Jason Huang

Sure! I shall do it later. By the way, I am now doing a computer graphic course project and I choose to use SPH fluid simulation with taichi. I use...

The project presentation is scheduled on next Tuesday, Beijing time. By that moment, I will have some good demo. I am exciting to share them will you later! On Thu,...

check our demo here! https://cloud.tsinghua.edu.cn/f/b546376dead84369ab87/ On Fri, Dec 29, 2023 at 8:01 AM Mingrui Zhang ***@***.***> wrote: > Thanks! The rigid body module in this repo is indeed not well...

the code is not open! check https://github.com/jason-huang03/SPH_Project On Tue, Jan 2, 2024 at 4:57 PM Jason Huang ***@***.***> wrote: > check our demo here! https://cloud.tsinghua.edu.cn/f/b546376dead84369ab87/ > > On Fri, Dec...

Is MLA supported now? If it is supported, could you point out how to use it?

Hi, you can refer to our work [SageAttention2](https://arxiv.org/abs/2411.10958), which is the first paper (even earlier than DeepSeek V3) revealing the accumulator issue of Hopper/Ada fp8 tensor core. Actually we find...

I checkout to v0.1.5 and rebuild using `pip install --no-cache-dir --force-reinstall -e .` . However, the problem persists. The whole error message is ``` CUDA Error: an illegal memory access...

Also I find that device_id in function `SinglePrefillWithKVCacheDispatched` in `python/include/flashinfer/attention/prefill.cuh` seems to be 0 regardless of the device_id set in the python code.

I use std::cout, device.index() here is empty, but device is correct (like cuda:1). I am now trying to use cuda 12.4 and torch 2.4 to see whether the problem can...

After using pytorch 2.4 and cuda 12.4, the error disappears. Thanks for your time. It seems that the device and device index api has undergone some changes in the cuda...