Fix GPT_neox incorrect output with batch query#1358
Merged
Conversation
Jianhong-Zhang
requested review from
bhargaveede,
ssarkar2 and
vivekgoe
as code owners
Collaborator
|
@dvarshney-habana @libinta can you please help with review? change is in text-generation, you may have a better idea about effect of changes here. |
libinta
reviewed
Oct 1, 2024
| if next_tokens[i] == eos_token_id: | ||
| idx_bs[i] = token_idx.item() | ||
| if token_idx > idx_bs[i]: | ||
| next_tokens[i] = pad_token_id |
Collaborator
There was a problem hiding this comment.
please do not add any logic inside the generation loop, it will impact performance. instead you can do it after generation to return right content based on eos
Contributor
Author
There was a problem hiding this comment.
moved logic to after generation loop and before return
Jianhong-Zhang
force-pushed
the
gpt_neox_output
branch
from
a8b7750 to
8c0ead3
Compare
libinta
reviewed
Oct 14, 2024
| idx_bs = idx | ||
| if idx > idx_bs: | ||
| input_ids[i][idx] = pad_token_id | ||
| idx_bs = generation_config.max_length |
Collaborator
There was a problem hiding this comment.
do we still want to return padding token in this case or just stop here?
Contributor
Author
There was a problem hiding this comment.
according to customer requirement, we need to padding token after <|endoftext|>
There was a problem hiding this comment.
https://huggingface.co/facebook/bart-large-cnn/blob/main/config.json
For bart decoder_start_token_id and eos_token_id are same.
This breaks bart generation
regisss
approved these changes
Oct 16, 2024
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
regisss
pushed a commit
that referenced
this pull request
Oct 17, 2024
ugolowic
added a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Dec 3, 2024
PR huggingface#1358 from upstream introduced a nested for loop which caused performance drop observed in T5 summarization. This commit translates the loop into tensor operations and restores performance. Co-authored-by: Adam Stachowicz <105052242+astachowiczhabana@users.noreply.github.com> Signed-off-by: Urszula Golowicz <urszula.golowicz@intel.com>
ugolowic
added a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Dec 3, 2024
PR huggingface#1358 from upstream introduced a nested for loop which caused performance drop observed in T5 summarization. This commit translates the loop into tensor operations and restores performance. Co-authored-by: Adam Stachowicz <105052242+astachowiczhabana@users.noreply.github.com> Signed-off-by: Urszula Golowicz <urszula.golowicz@intel.com>
ugolowic
added a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Dec 3, 2024
PR huggingface#1358 from upstream introduced a nested for loop which caused performance drop observed in T5 summarization. This commit translates the loop into tensor operations and restores performance. Co-authored-by: Adam Stachowicz <105052242+astachowiczhabana@users.noreply.github.com> Signed-off-by: Urszula Golowicz <urszula.golowicz@intel.com>
ugolowic
added a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Dec 3, 2024
PR huggingface#1358 from upstream introduced a nested for loop which caused performance drop observed in T5 summarization. This commit translates the loop into tensor operations and restores performance. Co-authored-by: Adam Stachowicz <105052242+astachowiczhabana@users.noreply.github.com> Signed-off-by: Urszula Golowicz <urszula.golowicz@intel.com>
ugolowic
added a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Dec 3, 2024
PR huggingface#1358 introduced a nested for loop which caused performance drop observed e.g. in T5 summarization. This commit translates the loop into tensor operations and restores performance. Co-authored-by: Adam Stachowicz <105052242+astachowiczhabana@users.noreply.github.com> Signed-off-by: Urszula Golowicz <urszula.golowicz@intel.com>
xinyu-intel
pushed a commit
to HabanaAI/optimum-habana-fork
that referenced
this pull request
Mar 4, 2025
PR huggingface#1358 from upstream introduced a nested for loop which caused performance drop observed in T5 summarization. This commit translates the loop into tensor operations and restores performance. Co-authored-by: Adam Stachowicz <105052242+astachowiczhabana@users.noreply.github.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
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.
No description provided.