b10415

ggml-org/llama.cppb10415Aug 13, 2026by github-actions[bot]same day as b10414

What changed since b10414

medium risk

b10415 adds automatic detection of MTP draft model types for speculative decoding, on top of b10414's Metal TQ2_0 support.

New features

  • Speculative decoding now auto-detects the MTP (Multi-Token Prediction) draft model type instead of requiring it to be specified manually.

Improvements

  • Carries forward b10414's Metal mul_mv kernel optimization (fewer integer ops, precomputed sums, hoisted coefficients, contiguous loads).
  • Retains Metal backend support for the TQ2_0 ternary 2-bit quantization type introduced in the prior build.

Upgrade notes

This is a single-commit increment over b10414 with no configuration or CLI changes required. If you use MTP-style draft models for speculative decoding, the draft type is now inferred, so any manual type flag can be dropped. Metal users keep the TQ2_0 and mul_mv improvements from the previous build.

Analyzed Oct 28, 2026

Upgrading?

Release notes summary

Adds automatic detection of MTP (Multi-Token Prediction) draft model types.

Key Highlights

  • Auto-detect mtp draft model type (#27005)

Release Notes (excerpt)

<details open> spec : auto-detect mtp draft model type (#27005) </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