Open stdio159 opened 2 months ago
Hi, I got the same error and solved it by correcting the nproc_per_node in script, basically I only have 4 gpu, should I have to change it from 8 to 4. Hope it helps.
Hello, I have encountered the same problem, have you solved it?
RuntimeErrorRuntimeError: CUDA error: invalid device ordinal Compile with
TORCH_USE_CUDA_DSA
to enable device-side assertions. : CUDA error: invalid device ordinal Compile withTORCH_USE_CUDA_DSA
to enable device-side assertions.