[core][compiled graphs] Support arbitrary torch.dtypes when passing through shared memory #48957
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
Signed-off-by: Stephanie Wang <[email protected]>
Signed-off-by: Stephanie Wang <[email protected]>
kevin85421
reviewed
Nov 27, 2024
| # Numpy does not have an equivalent dtype for all torch dtypes, so | ||
| # instead of casting directly to numpy, we first use a view with a | ||
| # common dtype and then view as numpy array. | ||
| return (tensor.view(torch.uint8).numpy(), tensor.dtype) |
There was a problem hiding this comment.
what will happen if tensor contains a value greater than 255?
There was a problem hiding this comment.
I think it will be broken down into multiple unit8 -- the underlying values don't change
ruisearch42
approved these changes
Nov 27, 2024
kevin85421
approved these changes
Nov 27, 2024
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Outdated suggestions cannot be applied.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.
Why are these changes needed?
Some torch.dtypes don't have a numpy equivalent. We use numpy to store the tensor data zero-copy in the object store. To support these tensors, we first view the array with a common dtype (uint8), and then view as a np array. During deserialization, we use another view back to the original dtype.
Closes #48141.