b10505

ggml-org/llama.cppb10505Aug 20, 2026by github-actions[bot]1 days after b10502

What changed since b10502

low risk

b10505 adds an opt-in server preset option to deduplicate cached models, with no breaking changes from b10502.

New features

  • New server preset option `dedup-cache-models` that deduplicates models held in the server cache to reduce memory use.

Improvements

  • Server cache management is more efficient when the deduplication preset is enabled.
  • Release artifacts continue to be published across the same platform matrix (macOS, Linux CPU/Vulkan/OpenVINO/SYCL), so packaging is unchanged.

Upgrade notes

The new behavior is opt-in via the server preset, so existing configurations keep working unchanged. If you run multiple models or repeated loads on one server, consider enabling `dedup-cache-models` to cut cache memory. No config migration or rebuild steps are required beyond swapping the binary.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Introduces a new server preset option to deduplicate cached models, allowing for more efficient memory usage and cache management.

Key Highlights

  • Add dedup-cache-models preset option
  • Optimize server cache management

Release Notes (excerpt)

<details open> server: add dedup-cache-models preset option (#27346) </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 7.14)DISABLED - 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