deepspeed-chat: fix bf16 stage2 accuracy for bloom-560m#772
Merged
tjruwase merged 1 commit intodeepspeedai:masterfrom Oct 17, 2023
Merged
deepspeed-chat: fix bf16 stage2 accuracy for bloom-560m#772tjruwase merged 1 commit intodeepspeedai:masterfrom
tjruwase merged 1 commit intodeepspeedai:masterfrom
Conversation
Bloom-560m model has high variance in its last LN layer weight. This causes accuracy issues in bf16 stage2 training. Therefore, reset the parameters of the last LN layer before training. This is a good practice in any case where we replace the classifier that follows the LN. In addition, in case we are using only optimize lora, we need to force the training of the LN parameters that were reset. Note that current fix uses plain initialization of final LN. A separate commit will provide support for zero3 initialization. Change-Id: I323d8947907eb4a1cc0fa6354bdaf0cbbf33a68d Signed-off-by: Moshe Island <misland@habana.ai>
mosheisland
requested review from
RezaYazdaniAminabadi,
ShadenSmith,
arashb,
awan-10,
conglongli,
duli2012,
eltonzheng,
jeffra,
minjiaz,
mrwyattii,
samyam,
tjruwase,
xiaoxiawu-microsoft and
yaozhewei
as code owners
tjruwase
requested review from
lekurile
and removed request for
RezaYazdaniAminabadi,
ShadenSmith,
arashb,
conglongli,
duli2012,
eltonzheng,
jeffra,
minjiaz,
mrwyattii,
samyam,
tjruwase,
xiaoxiawu-microsoft and
yaozhewei
hwchen2017
pushed a commit
that referenced
this pull request
Jun 8, 2025
Bloom-560m model has high variance in its last LN layer weight. This causes accuracy issues in bf16 stage2 training. Therefore, reset the parameters of the last LN layer before training. This is a good practice in any case where we replace the classifier that follows the LN. In addition, in case we are using only optimize lora, we need to force the training of the LN parameters that were reset. Note that current fix uses plain initialization of final LN. A separate commit will provide support for zero3 initialization. Change-Id: I323d8947907eb4a1cc0fa6354bdaf0cbbf33a68d Signed-off-by: Moshe Island <misland@habana.ai> Co-authored-by: Moshe Island <misland@habana.ai>
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
3 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.
Bloom-560m model has high variance in its last LN layer weight. This causes accuracy issues in bf16 stage2 training. Therefore, reset the parameters of the last LN layer before training. This is a good practice in any case where we replace the classifier that follows the LN.
In addition, in case we are using only optimize lora, we need to force the training of the LN parameters that were reset.
Note that current fix uses plain initialization of final LN. A separate commit will provide support for zero3 initialization.
Change-Id: I323d8947907eb4a1cc0fa6354bdaf0cbbf33a68d