viable/strict/1790620535: [c10d] Allow CPU device_id in init_process_group (#198340)
- PyTorch: 1593 events in the last 90 days
- PyTorch: 1574th Release in the last 90 days
- Previous: earlier the same day · trunk/346528339b76a711445d026b607d38efd90e0452
What happened
_new_process_group_helper rejected any CPU device_id, even one with a valid index (e.g. cpu:0), raising ValueError: init_process_group device_id parameter must be an accelerator with an index. This made it impossible to use device_id with the GLOO backend on CPU, despite init_process_group itself already treating CPU device_id as a valid case (it skips the accelerator-matching checks specifically for CPU, at line ~2…
Summary assembled by rule from the sources below