,11 @@
 from ..builder import DType
 
 try:
-    # pyre-ignore[21]: Undefined import.
     from fairseq2.nn.embedding import (
         Embedding as fsEmbedding,
         StandardEmbedding as fsStandardEmbedding,
     )
 
-    # pyre-ignore[21]: Undefined import.
     from fairseq2.nn.projection import Linear as fsLinear
 
     print("Using fairseq2 modules.")
@@ -98,10 +96,9 @@ def quantize(
 
         try:
             # torchao 0.3+
-            # pyre-ignore
             from torchao._eval import InputRecorder
         except ImportError:
-            from torchao.quantization.GPTQ import InputRecorder
+            from torchao.quantization.GPTQ import InputRecorder  # pyre-ignore
 
         from torchao.quantization.quant_api import Int8DynActInt4WeightGPTQQuantizer
 
@@ -113,7 +110,7 @@ def quantize(
         )
 
         inputs = (
-            InputRecorder(  # pyre-ignore
+            InputRecorder(
                 tokenizer,
                 calibration_seq_length,
                 None,  # input_prep_func

examples/models/mobilenet_v2/model.py

@@ -8,7 +8,6 @@
 
 import torch
 
-# pyre-ignore
 from torchvision.models import mobilenet_v2  # @manual
 from torchvision.models.mobilenetv2 import MobileNet_V2_Weights
 
@@ -21,7 +20,6 @@ def __init__(self):
 
     def get_eager_model(self) -> torch.nn.Module:
         logging.info("Loading mobilenet_v2 model")
-        # pyre-ignore
         mv2 = mobilenet_v2(weights=MobileNet_V2_Weights.DEFAULT)
         logging.info("Loaded mobilenet_v2 model")
         return mv2
@@ -36,7 +34,6 @@ def __init__(self):
         pass
 
     def get_eager_model(self) -> torch.nn.Module:
-        # pyre-ignore
         mv2 = mobilenet_v2()
         return mv2
 

examples/models/mobilenet_v3/model.py

@@ -19,9 +19,7 @@ def __init__(self):
 
     def get_eager_model(self) -> torch.nn.Module:
         logging.info("Loading mobilenet_v3 model")
-        # pyre-ignore
         mv3_small = models.mobilenet_v3_small(
-            # pyre-ignore[16]
             weights=models.MobileNet_V3_Small_Weights.IMAGENET1K_V1
         )
         logging.info("Loaded mobilenet_v3 model")

examples/models/resnet/model.py

@@ -8,7 +8,6 @@
 
 import torch
 
-# pyre-ignore
 from torchvision.models import (  # @manual
     resnet18,
     ResNet18_Weights,
@@ -25,7 +24,6 @@ def __init__(self):
 
     def get_eager_model(self) -> torch.nn.Module:
         logging.info("Loading torchvision resnet18 model")
-        # pyre-ignore
         resnet18_model = resnet18(weights=ResNet18_Weights.IMAGENET1K_V1)
         logging.info("Loaded torchvision resnet18 model")
         return resnet18_model
@@ -41,7 +39,6 @@ def __init__(self):
 
     def get_eager_model(self) -> torch.nn.Module:
         logging.info("Loading torchvision resnet50 model")
-        # pyre-ignore
         resnet50_model = resnet50(weights=ResNet50_Weights.IMAGENET1K_V1)
         logging.info("Loaded torchvision resnet50 model")
         return resnet50_model

examples/models/torchvision_vit/model.py

@@ -8,7 +8,6 @@
 
 import torch
 
-# pyre-ignore
 from torchvision.models import vit_b_16  # @manual
 
 from ..model_base import EagerModelBase
@@ -20,7 +19,6 @@ def __init__(self):
 
     def get_eager_model(self) -> torch.nn.Module:
         logging.info("Loading torchvision vit_b_16 model")
-        # pyre-ignore
         vit_b_16_model = vit_b_16(weights="IMAGENET1K_V1")
         logging.info("Loaded torchvision vit_b_16 model")
         return vit_b_16_model

exir/serde/export_serialize.py

@@ -1661,7 +1661,7 @@ def deserialize_argument_spec(self, x: Argument) -> ep.ArgumentSpec:
         elif x.type == "as_sym_int":
             return PySymIntArgument(name=x.as_sym_int.as_name)
         else:
-            return PyConstantArgument(value=self.deserialize_input(x))
+            return PyConstantArgument(name="", value=self.deserialize_input(x))
 
     def deserialize_module_call_signature(
         self, module_call_signature: ModuleCallSignature

exir/serde/serialize.py

@@ -756,7 +756,7 @@ def deserialize(
 
 def save(
     ep_save: ep.ExportedProgram,
-    f: Union[str, os.PathLike, io.BytesIO],
+    f: Union[str, os.PathLike[str], io.BytesIO],
     *,
     extra_files: Optional[Dict[str, Any]] = None,
     opset_version: Optional[Dict[str, int]] = None,
@@ -767,7 +767,7 @@ def save(
     artifact: export_serialize.SerializedArtifact = serialize(ep_save, opset_version)
 
     if isinstance(f, (str, os.PathLike)):
-        f = os.fspath(f)
+        f = os.fspath(str(f))
 
     with zipfile.ZipFile(f, "w") as zipf:
         # Save every field in the SerializedArtifact to a file.
@@ -786,13 +786,13 @@ def save(
 
 
 def load(
-    f: Union[str, os.PathLike, io.BytesIO],
+    f: Union[str, os.PathLike[str], io.BytesIO],
     *,
     extra_files: Optional[Dict[str, Any]] = None,
     expected_opset_version: Optional[Dict[str, int]] = None,
 ) -> ep.ExportedProgram:
     if isinstance(f, (str, os.PathLike)):
-        f = os.fspath(f)
+        f = os.fspath(str(f))
 
     extra_files = extra_files or {}
 

exir/tests/test_quantization.py

@@ -48,7 +48,7 @@ def test_resnet(self) -> None:
         with override_quantized_engine("qnnpack"):
             torch.backends.quantized.engine = "qnnpack"
             example_inputs = (torch.randn(1, 3, 224, 224),)
-            m = torchvision.models.resnet18().eval()  # pyre-ignore[16]
+            m = torchvision.models.resnet18().eval()
             m_copy = copy.deepcopy(m)
             # program capture
             m = torch._export.capture_pre_autograd_graph(

exir/wrap.py

@@ -20,7 +20,7 @@
     is_functionaltensor,
 )
 
-from torch._functorch.eager_transforms import _assert_wrapped_functional
+from torch._functorch.eager_transforms import _assert_wrapped_functional  # pyre-ignore
 
 
 def update_with_proxy(t: torch.Tensor, proxy: torch.fx.Proxy) -> torch.Tensor: