b9895

ggml-org/llama.cppb9895Jul 7, 2026by github-actions[bot]same day as b9894

What changed since b9894

low risk

b9895 fixes an out-of-bounds read in the ngram-map during speculative decoding prompt shrink, plus cleanup cutoff handling.

Improvements

  • Speculative decoding is more robust when prompts are shrunk, with corrected ngram-map cleanup cutoff behavior.
  • Memory handling during prompt shrink operations is safer, reducing the chance of crashes or corruption.

Bug fixes

  • Fixed an out-of-bounds read in the ngram-map that occurred when the prompt was shrunk during speculative decoding.

Upgrade notes

This is a small incremental build on top of b9894 with no API or configuration changes. Users of speculative decoding, especially with prompt shrinking or cache reuse, should upgrade to avoid the out-of-bounds read. No migration steps are required.

Analyzed Oct 28, 2026

Upgrading?

Release notes summary

Fixes out-of-bounds read in ngram-map during prompt shrinking operations, along with cleanup cutoff improvements for speculative decoding.

Key Highlights

  • Fix out-of-bounds read in ngram-map on prompt shrink
  • Improved speculative decoding robustness
  • Better memory management during prompt operations

New Features

  • Speculative decoding prompt shrink fix

Release Notes (excerpt)

<details open> speculative : fix out-of-bounds read in ngram-map on prompt shrink (#23936) * speculative : fix out-of-bounds read in ngram-map on prompt shrink * speculative : fix ngram-map cleanup cutoff after prompt shrink </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories