So I would like to train this model (GitHub - tianqi-wang1996/DeepAccident: Code for the benchmark - DeepAccident: A Motion and Accident Prediction Benchmark for V2X Autonomous Driving.) that uses CUDA 10.2, but the GPUs I have available are either an RTX 3090, an A40, or even newer than these. The lowest CUDA version they are compatible with is CUDA 11.1. Is it possible for me to run the code for this model at all, or would I need to rewrite its code or reimplement the model?
Yes, the CUDA version is irrelevant and you would need to install a PyTorch binary with CUDA runtime dependencies for your GPU.