[fx] fixed torchaudio conformer tracing (#1392)

tests/test_fx/test_tracer/test_torchaudio_model/test_torchaudio_tacotron.py

@@ -31,7 +31,7 @@ def _get_tacotron2_model(n_mels, decoder_max_step=2000, gate_threshold=0.5):
     )
 
 
-@pytest.mark.skip
+@pytest.mark.skip("Tracing failed")
 def test_tacotron_model():
     n_mels = 80
     n_batch = 3

tests/test_fx/test_tracer/test_torchaudio_model/test_torchaudio_transformer.py

@@ -4,7 +4,6 @@ from torchaudio.models import Emformer, Conformer
 import pytest
 
 
-@pytest.mark.skip
 def test_conformer():
     input_dim = 80
     batch_size = 10
@@ -27,10 +26,17 @@ def test_conformer():
         input = torch.rand(batch_size, int(lengths.max()), input_dim)
         return dict(input=input, lengths=lengths)
 
-    trace_and_compare(model, data_gen, need_meta=False, need_concrete=True)
+    def kwargs_transform(data):
+        new_data = {}
+
+        for k, v in data.items():
+            new_data[f'{k}_1'] = v
+        return new_data
+
+    trace_and_compare(model, data_gen, need_meta=False, need_concrete=True, kwargs_transform=kwargs_transform)
 
 
-@pytest.mark.skip
+@pytest.mark.skip("Tracing failed")
 def test_emformer():
     input_dim = 128
     batch_size = 10

tests/test_fx/test_tracer/test_torchaudio_model/test_torchaudio_wave2vec.py

@@ -40,7 +40,7 @@ def _smoke_test(model, device):
     trace_and_compare(model, data_gen, need_meta=True, need_concrete=False)
 
 
-@pytest.mark.skip
+@pytest.mark.skip("Tracing failed")
 def test_wav2vec():
     for model_fn in MODEL_LIST:
         _smoke_test(model_fn(), 'cpu')

tests/test_fx/test_tracer/test_torchaudio_model/torchaudio_utils.py

@@ -3,21 +3,24 @@ import torch
 from torch.fx import GraphModule, Tracer
 
 
-def trace_and_compare(model, data_gen, need_meta=False, need_concrete=False):
+def trace_and_compare(model, data_gen, need_meta=False, need_concrete=False, kwargs_transform=False):
     data = data_gen()
     concrete_args = data if need_concrete else {}
     meta_args = {k: v.to('meta') for k, v in data.items()} if need_meta else {}
     tracer = ColoTracer()
 
+    model.eval()
+
     graph = tracer.trace(root=model, concrete_args=concrete_args, meta_args=meta_args)
     gm = GraphModule(model, graph, model.__class__.__name__)
     gm.recompile()
 
-    model.eval()
-    gm.eval()
-
     with torch.no_grad():
         non_fx_out = model(**data)
+
+        if kwargs_transform:
+            data = kwargs_transform(data)
+
         fx_out = gm(**data)
     if isinstance(fx_out, tuple):
         for non_fx, fx in zip(non_fx_out, fx_out):