triton-inference-server / server

The Triton Inference Server provides an optimized cloud and edge inferencing solution.
https://docs.nvidia.com/deeplearning/triton-inference-server/user-guide/docs/index.html
BSD 3-Clause "New" or "Revised" License
8.35k stars 1.49k forks source link

Setting tensor name in python bls #6797

Open SunXuan90 opened 10 months ago

SunXuan90 commented 10 months ago

This is really for Python backend, but I can't find issue page there. I noticed that in source code of python backend, PbTensor class has a method SetName, but not implemented. This method is pretty useful for bls. Sometimes we need to pass some tensors from one model's output to another one's input, but the tensor name is different. Currently we can only do so using as_numpy or to_dlpack to get the content out and create a new tensor. It would be really nice that the method gets implemented.

nv-kmcgill53 commented 10 months ago

CC @Tabrizian

dyastremsky commented 9 months ago

Pinging @Tabrizian

XiaoxueWang1 commented 3 months ago

Hi, Has this issue been resolved?