[KV Cache] Overwrite Cache - SW Attention #297
Closed
Conversation
This file contains 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
5 tasks
davidpissarra
changed the title
tqchen
reviewed
Oct 21, 2023
src/runtime/relax_vm/lm_support.cc
Outdated
| @@ -159,6 +232,7 @@ class AttentionKVCache : public ObjectRef { | |||
| n->Append(init_data); | |||
| if (init_fill_count >= 0) { | |||
| n->fill_count = init_fill_count; | |||
| n->current_pos = init_fill_count; // sliding window attention only | |||
There was a problem hiding this comment.
window_attention_current_pos
src/runtime/relax_vm/lm_support.cc
Outdated
| * \brief Append value to the cache. | ||
| * \param value The value to overwrite previous elements. | ||
| */ | ||
| void Overwrite(NDArray value, int64_t max_cache_size) { |
|
Thanks @davidpissarra . The main comment is that given the impl is specialized to window, let us make sure tha API naming highlights the fact |
|
please also send the PR to unity branch of https://github.com/apache/tvm |
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Outdated suggestions cannot be applied.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.
Part of the effort on Sliding Window Attention (SWA) mlc-ai/mlc-llm#1003. Overwriting the cache is useful when computing SWA, so we can have a more efficient cache only containing the current window keys and values. Once the cache is full we start overwriting the older entries.