DSV4 Flash 0731 and MTP in llama.cpp?

<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1t3guzw/llamacpp_mtp_support_now_in_beta/"> <img alt="Llama.cpp MTP support now in beta!"

<!-- SC_OFF --><div class="md"><p>As in title - NVFP4 + MTP at once on llama.cpp<br /> <a

<!-- SC_OFF --><div class="md"><p>Some of you may remember old models from GLM: GLM Air or GLM Flash. I know they’re outdated, but I have a soft spot for them, so I am currently

<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1teryn8/llama_spec_mtp_support_by_am17an_pull_request/"> <img alt="llama + spec: MTP Support by am17an ·

<!-- SC_OFF --><div class="md"><p>If you're using MTP with Qwen Flash Next and llama.cpp, you can switch to:</p> <p>quants: <a

<!-- SC_OFF --><div class="md"><p>It does not like anything I have tried, including what works with llama-server. is it not built to work with speculative decoding?</p> </div><!--

<!-- SC_OFF --><div class="md"><p>DeepSeekv3 OG</p> <p>DeepSeekv3.2/4</p> <p>Qwen3.5</p> <p>GLM4.5+</p> <p>MiniMax2.5+</p> <p>Step3.5Flash</p> <p>Mimo v2+</p> <p>Until we get mtp

<!-- SC_OFF --><div class="md"><p>PR #22673 (commit 4f13cb7) landed MTP speculative decoding in mainline llama.cpp on May 16. I tested it on two separate rigs.</p> <p>Qwen3.6 27B,