Skip to main content
Workers disagreed on success or exception type. A subclass of RuntimeError raised by DistributedRunner.
Raised when some ranks fail and others succeed, or all ranks fail with different exception types. The message identifies the first failing rank. Its exception is available as __cause__. The runner returns no partial result and refuses further calls. Call shutdown() and construct a new runner. If a peer stays blocked in a collective, the call can instead reach its timeout.