GLM-4.7-Flash models with MTP
If you update llama.cpp to include this PR:
https://github.com/ggml-org/llama.cpp/pull/24868
then regenerating GLM-4.7-Flash will produce GGUFs with MTP support:
https://huggingface.co/zai-org/GLM-4.7-Flash
You could also regenerate some popular finetunes, for example:
https://huggingface.co/huihui-ai/Huihui-GLM-4.7-Flash-abliterated
https://huggingface.co/cerebras/GLM-4.7-Flash-REAP-23B-A3B
https://huggingface.co/koute/GLM-4.7-Flash-Derestricted
etc, etc
(but I assume standalone small MTP gguf is enough for them)
So far, I have only uploaded this one:
oi, missed it. We had a bit of backlog because I had a little break and my friend didnt queue anything =)
let me see...
ah yes, I might have noticed but didnt answer. We probably wont be doing that, because it's too much resources and time, unless you can provide me with new links, but idk if we will bother doing that, I need to see...
the link is: https://huggingface.co/zai-org/GLM-4.7-Flash, just create GGUF same way as before, it will contain MTP (like qwen or gemma models)
yeah I know, but we wont requant every small finetune of the model, since it's just way too many resources. Also usually MTP models require nolow quants, meaning we cannot provide small quants like q2, which some people find useful, so not sure if we even will handle this request