Garbled Output...
When I try to run Nexus-IKM-Mistral-7B, converted and quantized to gguf (for llama.cpp), I get garbled output. For comparison, your earlier GGUF's still function fine. Please fix this. 🙏
user: Hi!
assistant: argued checkura female longitude
arguedrowpostrans attended Italianaніura ab attended EXISTSte
DEind v end attended «iderwww ab attended profilesclient)) argued damals sulle EXISTSte
endscript ab Украи Johound subст ott Septinger specopieider to spec BibliothèqueennнеartCD h argued damals sulle French pse
DE un
DEind non things--- Italiana ott Jas ab attendedéste
end femaleot ab attended longitude
end dat attended Realprops Audio Italiana Question Jas,- walk profilesacionesse
DE un
argued un
atia
Item ent,-********jin Flyáss public \ out sinceiseste
argued
Ahh damn, I think the wrong one pushed to the hub...my bad!!! I have about a dozen different versions I am experimenting with and mislabeled this one.
I'll push the correct to the hub this morning! Thank you for catching this, apologies for the lobotomized LLM!
No worries! I appreciate your endeavors.
Working version up now, I appreciate the patience! This was the version that was up before the recent update. There will be the improved model I was trying to load up later today. Thanks for your interest in this one!
Improved version that I had originally meant to push should be up now! It is the working version of the V5 experimental repo so it has the latest and best training *Laser (took a few tries to really get the model to not fry)
The latest push to the repo works! Thanks, Severian.