b10048

ggml-org/llama.cppb10048Jul 16, 2026by github-actions[bot]same day as b10047

What changed since b10047

low risk

b10048 fixes Transformer Parallelism support for Phi3, Bert, Plamo2/3, and ChatGLM architectures.

Improvements

  • Transformer Parallelism now works correctly for Phi3, Bert, Plamo2/3, and ChatGLM model families.
  • Pre-built binaries refreshed across macOS, Linux, Windows, and Android, with KleidiAI still disabled on Apple Silicon.

Bug fixes

  • Corrected TP handling for Phi3 models.
  • Corrected TP handling for Bert models.
  • Corrected TP handling for Plamo2/3 models.
  • Corrected TP handling for ChatGLM models.

Upgrade notes

This is a narrow bug-fix build on top of b10047, so no config or API changes are expected. Users running tensor/transformer-parallel inference on Phi3, Bert, Plamo2/3, or ChatGLM should upgrade to get correct behavior. The BoringSSL bump from b10047 carries over unchanged.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

This release fixes Transformer Parallelism (TP) support for specific model architectures including Phi3, Bert, Plamo2/3, and ChatGLM.

Key Highlights

  • Fixed TP for Phi3, Bert, Plamo2/3, and ChatGLM models.
  • Included standard cross-platform binaries for macOS, Linux, Android, and Windows.
  • KleidiAI support is disabled on macOS Apple Silicon for this build.

Release Notes (excerpt)

<details open> TP: fix Phi3, Bert, Plamo2/3, ChatGLM (#25536) </details> **Website:** - <https://llama.app> **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:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories