Merge pull request #2108 from huggingface/onnx_dynamo

Add support for dynamo based onnx export

Author: rwightman

onnx_export.py

@@ -57,6 +57,8 @@
                     help='Export in training mode (default is eval)')
 parser.add_argument('--verbose', default=False, action='store_true',
                     help='Extra stdout output')
+parser.add_argument('--dynamo', default=False, action='store_true',
+                    help='Use torch dynamo export.')
 
 def main():
     args = parser.parse_args()
@@ -90,6 +92,7 @@ def main():
         check_forward=args.check_forward,
         training=args.training,
         verbose=args.verbose,
+        use_dynamo=args.dynamo,
         input_size=(3, args.img_size, args.img_size),
         batch_size=args.batch_size,
     )

timm/utils/onnx.py

@@ -28,6 +28,7 @@ def onnx_export(
         dynamic_size: bool = False,
         aten_fallback: bool = False,
         keep_initializers: Optional[bool] = None,
+        use_dynamo: bool = False,
         input_names: List[str] = None,
         output_names: List[str] = None,
 ):
@@ -53,7 +54,8 @@ def onnx_export(
     # Opset >= 11 should allow for dynamic padding, however I cannot get it to work due to
     # issues in the tracing of the dynamic padding or errors attempting to export the model after jit
     # scripting it (an approach that should work). Perhaps in a future PyTorch or ONNX versions...
-    original_out = model(example_input)
+    with torch.no_grad():
+        original_out = model(example_input)
 
     input_names = input_names or ["input0"]
     output_names = output_names or ["output0"]
@@ -68,28 +70,40 @@ def onnx_export(
     else:
         export_type = torch.onnx.OperatorExportTypes.ONNX
 
-    torch_out = torch.onnx._export(
-        model,
-        example_input,
-        output_file,
-        training=training_mode,
-        export_params=True,
-        verbose=verbose,
-        input_names=input_names,
-        output_names=output_names,
-        keep_initializers_as_inputs=keep_initializers,
-        dynamic_axes=dynamic_axes,
-        opset_version=opset,
-        operator_export_type=export_type
-    )
+    if use_dynamo:
+        export_options = torch.onnx.ExportOptions(dynamic_shapes=dynamic_size)
+        export_output = torch.onnx.dynamo_export(
+            model,
+            example_input,
+            export_options=export_options,
+        )
+        export_output.save(output_file)
+        torch_out = None
+    else:
+        torch_out = torch.onnx._export(
+            model,
+            example_input,
+            output_file,
+            training=training_mode,
+            export_params=True,
+            verbose=verbose,
+            input_names=input_names,
+            output_names=output_names,
+            keep_initializers_as_inputs=keep_initializers,
+            dynamic_axes=dynamic_axes,
+            opset_version=opset,
+            operator_export_type=export_type
+        )
 
     if check:
         onnx_model = onnx.load(output_file)
         onnx.checker.check_model(onnx_model, full_check=True)  # assuming throw on error
         if check_forward and not training:
             import numpy as np
             onnx_out = onnx_forward(output_file, example_input)
-            np.testing.assert_almost_equal(torch_out.data.numpy(), onnx_out, decimal=3)
-            np.testing.assert_almost_equal(original_out.data.numpy(), torch_out.data.numpy(), decimal=5)
-
+            if torch_out is not None:
+                np.testing.assert_almost_equal(torch_out.numpy(), onnx_out, decimal=3)
+                np.testing.assert_almost_equal(original_out.numpy(), torch_out.numpy(), decimal=5)
+            else:
+                np.testing.assert_almost_equal(original_out.numpy(), onnx_out, decimal=3)