Skip to content

Thought Exercise - Chained Semantic Prefix Cache Matching #282

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
sempervictus wants to merge 4 commits into from
Closed

Thought Exercise - Chained Semantic Prefix Cache Matching #282

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7bfed40
Chained Semantic Prefix Cache Matching
Mar 30, 2026
118cbb4
Waterfall match attempts for each block
Mar 31, 2026
c1be551
Use last hash for seeding
Mar 31, 2026
6308017
Store parent sem hash in PrefixCache
Mar 31, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 6 additions & 2 deletions src/core/block_manager.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,7 +155,10 @@ impl BlockManager {
if self.prefix_cache.is_some() {
let mut prefix_cache = self.prefix_cache.take().unwrap();
let seed = seq.images.as_ref().map(Self::image_prefix_seed);
let prefix_match = prefix_cache.match_prefix_with_seed(&seq.token_ids, seed);

// Use relaxed matching with 5% tolerance for tokenization differences
let prefix_match = prefix_cache.match_prefix_relaxed(&seq.token_ids, seed, 0.05);

let matched_blocks = self.resolve_mamba_matched_blocks(
&prefix_cache,
seq.id,
Expand Down Expand Up @@ -325,7 +328,8 @@ impl BlockManager {

if prefix_cache.enabled() {
let seed = seq.images.as_ref().map(Self::image_prefix_seed);
let prefix_match = prefix_cache.match_prefix_with_seed(tokens, seed);
// Use relaxed matching to handle tokenization variations
let prefix_match = prefix_cache.match_prefix_relaxed(tokens, seed, 0.05);
last_hash = prefix_match.last_hash;
raw_matched_blocks =
self.adjusted_matched_blocks(tokens.len(), prefix_match.matched_blocks);
Expand Down
Loading