Nick Comly comments

Results 68 comments of


                                            Nick Comly

trafficstars

🐛 [Bug] FX intro example broken

CC: @narendasan @andi4191 Do we have test cases in DLFW for FX?

🐛 [Bug] Conversion error when using torch-TRT to run the bert model after qat quantization

Hi @lixiaolx is the model trained using the [PyTorch QAT toolkit](https://github.com/NVIDIA/TensorRT/tree/main/tools/pytorch-quantization)?

✨[Feature] torchtrtc should have the way to accept compile spec from command line

@borisfom is this issue still present with the latest APIs? Do you have a preference for CLI vs API?

❓ [Question] Speed problem about TRTorch and Torch-TensorRT - Device Compatibility Check

Between these two versions there was a constant time operation that was added to check compatibility of the current device with the compiled model. This is likely the overhead you...

🐛 [Bug] Encountered bug when using Torch-TensorRT (We don't have an op for aten::floor_divide but it isn't a special case)

@narendasan seems a lowering pass could be a good WAR here.

❓ [Question] How do you override or remove evaluators

@peri044 bump. @narendasan we should bring up with TRT team in our sync Friday

🐛 [Bug] Unsupported operator: aten::lstm.input, aten::Int.Tensor

The outstanding layers have been added as feature requests. Adding the bug tag since `require_full_compilation = False` so partial compilation should work.

🐛 [Bug] Encountered bug when using Torch-TensorRT for INT8

@peri044 for another PTQ bug. P1

✨[Feature] Will torch-TensorRT plan to support runtime subgraph optimization like TFTRT?

This is a constraint on data dependent shapes (DDS), currently slated for v1.4 end of year.

❓ [Question] No improvement when I use sparse-weights?

Hi @wzywzywzy this is because of TensorRT's kernel autotuning. TRT selects the fastest kernels for your model, regardless of sparsity, so in this case the dense kernels may be faster...