Add support for unused params (pytorch#5694)

test/stablehlo/test_saved_model.py

@@ -39,6 +39,24 @@ def test_resnet18_save_load(self):
       output2 = torch.tensor(res.numpy())
       self.assertTrue(torch.allclose(output, output2, atol=1e-5))
 
+  def test_unused_param(self):
+
+    class M(torch.nn.Module):
+
+      def forward(self, a, b):
+        return torch.sin(b)
+
+    model = M()
+    data = (torch.randn(4, 3, 224, 224), torch.randn(1, 100))
+    output = model(*data)
+
+    with tempfile.TemporaryDirectory() as tempdir:
+      save_torch_module_as_tf_saved_model(model, data, tempdir)
+      loaded_m = tf.saved_model.load(tempdir)
+      res = loaded_m.f(data[0].detach().numpy(), data[1].detach().numpy())[0]
+      output2 = torch.tensor(res.numpy())
+      self.assertTrue(torch.allclose(output, output2, atol=1e-5))
+
 
 if __name__ == '__main__':
   test = unittest.main()

torch_xla/stablehlo.py

@@ -135,6 +135,8 @@ class StableHLOFunctionMeta:
   # the arguments the user supplied, OR a parameter, OR a constant
   input_locations: List[InputLocation]
 
+  unused_inputs: List[Tuple[InputLocation, VariableSignature]]
+
   # input_pytree_spec
   input_pytree_spec: Optional[str] = None
   output_pytree_spec: Optional[str] = None
@@ -299,10 +301,16 @@ def _exported_program_to_stablehlo_bundle(exported_model,
       if isinstance(tensor, torch.Tensor)
   }
 
+  # there might be inputs that is part of input but not consumed by HLO graph
+  unused_input_positions = set(range(len(input_args)))
+
   for hlo_input_pos, (tensor_id, tensor_value) in enumerate(
       zip(graph_input_tensor_ids, graph_input_xla_values)):
     if tensor_id in input_ids:  # this is input
-      location = InputLocation.input_arg(position=input_ids[tensor_id])
+      pos_id = input_ids[tensor_id]
+      location = InputLocation.input_arg(position=pos_id)
+      if pos_id in unused_input_positions:
+        unused_input_positions.remove(pos_id)
     elif tensor_id in tensor_id_to_state_name:
       location = InputLocation.parameter(
           name=tensor_id_to_state_name[tensor_id])
@@ -315,6 +323,21 @@ def _exported_program_to_stablehlo_bundle(exported_model,
             shape=list(tensor_value.shape),
             dtype=str(tensor_value.dtype).replace('torch.', '')))
 
+  unused_inputs = []
+  for i in unused_input_positions:
+    pos = InputLocation.input_arg(position=i)
+    arg = input_args[i]
+    if isinstance(arg, torch.Tensor):
+      signature = VariableSignature(
+          shape=list(arg.shape), dtype=str(arg.dtype).replace('torch.', ''))
+    else:
+      signature = VariableSignature(
+          shape=[],
+          dtype=str(type(arg)),
+      )
+
+    unused_inputs.append((pos, signature))
+
   output_signature = [
       VariableSignature(
           shape=list(tensor.shape),
@@ -330,6 +353,7 @@ def _exported_program_to_stablehlo_bundle(exported_model,
       input_signature=input_signatures,
       output_signature=output_signature,
       input_locations=input_locations,
+      unused_inputs=unused_inputs,
       input_pytree_spec=pytree.treespec_dumps(exported_model.call_spec.in_spec),
       output_pytree_spec=pytree.treespec_dumps(
           exported_model.call_spec.out_spec),

torch_xla/tf_saved_model_integration.py

@@ -1,3 +1,4 @@
+import itertools
 import sys
 import os
 from typing import List, Tuple, Any
@@ -44,7 +45,8 @@ def _make_input_signatures(
     meta: stablehlo.StableHLOFunctionMeta) -> List[tf.TensorSpec]:
   input_pos_to_spec = {
       loc.position: spec
-      for loc, spec in zip(meta.input_locations, meta.input_signature)
+      for loc, spec in itertools.chain(
+          zip(meta.input_locations, meta.input_signature), meta.unused_inputs)
       if loc.type_ == stablehlo.VariableType.INPUT_ARG
   }
   for i in range(len(input_pos_to_spec)):