-
Notifications
You must be signed in to change notification settings - Fork 1.1k
[AINode] Fix bug of sundial and forecast udf #16768
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
[AINode] Fix bug of sundial and forecast udf #16768
Conversation
CRZbulabula
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
PTAL.
| elif past_length < token_num: | ||
| # TODO: Actually, we need to know the output_token_lens used in the last generation step. | ||
| # Sundial will pad the input when it is non-divisible, so we cannot use past_length to slice input_ids | ||
| input_ids = input_ids[:, -self.config.output_token_lens[0] :] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why change the computation of input_ids here?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Because Sundial will pad the input when it is non-divisible, so we cannot use past_length to slice input_ids, it will slice more sequence than we expected.
0dc23cf to
b18da36
Compare
CRZbulabula
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM!
(cherry picked from commit 2b47be7)
(cherry picked from commit 2b47be7)
* [AINode] Refactor code base * [AINode] Implement concurrent inference framework (#16311) (cherry picked from commit 7b9ec7e) * [AINode] Fix bugs for SHOW LOADED MODELS (#16410) (cherry picked from commit 40b2b33) * [AINode] Add a batcher for inference (#16411) (cherry picked from commit 7734331) * [AINode][Bug fix] Concurrent inference (#16518) * trigger CI * bug fix 4 show loaded models (cherry picked from commit b4dde12) * [AINode] Concurrent inference bug fix (#16595) (cherry picked from commit 46a0c6a) * [AINode] Adjust the maximum inference input length (#16640) (cherry picked from commit 2c9064f) * [AINode] Fix bug of sundial and forecast udf (#16768) (cherry picked from commit 2b47be7) * [AINode] Package AINode via PyInstaller (#16707) (cherry picked from commit 49c625b) * [AINode] Enable AINode start as background (-d) (#16762) (cherry picked from commit 1ebb951) * [AINode] Update AINodeClient for DataNode to borrow (#16647) (cherry picked from commit d49d7dd) * [AINode] Fix bug that AINode cannot compile in Windows (#16767) (cherry picked from commit cd443ba) * [AINode] Delete poetry.lock for easier maintain different operating systems (#16793) (cherry picked from commit 50f92e4) * [AINode] Fix cp errors --------- Co-authored-by: Leo <[email protected]> Co-authored-by: jtmer <[email protected]> Co-authored-by: Zeyu Zhang <[email protected]>
(cherry picked from commit 2b47be7)
Description
Test
Before
After