Question about this repo?

[aahmozart]

1. Is there a prebuilt MTP sidecar anywhere (HF, release asset, Discord pin)? Their README documents --spec-draft-adaptive as speeding up MTP, so someone must be running it — I just can't tell what file they feed -md.
2. What is the exact 37-tensor set the loader expects? If it's the 33 non-indexer tensors plus four more, I may be able to build a compatible sidecar from the target GGUF without the 250 GB safetensors download.
3. Would they accept upstream's exports — i.e. support the qwen4exp.nextn_shared_target_tensors flag from PR models: Qwen3.8-Flash-Next MTP ggml-org/llama.cpp#28243, which would make the fork drop-in for anyone already on unsloth's GGUFs.

[voidsurfer]

Hey! Theres no dedicated weights for this repo, but we do have a discord through https://strixhalo.wiki/