Chao Zhang

Results 63 comments of Chao Zhang

🐛 [Bug] Compilation causes error: `RuntimeError: [Error thrown at core/partitioning/shape_analysis.cpp:66] Expected ivalues_maps.count(input) to be true but got false Could not find torch::jit::Value* 47 produced from %47 : int = prim::dtype(%52) in lowering graph for mini graph input.`

Looks like the issue is in this line: ``` tril = torch.zeros(1, 1, 1, device=a.device, dtype=out.dtype) ``` as changing this line to: ``` tril = torch.zeros(1, 1, 1).cuda() ``` appears...

🐛 [Bug] Compilation causes error: `RuntimeError: [Error thrown at core/partitioning/shape_analysis.cpp:66] Expected ivalues_maps.count(input) to be true but got false Could not find torch::jit::Value* 47 produced from %47 : int = prim::dtype(%52) in lowering graph for mini graph input.`

You're probably facing a different but possibly related issue. Can you file a new bug report with the above information?

[`torch.Tensor.__getitem__`] Add advanced indexing (GatherND) support to `torch.Tensor.__getitem__`

Updated the algorithm to remove unnecessary gather operations... now only one gather operation is needed whereas multiple were used before. Besides being more efficient, this also side steps a problem...

[Bug] `torch.clamp` fails to convert with error `TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.`

I'll put up a fix shortly.

[Bug] `torch.clamp` fails to convert with error `TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.`

Ah I see, the bigger issue is that the `torch.clamp` implementations currently only handle single scalars instead of `torch.tensors` for min and max args (at least, that seems to be...

[Feature Request] TensorRT does not automatically promote dtypes when comparing inputs

I'll have a PR up shortly.

[Bug] Comparing a tensor with an IntWrapper results in error `TypeError: add_constant(): incompatible function arguments.`

I'll put up a fix shortly.

[Bug] `torch.squeeze` removes non-singular dimensions

I'll take a look at this issue if needed, but I think this really only comes up with dynamic shapes. In these situations, the user usually knows a priori that...

[Bug] `torch.squeeze` fails to squeeze negative dimensions.

I'll put up a fix shortly.

[Bug] Comparison between tensor and scalar fails with error `[elementWiseNode.cpp::computeOutputExtents::19] Error Code 2: Internal Error (Assertion x.nbDims == y.nbDims failed. )`

I'll put up a fix shortly.