add llama2 autoTP support in replace_module #4022
Merged
Conversation
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
dc3671
requested review from
RezaYazdaniAminabadi,
arashb,
awan-10,
cmikeh2,
jeffra and
mrwyattii
as code owners
mrwyattii
approved these changes
Jul 24, 2023
blzheng
pushed a commit
to blzheng/DeepSpeedSYCLSupport
that referenced
this pull request
Aug 15, 2023
delock
pushed a commit
to delock/DeepSpeedSYCLSupport
that referenced
this pull request
Aug 16, 2023
* add lm_head tensor parallel * fix conflict * add embed_out tp * add llama2 autoTP support in replace_module (deepspeedai#4022) --------- Co-authored-by: Lai, Yejing <yejing.lai@intel.com> Co-authored-by: Dino Chen <zhenhuan.chen@intel.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.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.
Problem
After llama2 update in transformers [Llama2] Add support for Llama 2, the autoTP for llama is broken.
Reason
After dig into transformers modification, I noticed that there is another
self.num_key_value_headsattribute in the attention module:https://github.com/huggingface/transformers/pull/24891/files#diff-06392bad3b9e97be9ade60d4ac46f73b6809388f4d507c2ba1384ab872711c51R253
Solution
In DeepSpeed, this part is controlled by a hardcoded name list in
update_mp_paramsof autoTP'sreplace_wo_policy(see this PR's modification). So I added one more key in this list, and the problem is fixed.@tjruwase @jeffra please kindly review, thanks~