Change8

b10238

📦 llama-cppView on GitHub →
1 features🐛 6 fixes🔧 3 symbols

Summary

Introduced MTP support for Qwen3-Next models and resolved several Python type-checking and computation bugs. This release also includes updates to model-specific C++ files and constants.

✨ New Features

  • Added MTP support for Qwen3-Next.

🐛 Bug Fixes

  • Fixed Python type-check issues.
  • Corrected computation of num_mtp from the mtp layer.
  • Defined opt_num_mtp_layers in _QwenMtpMixin and improved comments.
  • Updated gguf-py/gguf/constants.py.
  • Rebased changes and added load_mtp flags.
  • Updated src/models/qwen3next.cpp.

Affected Symbols