Skip to content

Optionally allow models to keep the last cache item at prefill time. #653

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
copybara-service merged 1 commit into from
May 22, 2026
Merged

Optionally allow models to keep the last cache item at prefill time. #653

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
9 changes: 6 additions & 3 deletions gemma/gm/text/_prefill.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,9 +217,12 @@ def prefill(
# A cleaner implementation could be to have a per-batch cache index, to
# remove padding. But I leave this to my future self (or to future Gemini).

new_used_cache_length = (
prev_turns.used_cache_length + input.length_with_mm - 1
)
if hasattr(model, 'keep_last_prefill_kv') and model.keep_last_prefill_kv:
new_used_cache_length = prev_turns.used_cache_length + input.length_with_mm
else:
new_used_cache_length = (
prev_turns.used_cache_length + input.length_with_mm - 1
)
cache = cache.set_end_index(new_used_cache_length)

# TODO(epot): The first token was predicted, so could use this, but would
Expand Down
Loading