b9551

ggml-org/llama.cppb9551Jun 7, 2026by github-actions[bot]same day as b9550

What changed since b9550

low risk

b9551 is a single-commit KV cache optimization that avoids unnecessary cell copies, with no functional changes beyond b9550.

Improvements

  • KV cache now avoids redundant cell copies, reducing memory traffic and per-token overhead during cache management.
  • Lower memory churn in the shared KV cache path, which complements the size-following fix landed in b9550.

Upgrade notes

This is a drop-in patch bump with no config, API, or CLI changes. The KV cache behavior is internal, so existing contexts and saved sessions should continue to work; only re-validate if you rely on precise memory-usage profiling. Note that the b9550 tensor-overflow fix is already included as the baseline.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

This release optimizes the key-value (KV) cache memory management by avoiding unnecessary cell copies, aiming to improve efficiency and reduce overhead.

Key Highlights

  • Optimization of KV cache memory usage.
  • Avoidance of unnecessary cell copies in the cache.

New Features

  • Optimized KV cache memory management

Release Notes (excerpt)

<details open> kv-cache : avoid kv cells copies (#24277) </details> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:** - Ubuntu x64 (CPU) - Ubuntu arm64 (CPU) - Ubuntu s390x (CPU) - Ubuntu x64 (Vulkan) - Ubuntu arm64 (Vulkan) - Ubuntu x64 (ROCm 7.2) - Ubuntu x64 (OpenVINO) - Ubuntu x64 (SYCL FP32) DISABLED **Android:** - Android arm64 (CPU) **Windows:** - Windows x64 (CPU)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories