b10412

ggml-org/llama.cppb10412Aug 13, 2026by github-actions[bot]same day as b10411

What changed since b10411

low risk

Enables backend sampling for speculative decoding with dflash and dspark models.

New features

  • Enables backend sampling support for speculative decoding
  • Supports backend sampling for dflash and dspark models
  • Allows p_min > 0 in backend sampling

Upgrade notes

This is a patch release with no breaking changes. Existing configurations should work without modification. Users utilizing speculative decoding may need to update model specifications to leverage the new backend sampling support.

Analyzed Oct 17, 2026

Upgrading?

Release notes summary

Enables backend sampling support for speculative decoding, allowing it to work with both dflash and dspark models.

Key Highlights

  • Enable backend sampling for both dflash & dspark (#26958)
  • Enable p_min > 0 in backend sampling

Release Notes (excerpt)

<details open> spec: enable backend sampling for both dflash & dspark (#26958) * dflash: enable backend sampling for both dflash & dspark * enable p_min > 0 in backend sampling and add guard * cont : add TODO --------- Co-authored-by: Georgi Gerganov <ggerganov@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