Columbia IBM Project: Implementing Paged Attention with Flex Attention#421
Open
thomasjoshi wants to merge 65 commits into
Open
Columbia IBM Project: Implementing Paged Attention with Flex Attention#421thomasjoshi wants to merge 65 commits into
thomasjoshi wants to merge 65 commits into
Conversation
* Add `_flex_attn` attribute with lazy instantiation so we avoid recreating `FlexAttention` on every forward pass; sync dropout dynamically with training / eval mode. * Improve `_validate_paged_attention` with positive‑length check. * Extend class docstring to document `paged_attention_config`.
…ble to perform under such loads for capacity planning.
Created new sequence benchmark files
update attention test model setup
Remove backward pass benchmark
update tokenzier
create csv and plot
remove 8192 case
add matplotlib to req.txt
update plot code
remove 4096 test
update csv write
memory
Hss2173/seq bench
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Our project aims to integrate PyTorch's Paged Attention into the Foundation Model Stack (FMS) using Flex Attention. We intend to enhance memory efficiency and inference speed for long-context language models without sacrificing model accuracy. Specifically, we will implement a dynamic, paged key-value (KV) cache that minimizes memory fragmentation, benchmark its performance against standard attention mechanisms, and evaluate the impact of various paging strategies on overall model performance.