Update deepseek modeling file for PatchedVLLMKVCache #1009
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
Signed-off-by: Mengni Wang <mengni.wang@intel.com>
mengniwang95
requested review from
afierka-intel,
kzawora-intel,
madamczyk-intel,
mgawarkiewicz,
michalkuligowski and
vivekgoe
as code owners
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Outdated suggestions cannot be applied.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.
Previously, when we use INC to convert deepseek FP8 model, we need this commit to remove extra converts in KVCache but actually GC can remove them during graph optimization theoretically.
Furthermore, the change in commit is not aligned with the design of INC patched module, which wants to keep the returned tensor BF16 because we can't make sure users' next operation.
So, I update the modeling file to make GC can work for patched KVCache pattern of deepseek model.
Since next release is very close and GC currently can not work as expection during decode stage, it is still a workround. We will root cause and fix it from source in next relase.
This PR should work together with this PR: intel/neural-compressor#2165