MTP
#2925
by Ahmad648794 - opened
does it have an mtp? if yes, then maybe it's already existing: https://hf.tst.eu/model#Qwen3.8-4B-Distill-GGUF
This model supports MTP and has been trained, but I need its .gguf version. The gguf version you have created is not MTP. Please create it. Thank you.
Please respond.
I could nuke and requant it with latest llama.cpp but if MTP still doesn't work it's an issue with either the model or llama.cpp. We always provide MTP by default for every model that supports it and this model is new enough that I don’t believe llama.cpp still lacked MTP for this architecture at the time we did it.