Skip to content

perf: bound KV page reclamation candidate scans #578

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

Open
T4t4KAU wants to merge 2 commits into
base: InfiniLM-v0.2.9c
Choose a base branch
from
Open

perf: bound KV page reclamation candidate scans #578

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c518204
perf: bound KV page reclamation candidate scans
T4t4KAU Sep 17, 2026
390b975
chore: remove added unit tests
T4t4KAU Sep 17, 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
14 changes: 9 additions & 5 deletions python/infinilm/llm/cache_manager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -336,11 +336,15 @@ def free_blocks(self, block_table: Sequence[int]) -> None:

def try_free_blocks(self, num_required: int) -> bool:
"""Evict unreferenced blocks until the requested capacity is available."""
to_free = [
block_id
for block_id in self.used_block_ids
if self.blocks[block_id].ref_count == 0
]
# Preserve one eviction, if possible, even when capacity is already enough.
num_to_free = max(1, num_required - len(self.free_block_ids))
to_free = []
for block_id in self.used_block_ids:
if self.blocks[block_id].ref_count == 0:
to_free.append(block_id)
if len(to_free) >= num_to_free:
break
# Collect candidates before mutating the set being traversed.
for block_id in to_free:
self._deallocate_block(block_id)
if self.can_allocate(num_required):
Expand Down