New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
AOT Autograd - LSTM - grads not generated (model tts_angular) #586
Comments
anijain2305
changed the title
AOT Autograd - LSTM - grads not generated
AOT Autograd - LSTM - grads not generated (model tts_angular)
Mar 10, 2022
18 tasks
5 tasks
This now is erroring
Last time I saw this it was because we tried to copy a Proxy |
I think the minimal repro may no longer be valid. When I run the original tts_angular it passes
|
OK it turns out inductor still triggers this
|
ezyang
added a commit
to pytorch/torchdynamo
that referenced
this issue
Aug 23, 2022
They call Tensor.set_ internally with Storage, which is no go for AOTAutograd. Inline into them so that we can graph break. Fixes pytorch/functorch#586 Test strategy: ``` ./benchmarks/torchbench.py --inductor -dcuda --no-skip -k tts_angular ``` Note that inductor is still failing, but differently, after this PR. Signed-off-by: Edward Z. Yang <ezyang@fb.com>
ezyang
added a commit
to pytorch/torchdynamo
that referenced
this issue
Aug 23, 2022
They call Tensor.set_ internally with Storage, which is no go for AOTAutograd. Inline into them so that we can graph break. Fixes pytorch/functorch#586 Test strategy: ``` ./benchmarks/torchbench.py --inductor -dcuda --no-skip -k tts_angular ``` Note that inductor is still failing, but differently, after this PR. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This is a subgraph from
tts_angular
modelThe generated backward pass has many
None
outputs, suggesting that thatrequires_grad
is somehow not passed correctly when LSTM cell is used.The text was updated successfully, but these errors were encountered: