Skip to content
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

deepspeed-chat: support explicit configuration of dropout #746

Merged
merged 2 commits into from
Oct 3, 2023

Conversation

mosheisland
Copy link
Contributor

Currently, only disable_dropout configuration is supported. However, some models (e.g. Bloom) have a default of dropout=0 in model config. Therefore, modify to support explicit dropout configuration. Also, update accordingly existing training scripts.

Change-Id: I5ee96a77ca2b58d9787573a48009e2af36a270b0

@mosheisland
Copy link
Contributor Author

@microsoft-github-policy-service agree company="Intel"

@mosheisland
Copy link
Contributor Author

The formatting error is not due to this commit.
"applications/DeepSpeed-Chat/training/utils/ds_utils.py:6:1: F401 'torch' imported but unused"

@tjruwase
Copy link
Contributor

tjruwase commented Oct 3, 2023

The formatting error is not due to this commit. "applications/DeepSpeed-Chat/training/utils/ds_utils.py:6:1: F401 'torch' imported but unused"

Please check again. I think there is a different error now.

Currently, only disable_dropout configuration is supported.
However, some models (e.g. Bloom) have a default of dropout=0 in model config.
Therefore, modify to support explicit dropout configuration.
Also, update accordingly existing training scripts.

Change-Id: I5ee96a77ca2b58d9787573a48009e2af36a270b0
Signed-off-by: Moshe Island <misland@habana.ai>
@tjruwase tjruwase merged commit 4bf1924 into microsoft:master Oct 3, 2023
2 checks passed
@mosheisland mosheisland deleted the 3_dropout branch October 4, 2023 07:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants