Skip to content

Fixing edge case #2249

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

Merged
merged 1 commit into from
Aug 2, 2025
Merged

Fixing edge case #2249

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
14 changes: 9 additions & 5 deletions lib/bindings/python/rust/llm/block_manager/vllm/connector/leader/slot.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -510,11 +510,13 @@ impl Slot for VllmConnectorSlot {
.sequence()
.blocks()
.iter()
.skip(num_computed_blocks)
.map(|b| b.sequence_hash())
.collect::<Vec<_>>();

tracing::debug!("matching against {} block hashes", sequence_hashes.len());
// we start matching non-device blocks after the device blocks
let search_offset = num_computed_blocks;

tracing::debug!("matching against {} block hashes", sequence_hashes[search_offset..].len());

// we should do this opportunistically after this operation is done
// ideally it was triggered by the match_sequence_hashes_blocking calls directly
Expand All @@ -527,9 +529,6 @@ impl Slot for VllmConnectorSlot {
// disk.touch_blocks_blocking(&sequence_hashes)?;
// }

// we start matching non-device blocks after the device blocks
let search_offset = num_computed_blocks;

let mut host_blocks = self
.block_manager
.host()
Expand Down Expand Up @@ -568,6 +567,11 @@ impl Slot for VllmConnectorSlot {
return Ok(());
}

// early exit if we need to onboard 0 blocks
if (num_computed_blocks + num_matched_blocks) * block_size == self.sequence().total_tokens() {
return Ok(());
}

let mut num_new_matched_tokens = num_matched_blocks * block_size;

// we are on a block boundary, so we need to throw away the last block
Expand Down
Loading