Minor GGUF and speculative decoding PRs

Two low-activity pull requests touched model loading and speculative decoding paths in major open-source ML stacks. Neither drew substantive human discussion beyond automated checks or a single clarifying question.

LFM2 support added to Transformers GGUF loader

A pull request in the Hugging Face transformers repository proposes adding LFM2 to the GGUF loading path. Activity consisted only of bot messages, and continuous integration completed cleanly. The change extends format compatibility for users who load models through that path.

Draft-vocabulary trim for Qwen MTP in llama.cpp

A llama.cpp pull request adds d2t draft-vocabulary trim support for Qwen 3.5 multi-token prediction speculative decoding. A single comment asked about the practical use case for the feature. The work remains a narrowly scoped proposal inside the ggml-org codebase.