TensorRT
TensorRT copied to clipboard
slice_scatter decomposition
Fixes #2434 This PR would be dependant on #2664 and #2669. Major changes
- 2664- Implementation makes use of
aten::scatter.src - 2669- Constants getting converted to fake tensors in
get_attrcall due to which different device locationmetaandcpuin torch
Monitoring the CI to see if this error comes in the test-
torch._dynamo.exc.BackendCompilerFailed: backend='functools.partial(<function fx_dynamo_testing_backend at 0x7f19514f7af0>, store_intermediate_graphs=[], min_
block_size=1, torch_executed_ops=set(), use_fast_partitioner=True)' raised:
RuntimeError: Attempted to set the storage of a tensor on device "meta" to a storage on different device "cpu". This is no longer allowed; the devices must match.