b9844

ggml-org/llama.cppb9844Jun 30, 2026by github-actions[bot]same day as b9843

What changed since b9843

low risk

b9844 adds NVIDIA FP4 (NVFP4) quantization support to the WebGPU backend, with no other functional changes.

New features

  • WebGPU backend can now load and run models quantized with NVIDIA FP4 (NVFP4), reducing memory footprint on supported hardware.

Improvements

  • WebGPU gains a lower-precision compute path that can improve throughput and memory efficiency for NVFP4-capable GPUs.

Upgrade notes

This is a drop-in binary update with no config or CLI changes required; existing WebGPU users keep their current behavior unless they explicitly switch to an NVFP4-quantized model. The revert of the split-compute synchronization optimization from b9843 remains in effect, so no action is needed there. Verify your GPU/driver stack actually supports FP4 before relying on the new path.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

This release introduces support for NVIDIA FP4 precision in the WebGPU backend, optimizing memory usage and performance for specific hardware. It provides standard binaries for all supported platforms.

Key Highlights

  • Adds NVFP4 (NVIDIA 4-bit floating point) support to WebGPU backend
  • Includes standard binaries for macOS, Linux, Android, Windows, and openEuler
  • KleidiAI support remains disabled

Release Notes (excerpt)

<details open> ggml-webgpu: add support for NVFP4 (#25143) </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