I see that there is warning when run a model using torch.compile(model) way
Warning:
2128922-2023-01-18 10:11:40:751 INFO [logging.py:121] — Step 1: torchdynamo start tracing forward
2128922-2023-01-18 10:11:40:756 INFO [logging.py:121] — Step 1: torchdynamo done tracing forward (RETURN_VALUE)
2128922-2023-01-18 10:11:40:757 INFO [logging.py:121] — Step 2: calling compiler function debug_wrapper
2128922-2023-01-18 10:11:40:884 INFO [logging.py:121] — Step 3: torchinductor compiling FORWARDS graph 0
2128922-2023-01-18 10:11:40:887 WARNING [graph.py:280] — Creating implicit fallback for:
target: aten._adaptive_avg_pool3d.default
args[0]: TensorBox(StorageBox(
InputBuffer(name=‘primals_1’, layout=FixedLayout(‘cpu’, torch.float32, size=[8, 2048, 7, 7], stride=[100352, 49, 7, 1]))
))
args[1]: [32, 7, 7]
2128922-2023-01-18 10:11:40:894 WARNING [ir.py:2942] — Using FallbackKernel: torch.ops.aten._adaptive_avg_pool3d.default
can someone help me to understand above the reason of fallback? is it running in eager mode?
what can be done to execute in compile mode?