TopK 3840 limitation and future plans for this operator

NVIDIA / TensorRT

NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.

https://developer.nvidia.com/tensorrt

Apache License 2.0

10.86k stars 2.14k forks source link

TopK 3840 limitation and future plans for this operator #4244

Open amadeuszsz opened 2 weeks ago

amadeuszsz commented 2 weeks ago

Are there any plans for removing 3840 limitation from TopK operator? Some time ago I reached a post written by @zerollzeng

It's known limitation, and we are actively working on remove it.

I wonder if some work on this topic is still ongoing. If so, which TensorRT release it targets?

poweiw commented 1 week ago

@zerollzeng can you share some inputs? Do we have an internal bug tracking this?