b10285

ggml-org/llama.cppb10285Aug 5, 2026by github-actions[bot]same day as b10284

What changed since b10284

low risk

b10285 adds multi-row batching for DeepSeek-OCR in mtmd, on top of b10284's MTP memory-allocation fix.

New features

  • mtmd now supports multi-row batching for the deepseek-ocr model, allowing multiple image rows to be processed together.

Improvements

  • DeepSeek-OCR rows are woven in a single pass rather than row-by-row, improving OCR throughput.
  • The MTP layer memory-allocation fix from b10284 is carried forward in this build.

Upgrade notes

This is a one-commit increment over b10284, so upgrading is a drop-in binary swap with no config or model-format changes. Only DeepSeek-OCR multi-modal workloads are affected; other models behave identically. If you skipped b10284, you also inherit its MTP memory-allocation fix.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Enhances multi-modal processing by supporting multi-row batching for the DeepSeek OCR model.

Key Highlights

  • Supports multi-row batching for deepseek-ocr
  • Improves efficiency for OCR tasks by weaving rows in one shot

Release Notes (excerpt)

<details open> mtmd: support multi-row batching for deepseek-ocr (#26154) * mtmd: support multi-row batching for deepseek-ocr * mtmd: weave deepseek-ocr rows in one shot instead of per row (#26615) --------- Co-authored-by: Saba Fallah <sabafallah@gmail.com> </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories