Zhexin Li

icon indicating an email [email protected]

icon company DAMO, Alibaba-inc icon location Hangzhou, China icon location Master of National Laboratory of Pattern Recognition (NLPR), Institute of Automation (IA), Chinese Academy of Sciences (CAS)

Results 4 issues of Zhexin Li

confused with an undefined function

2 comment

hello hustzxd! Your repo is very helpful for my current work. But it confuses me that in the class _Conv2dQ there is an undefined function get_default_kwargs_q(). I can't find it...

Is there a way to activate int8 MHA_v2 kernel when SeqLen > 512?

4 comment

## Description Hi, I notice form [Issue](https://github.com/NVIDIA/TensorRT/issues/3243#issuecomment-1714849183) that the int8 MHA_v2 kernel only supports SeqLen > 512. I use pytorch_quantization to insert QDQ in the MHA and convert to TRT....

triaged

[feat]: support navit

See README.md

关于使用 cutlass 实现 per-channel 量化 的 int8 conv

2 comment

大佬你好，我最近在参考你的 cutlass 写一个 int8 quantized 的 conv。在你的代码中，int8 conv 和 dequantize 是两个 kernel。我想将 DQ fuse 到 conv 中作为一个 kernel 执行以节省访存。我希望借助 DefaultConv2dFprop 中的 EpilogueOp 来实现，也就是将 alpha 设为 input_scale * per_channel_weight_scale，计算公式是 alpha *...