Description
I checked some comments on the issues, the commit for neuralsim is d0bee89, and for nr3d_lib it's b3d9627. My computer runs Ubuntu 22.04, and my graphics card is the NVIDIA RTX 4000 Ada. I saw your @ventusff reply under other comments, and I downloaded CUDA version 11.7 and torch version 2.0.0. I no longer have the issue with "If capturable=True, params and state_steps must be CUDA tensors."
However, I encountered " RuntimeError: CUDA error: device-side assert triggered. Compile with `TORCH_USE_CUDA_DSA' to enable device-side assertions. “ After making some modifications, the training still couldn't be completed, and it would throw a ”RuntimeError: CUDA error: device-side assert triggered。“ at 2% or 3%. Is it because the code version installed is too old, or what? Or has anyone else encountered this issue that we could discuss?
我看issues的一些评论，neuralsim的commit是d0bee89， nr3d_lib的commit是b3d9627。我电脑的版本是ubuntu22.04，显卡是NVIDIA RTX 4000 Ada，看到了您@ventusff在其他评论下的回复，我下载的cuda版本为11.7，torch版本为2.0.0，没有了If capturable=True, params and state_steps must be CUDA tensors.的问题。遇到了RuntimeError: CUDA error: device-side assert triggered。Compile with TORCH_USE_CUDA_DSA to enable device-side assertions. 的问题，做了一些修改，但是训练一直不能完成，到2%或者3%就会报RuntimeError: CUDA error: device-side assert triggered 的错误。是代码版本安装的比较老吗还是什么？ 或者谁有遇到过这个问题可以探讨一下。
