Hi,
I'm trying to run DeepPrep on an RTX 5090 (Blackwell architecture, sm_120) but encountered CUDA compatibility issues since the current Docker image uses PyTorch that only supports up to sm_90.
I've created a guide with workarounds to get DeepPrep running on Blackwell GPUs:
https://github.com/BoletuPeng/DeepPrep-blackwell-support
Would it be possible to:
- Update the official Docker image to support CUDA 12.8+ / PyTorch 2.7+?
- Or add documentation for RTX 50 series users?
Happy to help if needed.