TensorRT
TensorRT copied to clipboard
pytorch
PyTorch/TorchScript/FX compiler for NVIDIA GPUs using TensorRT
Fixes #2434 This PR would be dependant on #2664 and #2669. Major changes 1. 2664- Implementation makes use of `aten::scatter.src` 2. 2669- Constants getting converted to fake tensors in `get_attr`...
Covers the converters- aten.rand, aten.randn, aten.randperm
Fixes #2436 This PR would be dependant on https://github.com/pytorch/TensorRT/pull/2519, https://github.com/pytorch/TensorRT/pull/2664 and https://github.com/pytorch/TensorRT/pull/2669. Major changes 2519- Decomposition of aten::slice_scatter 2664- Implementation makes use of aten::scatter.src 2669- Constants getting converted to fake...
## Bug Description Cannot compile the [SwinIR model](https://github.com/JingyunLiang/SwinIR). Error message: ``` Traceback (most recent call last): File "main.py", line 61, in compile_tensorrt_model(torch.float) File "main.py", line 56, in compile_tensorrt_model compiled_model =...
# Description Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change....
# Description Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change....
# Description This PR enables the conv1d -> conv2d converter mapping work when the weights are Tensor. PS: Kernel tensor is also unsqueezed when the filterDims are unsqueezed for conv1d...
## ❓ Question torch_tensorrt.dynamo.compile hangs indefinitely mid compilation cpu usage is through the roof and having debug = True shows that there's a step where it fails ## What you...
## Bug Description got the export error with torch_tensorrt 2.2.0 use dynamo IR ```bash torch._export.verifier.SpecViolationError: Node.meta max_pool2d_default is missing val field. ``` ## To Reproduce ```python optimized_model = torch_tensorrt.compile( self.model,...
