Apurba Bose

Hi @gs-olive In the test case below: ``` class TestGatherConverter(DispatchTestCase): def test_gather_zero_two_dim(self): class TestModule(nn.Module): def __init__(self): super().__init__() def forward(self, x, indices): # index0 = torch.randint(0, 1, (1, 1)) out =...

I removed the `gather` test here, since the above task was to expose the gather layer. The TensorRT output for `torch.ops.aten.gather` and ` ctx.net.add_gather` is different. Example ``` class gather(torch.nn.Module):...

Had a doubt on this one. Does this require a test. In the following test: ``` def test_lowering_empty_like(self): class emptyLike(torch.nn.Module): def __init__(self, *args, **kwargs) -> None: super().__init__(*args, **kwargs) def forward(self,...

I do not think that the graph would be empty since it would reduce to the lowering operations of `aten::size` and `torch.Tensor()` of the corresponding size getting created. So the...

I verified the above test case with three cases- 1. Case 1: ``` class emptyLike(torch.nn.Module): def __init__(self, *args, **kwargs) -> None: super().__init__(*args, **kwargs) def forward(self, x): y = torch.ops.aten.empty_like.default(x) return...

Ok I will go ahead and make a separate PR for `empty_permute`. For now this PR can be merged then?

Monitoring the CI to see if this error comes in the test- ``` torch._dynamo.exc.BackendCompilerFailed: backend='functools.partial(, 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...

Thanks @gs-olive for pointing the above. But I think the implementation using `slice_scatter` decomposition should also work in our case. For eg: in the above case the unsqueeze dimension with...

I misread the case pointed by you. ``` >>> import torch >>> a = torch.zeros(2, 2) >>> b = torch.ones(2) >>> torch.select_scatter(a, b, 0, 0) tensor([[1., 1.], [0., 0.]]) >>>...

The issue is that you need to convert the model to fp16 after loading it. Specifying the enabled_precisions= {torch.half} would not be enough since it would just direct TensorRT to...