torch.compile crashes with InternalTorchDynamoError when an OrderedDict is mutated concurrently
GitHub issue · pytorch/pytorch #196208
Becomes a ticket, AIOPS-21
It arrived
`torch.compile` crashes with an `InternalTorchDynamoError` when an `OrderedDict` argument is concurrently mutated while TorchDynamo is constructing its symbolic representation. The corresponding eager execution succeeds. The failure occurs with `backend="eager"`, which suggests the issue is in…
Nothing else looked like it
Written reports are grouped when their wording is close enough, first by shared words, then by meaning. It became a problem of its own, and the model read it on its own.
A model labelled it, once
Everything the model returns is below. It does not decide what happens next.
- Kind of problem
- crash
- How bad
- high
- How sure
- In one sentence
torch.compile crashes with InternalTorchDynamoError when an OrderedDict is mutated concurrently during tracing
The rulebook decided
Read top to bottom; the lit rule matched first.
- severity is criticala person looks first
- confidence at or above 0.9acts on it alone(the rule that decided this case)
- confidence at or above 0.7drafts it for approval
- severity is higha person looks first
- anything elserecords it as unsure
When it acts alone, it sends a ticket only for crashes, errors, slowdowns and feature requests. Anything else it is sure about is routine, and dropped.
Rule 2 matched: confidence 0.90 is at or above the 0.9 auto threshold.
Becomes a ticket
Sure enough, and something a team should fix. The rulebook sends it to the tracker.
AIOPS-21 Bug: [crash] torch.compile crashes with InternalTorchDynamoError when an OrderedDict is mutated concurrently during tracing
The demo tracker holds a sample of 26 tickets; the rulebook would file every match.