Filter tensor arguments from traced model. #5689
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information.
|
Thanks, can you add a test cast that would fail without this change? |
JackCaoG
reviewed
Oct 11, 2023
ysiraichi
changed the title
ysiraichi
commented
Oct 11, 2023
JackCaoG
reviewed
Oct 11, 2023
JackCaoG
reviewed
Oct 11, 2023
wonjoolee95
approved these changes
Oct 12, 2023
ysiraichi
commented
Oct 12, 2023
test/dynamo/test_bridge.py
Outdated
| module = Emb() | ||
| module.to(device) | ||
|
|
||
| @torch.compile(backend="openxla") |
There was a problem hiding this comment.
Actually, openxla doesn't have this problem. This only happens with openxla_eval. Since we are trying to get rid of openxla_eval, are we still interested in merging this PR?
There was a problem hiding this comment.
I think it is aot-autograd handles the argument. What's the non-tensor input passed to openxla_eval?
There was a problem hiding this comment.
Yep, exactly. One of the inputs passed was a custom class that inherited from nn.Embedding.
There was a problem hiding this comment.
Let me check with Brian whether it is possible for us to get non-tensor input after aot-autograd
There was a problem hiding this comment.
actually @bdhirsh any thought?
|
I think it is safer to have this change in, I am going to merge it. |
zpcore
pushed a commit
that referenced
this pull request
Oct 19, 2023
* Filter tensor arguments from traced model. This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information. * Add test. * Fix lint issues. * Simplified test. * Use `openxla` instead of `openxla_eval` backend. * Rename variables for readability. * Use `openxla_eval` instead of `openxla`.
ghpvnist
pushed a commit
to ghpvnist/xla
that referenced
this pull request
Oct 31, 2023
* Filter tensor arguments from traced model. This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information. * Add test. * Fix lint issues. * Simplified test. * Use `openxla` instead of `openxla_eval` backend. * Rename variables for readability. * Use `openxla_eval` instead of `openxla`.
mbzomowski
pushed a commit
to mbzomowski-test-org/xla
that referenced
this pull request
Nov 16, 2023
* Filter tensor arguments from traced model. This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information. * Add test. * Fix lint issues. * Simplified test. * Use `openxla` instead of `openxla_eval` backend. * Rename variables for readability. * Use `openxla_eval` instead of `openxla`.
chunnienc
pushed a commit
to chunnienc/xla
that referenced
this pull request
Dec 14, 2023
* Filter tensor arguments from traced model. This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information. * Add test. * Fix lint issues. * Simplified test. * Use `openxla` instead of `openxla_eval` backend. * Rename variables for readability. * Use `openxla_eval` instead of `openxla`.
golechwierowicz
pushed a commit
that referenced
this pull request
Jan 12, 2024
* Filter tensor arguments from traced model. This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information. * Add test. * Fix lint issues. * Simplified test. * Use `openxla` instead of `openxla_eval` backend. * Rename variables for readability. * Use `openxla_eval` instead of `openxla`.
bhavya01
pushed a commit
that referenced
this pull request
Apr 22, 2024
* Filter tensor arguments from traced model. This PR filters tensor arguments from the list of arguments that would be given to the model. **Problem:** dynamo bridge assumed all arguments were tensors. **Solution:** filter tensor arguments so that we correctly collect tensor information. * Add test. * Fix lint issues. * Simplified test. * Use `openxla` instead of `openxla_eval` backend. * Rename variables for readability. * Use `openxla_eval` instead of `openxla`.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR filters tensor arguments, when collecting tensor information, from the list of arguments that would be given to the model.
Problem: dynamo bridge assumed all arguments were tensors.
Solution: filter tensor arguments so that we correctly collect tensor information.