b9905

ggml-org/llama.cppb9905Jul 8, 2026by github-actions[bot]same day as b9904

What changed since b9904

low risk

b9905 fixes quantized KV-cache handling for DeepSeek V4; otherwise identical to b9904.

Bug fixes

  • Corrected quantized key-value cache state management for DeepSeek V4 (dsv4), restoring consistency during inference.

Upgrade notes

This is a single targeted fix on top of b9904, so no config or API changes are required. Users running DeepSeek V4 with a quantized KV cache should upgrade to avoid incorrect cache state; others can treat it as a routine bump.

Analyzed Oct 30, 2026

Upgrading?

Release notes summary

Addresses a bug related to the quantized key-value cache for dsv4 (DeepSeek V4) to ensure correct state management and consistency.

Key Highlights

  • Fix quantized KV-cache for DeepSeek V4 (dsv4)
  • Update binaries for all supported platforms

New Features

  • KV-cache consistency fix

Release Notes (excerpt)

<details open> llama: fix quantized kv-cache for dsv4 (#25202) </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