b9669

ggml-org/llama.cppb9669Jun 16, 2026by github-actions[bot]same day as b9668

What changed since b9668

medium risk

b9669 adds backend sampling support for the Eagle3 speculative decoding model, with no other functional changes.

New features

  • Backend sampling is now supported for the Eagle3 speculative decoding model, allowing sampling to run in the backend rather than on the host.

Improvements

  • Speculative decoding with Eagle3 can now use backend-side sampling, consistent with other spec-decoding implementations.

Upgrade notes

This is a narrow, additive change scoped to Eagle3 speculative decoding; no config or CLI migration is required. Users not running Eagle3 drafts will see no behavioral difference, and prebuilt binaries are otherwise equivalent to b9668.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

This release adds backend sampling support for the Eagle3 model, enhancing its integration capabilities. It provides pre-compiled binaries for macOS, Linux, Windows, Android, and openEuler across various hardware architectures and accelerators.

Key Highlights

  • Backend sampling support for Eagle3 model
  • Availability of macOS binaries (Apple Silicon and Intel)
  • Availability of Linux binaries (Ubuntu x64, arm64, s390x) with Vulkan, ROCm, OpenVINO, and SYCL support
  • Availability of Windows binaries (x64) with CUDA 12.4 and 13.3, Vulkan, SYCL, and HIP support
  • Availability of UI binary

Release Notes (excerpt)

<details open> spec: add backend sampling support for eagle3 (#24655) </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) - Ubuntu x64 (SYCL FP16) **Android:** - Android arm64 (CPU) **Windows:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories