Use head_dim in config if exists #2622
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
simon-mo
approved these changes
Jan 27, 2024
NikolaBorisov
pushed a commit
to deepinfra/vllm
that referenced
this pull request
Jan 31, 2024
hongxiayang
pushed a commit
to hongxiayang/vllm
that referenced
this pull request
Feb 13, 2024
alexm-redhat
pushed a commit
to neuralmagic/nm-vllm
that referenced
this pull request
Feb 13, 2024
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Outdated suggestions cannot be applied.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.
The change is for some models which set
head_dimexplicitly instead of usinghidden_size // num_attention_heads.The wrong
head_dimread leads to the wrong size kv_cache created, which makes the PagedAttention CUDA kernel writes the new KV to unexpected GPU memory. During our case, with thehead_dimsetting wrong, the PagedAttention kernel writes values into model weights, which causes huge confusion and higher debugging cost (because the model can run without any CUDA memory error raised, and the model weights are modified unexpectedly during forward).