b10509

ggml-org/llama.cppb10509Aug 20, 2026by github-actions[bot]same day as b10507

What changed since b10507

low risk

Adds new RoPE offset parameter with CPU, Metal, CUDA, and Vulkan support.

New features

  • Added ggml_rope_set_offset parameter
  • Metal backend support for rope offset
  • CUDA backend support for rope offset
  • Vulkan backend support for rope offset

Upgrade notes

Users can now control RoPE offsets using the new ggml_rope_set_offset parameter. This feature is supported on CPU, Metal, CUDA, and Vulkan backends.

Analyzed Oct 15, 2026

Upgrading?

Release notes summary

Adds a new parameter `ggml_rope_set_offset` to control Rotary Positional Embedding (RoPE) offsets, with full support implemented for CPU, Metal, CUDA, and Vulkan backends.

Key Highlights

  • Added ggml_rope_set_offset parameter
  • Metal backend support for rope offset
  • CUDA backend support for rope offset
  • Vulkan backend support for rope offset

Release Notes (excerpt)

<details open> ggml: add ggml_rope_set_offset (+ metal support) (#27120) * add params * cpu kernel * metal kernel * add test backend ops * gate other backends * ggml: (cuda) support ggml_rope_set_offset (#27121) * rm cuda supports_op guard, fix webgpu clang-format * ggml: support ggml_rope_set_offset on vulkan (#27344) * ggml: support ggml_rope_set_offset on vulkan * remove inplace optimization </details> **Website:** - <https://llama.app> **macOS/iOS:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories