-
Notifications
You must be signed in to change notification settings - Fork 4.1k
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
state_dict_factory: llama checkpoint - support SWIGLU #5601
Open
nelyahu
wants to merge
3
commits into
microsoft:master
Choose a base branch
from
nelyahu:fix_llama_3d_checkpoint_load
base: master
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
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
DeepSpeed supports loading a checkpoint for inference with different DP/TP/PP. This requires to split/merge parameters based on their TP attributes. Currently, this is done by using model specific parameter names. This is not a good practice and should be modified. This commit handles the required changes to support MDS LLaMA model. There are 2 changes: - Support for lm_head.weight - Support for mlp.h_to_4h.weight for SWIGLU SWIGLU requires different handling, however there is no meta data available that identifies mlp.h_to_4h.weight as SWIGLU. Therefore, for now we use a hack to detect it.
tjruwase
requested review from
samadejacobs and
tohtana
and removed request for
mrwyattii
June 21, 2024 22:27
tjruwase
reviewed
Jun 21, 2024
new_client_sd[key] = torch.cat(value_list, axis=0) | ||
elif "mlp.dense_h_to_4h.weight" in key: |
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.
@nelyahu, this logic is very old and hacky. Could it not be replaced with Universal Checkpointing?
tohtana
reviewed
Jun 21, 2024
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.
@nelyahu Do we have any update? |
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.
DeepSpeed supports loading a checkpoint for inference with different DP/TP/PP. This requires to split/merge parameters based on their TP attributes. Currently, this is done by using model specific parameter names. This is not a good practice and should be modified.
This commit handles the required changes to support MDS LLaMA model. There are 2 changes:
SWIGLU requires different handling, however there is no meta data available that identifies mlp.h_to_4h.weight as SWIGLU. Therefore, for now we use a hack to detect it.