Qwen3.5/3.6 MTP-only GGUF subsets released for grafting and acceleration

A user has published MTP-only GGUF subsets of Qwen3.5 and Qwen3.6 Middle/Large models (27B parameters and above) to facilitate experimentation with Qwen-based fine-tunes.

• The files accelerate token generation for Qwen-based models that lack MTP tensors by providing necessary draft weights.
• They demonstrate how Qwen3.5/3.6 MTP tensors can be embedded inside existing GGUF files, addressing issues found in other quantizations like Ornith-1.0-35B.
• The subsets support dual uses: serving as separate draft model files for llama.cpp or as grafting assets for transplanting MTP-related tensors.

These resources aim to make testing and experimenting with Qwen3.5/3.6-based fine-tunes easier by providing accessible, easy-to-test model files.