Closed yfyeung closed 1 day ago
Can you explicitly pass rank and world_size arguments to the sampler and see if the issue persists? It attempts auto-detection but maybe it failed for some reason in your configuration (if that’s the case let’s try to find out why).
Can you explicitly pass rank and world_size arguments to the sampler and see if the issue persists? It attempts auto-detection but maybe it failed for some reason in your configuration (if that’s the case let’s try to find out why).
After doing that, it's working as expected. It seems this issue occurs only on the virtual machine node, likely due to the specific environment of the machine itself.
branch: current master environment: torch2.0.1 + Python3.8.19 + cuda11.8 recipe: egs/librispeech/zipformer commend:
log: