add fused linear-loss function in Domino #965
Open
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
|
Hi @duanhx1037 , thx for this pr. Please solve above:
|
Signed-off-by: dhx <[email protected]>
Signed-off-by: dhx <[email protected]>
duanhx1037
force-pushed
the
liger_integration
branch
from
a297a52 to
f4eefa1
Compare
Signed-off-by: dhx <[email protected]>
duanhx1037
force-pushed
the
liger_integration
branch
from
06914b0 to
8a02385
Compare
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Outdated suggestions cannot be applied.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.
Changes
Add a fused and chunked function for linear and cross-entropy loss computation in Domino, based on [Liger-Kernel](https://github.com/linkedin/Liger-Kernel).
Effect on memory usage
Reduce training memory usage, especially peak memory usage in the vocabulary layer. Using a setup of
num-layers=4, seq-length=512, batch-size=8intraining/DeepSpeed-Domino/pretrain_gpt3_2.7b.sh, the average memory usage (GB) measured bytorch.cuda.max_memory_allocated()in each training iteration will drop from 6.158 to 5.0458.Effect on loss
Almost identical loss curve in a 1000-iteration experiment.
