b9716

ggml-org/llama.cppb9716Jun 19, 2026by github-actions[bot]same day as b9715

What changed since b9715

medium risk

b9716 adds batching support for InternVL multimodal inference, a small incremental step over b9715's CUDA COL2IM_1D work.

New features

  • InternVL models in the mtmd multimodal path can now process batched inputs instead of one item at a time.

Improvements

  • Batched InternVL processing improves throughput for multi-image or multi-input multimodal workloads.
  • Release artifacts now include an Ubuntu x64 OpenVINO 2026.2 build alongside the existing CPU, Vulkan, ROCm and SYCL packages.

Upgrade notes

No configuration or API changes are required; this is a drop-in build bump from b9715. Rebuild or pull the new binaries only if you use InternVL through the multimodal (mtmd) path or want the newly published OpenVINO package. The CUDA COL2IM_1D support introduced in b9715 carries over unchanged.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Implements batching support for the InternVL multimodal model, improving efficiency when processing multiple inputs.

Key Highlights

  • Added batching support for InternVL
  • Improved performance for internvl workloads
  • Enhanced multimodal processing capabilities

Release Notes (excerpt)

<details open> mtmd: add batching support for internvl (#24775) </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:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories