Add page attention manager and kvcache manager #167
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
FanhaiLu1
requested review from
bhavya01,
lsy323,
qihqi,
sixiang-google and
wang2yn84
qihqi
approved these changes
Aug 6, 2024
| page_size: int, | ||
| max_pages_per_sequence: int, | ||
| ): | ||
| self.unused_pages = queue.Queue() |
There was a problem hiding this comment.
Current jet stream implementation's detokenize_threads and _generate_threads are different thread, both of them need to access this queue. So the queue should be thread safe, but deque is not thread safe.
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Outdated suggestions cannot be applied.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.
This PR adds two classes, fundamental for page attention in JetStream:
PageAttentionManager:
This class manages and frees page resources, calculates page metadata, and supports cache insertion.
PageKVCacheGenerate:
This class updates decode caches in a page-attention format. Unlike the standard LLM KV cache shape ([batch_size, num_heads, seq_len, head_dim]), PageKVCache uses the shape [num_heads, total_num_pages, page_size, head_dim].