b9978

ggml-org/llama.cppb9978Jul 12, 2026by github-actions[bot]same day as b9977

What changed since b9977

low risk

Server now evicts checkpoints that are too close in step count to curb redundant file accumulation.

New features

  • Server checkpoint management now evicts checkpoints whose step counts fall within a minimum-step window of each other.

Improvements

  • Reduces redundant checkpoint files, lowering disk usage for long-running server sessions.
  • Storage management for checkpoints is now handled automatically rather than accumulating every saved state.

Upgrade notes

This is a single-commit server-side change with no API or config surface documented in the notes. If you rely on retaining closely spaced checkpoints for fine-grained rollback, verify the eviction threshold suits your workflow. Otherwise the upgrade is a drop-in binary swap.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Improves checkpoint management in the server by evicting checkpoints that are too close in step count, helping to optimize storage usage and prevent accumulation of redundant files.

Key Highlights

  • Checkpoint eviction optimization
  • Prevents redundant checkpoint accumulation
  • Server storage management improvement

Release Notes (excerpt)

<details open> server : evict checkpoints within min-step of each other (#25472) </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories