[Feature Request] MTP support when loading corresponding gguf

[johnlabor]

Describe the Issue

There is a new PR regarding MTP support in llama.cpp:

• llama + spec: MTP Support  ggml-org/llama.cpp#22673 (superseded llama : add MTP API ggml-org/llama.cpp#18886)

When searching in this repo I see this comment

// NextN/MTP tensors are currently ignored (reserved for future MTP support)

I am currently having some MTP preserved MLP NVFP4 gguf.
They can be loaded by kobold but I think MTP is being ignored due to no support.

Wonder what is the effort to support it in Koboldcpp once merging the upstream ggml/llama.cpp changes.

Additional Information:
The gguf I am using to test:
https://huggingface.co/llmfan46/Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved-NVFP4-GGUF

Thanks.

[henk717]

We'll have to see what needs to happen on our side once this gets merged, but this is the kind of stuff we do tend to add.
Either we'll do nothing and it just works, or it needs us to pass trough new things to llamacpp which will take some work on our side.
Normally we always wait until things actually land in llamacpp before integrating it, we don't cherry pick their PR's like this.