b10021

ggml-org/llama.cppb10021Jul 15, 2026by github-actions[bot]same day as b10020

What changed since b10020

low risk

b10021 reduces DeepseekV4 graph splits for faster inference; the SYCL get_rows quantization fix from b10020 carries over unchanged.

Improvements

  • DeepseekV4 inference now uses fewer graph splits, cutting per-step graph overhead.
  • Release binaries rebuilt for all previously supported platforms (macOS, Linux CPU/Vulkan/ROCm/OpenVINO/SYCL).

Upgrade notes

This is a single-commit build bump; no config, CLI, or model-format changes are indicated. Users on DeepseekV4 should see modest speedups, while other architectures are effectively unaffected. The SYCL Q2_K/Q4_K/Q5_K get_rows fix from b10020 remains included, so no separate action is needed.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Optimizes DeepseekV4 graph splitting logic to reduce overhead, improving performance for inference tasks on this specific model architecture.

Key Highlights

  • Reduces graph splits for DeepseekV4
  • Standard release binaries for all supported platforms

Release Notes (excerpt)

<details open> DeepseekV4: reduce graph splits (#25702) </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:** - Windows x64 (CPU)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories