b10259

ggml-org/llama.cppb10259Aug 4, 2026by github-actions[bot]same day as b10258

What changed since b10258

low risk

b10259 adds support for reshaping tensors during model loading, a single model-loader change over b10258.

New features

  • Tensors can now be reshaped while a model is being loaded, widening the range of checkpoints that load without conversion.

Improvements

  • Model loading is more flexible, reducing the need to pre-process or convert weights whose shapes differ from the expected layout.

Upgrade notes

This is a one-commit step from b10258; the sampler refactor that moved n_vocab into penalty_sampler was already shipped in the older build, so no additional sampler changes are needed. No config or CLI changes are documented, so a rebuild and re-pull of the matching binaries should suffice. Custom model loaders or code touching tensor shapes should be re-tested against the new load path.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Enables the reshaping of tensors during model loading to improve flexibility.

Key Highlights

  • Allow reshape of tensors during load
  • Improved model loading flexibility

Release Notes (excerpt)

<details open> model : allow reshape of tensors during load (#26531) </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