Add reset(to:), processedTokenCount, and implicit prefix caching#51
Merged
Conversation
stikves
force-pushed
the
sukru/engine-partial-reset
branch
2 times, most recently
from
c3f7d0a to
a5c51ce
Compare
stikves
requested review from
alejandro-isaza,
carinapeng,
kevchengcodes and
tjia1818
carinapeng
reviewed
Jun 17, 2026
stikves
force-pushed
the
sukru/engine-partial-reset
branch
2 times, most recently
from
433d83b to
5482d42
Compare
stikves
force-pushed
the
sukru/engine-partial-reset
branch
from
8623ada to
8ddd9ba
Compare
Extends the InferenceEngine protocol with partial KV cache reset and automatic prefix detection: - reset(to: tokenIndex): truncate KV cache to keep first N positions - processedTokenCount: read-only, tracks engine's current cache position - TokenHistory: shared utility for memcmp-based prefix detection - Pipelined engine: implicit prefix caching — engine auto-detects common prefix between consecutive generate() calls, only processes new tokens - Sequential/Static: implicit rewind when input is shorter than cache (full divergence triggers zero-fill, prefix extension rewinds counter) All engines pass 20/20 explicit reset parity test. Pipelined passes 11/11 implicit prefix caching test (extend + diverge). Sequential/Static pass 10/11 (divergence auto-reset works, extend works via counter rewind).
stikves
force-pushed
the
sukru/engine-partial-reset
branch
from
8ddd9ba to
9bc7aa3
Compare
alejandro-isaza
approved these changes
Jun 18, 2026
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Extends the InferenceEngine protocol with partial KV cache reset and automatic prefix detection:
All engines pass explicit reset parity test. Pipelined passes implicit prefix caching test (extend + diverge). Sequential/Static pass by wiring explicit reset.