Skip to content

[executorch] Take Tensor for start_pos in llama custom SDPA - #3580

Closed
swolchok wants to merge 1 commit into
gh/swolchok/4/basefrom
gh/swolchok/4/head
Closed

[executorch] Take Tensor for start_pos in llama custom SDPA#3580
swolchok wants to merge 1 commit into
gh/swolchok/4/basefrom
gh/swolchok/4/head

Conversation

@swolchok

@swolchokswolchok commented May 10, 2024

Copy link
Copy Markdown
Contributor

Stack from ghstack (oldest at bottom):

Several different runtimes don't seem to like the custom SDPA taking a single scalar pulled from a Tensor as an argument:

  • dynamo doesn't support item()
  • CoreML doesn't seem to like this either

So, just take a Tensor.

Differential Revision: D57233278

Several different runtimes don't seem to like the custom SDPA taking a single scalar pulled from a Tensor as an argument:
- dynamo doesn't support item()
- CoreML doesn't seem to like this either
So, just take a Tensor.
Differential Revision: [D57233278](https://our.internmc.facebook.com/intern/diff/D57233278/)
[ghstack-poisoned]
@pytorch-bot

pytorch-botBot commented May 10, 2024

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/3580

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 9ea21d0 with merge base dd81fc7 (image):
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@facebook-github-botfacebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label May 10, 2024
swolchok added a commit that referenced this pull request May 10, 2024
Several different runtimes don't seem to like the custom SDPA taking a single scalar pulled from a Tensor as an argument:
- dynamo doesn't support item()
- CoreML doesn't seem to like this either
So, just take a Tensor.
Differential Revision: [D57233278](https://our.internmc.facebook.com/intern/diff/D57233278/)
ghstack-source-id: 225980366
Pull Request resolved: #3580
@facebook-github-bot

Copy link
Copy Markdown
Contributor

This pull request was exported from Phabricator. Differential Revision: D57233278

@swolchok

Copy link
Copy Markdown
ContributorAuthor

I found out we shouldn't be using the custom SDPA op for CoreML, so we don't have "several" runtimes that want this. Closing.

Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA SignedThis label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.fb-exported

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants

@swolchok@facebook-github-bot@JacobSzwejbka