Text Generation
Transformers
Safetensors
llama
UNA
juanako
cybertron
xaberius
Eval Results
Inference Endpoints
text-generation-inference

Hallucinations

#2
by PierreMenard - opened

This model is hallucinating much worse than Yi-Chat. It's nearly impossible to get any factually correct information out of it sadly.

Pierre,

Firstly thanks for giving a shot to my beta model, remember bro.. its a beta :)
I guess you already experienced the power of Xaberius (partially) .. and u want more.. fair enough.
IF you want your issue solved by V1-RC or FINAL.. i will need from you and the community:

  • HyperParams (temp, k, t, penalty, etc)
  • Prompt.. conversation, etc.. overall I need to be able to reproduce the issue..
  • Model (better a url of it .. like TheBloke/model-etc/quantz.gguf file)
    Thanks

istock-145839390-0d4d64debfba2a5a46fa4c585bd3ad50.webp

fblgit changed discussion status to closed
PierreMenard changed discussion status to open

Can U provide the requested details?

Updated to the right tokenizer, give it a try or provide more reproducible steps / prompts / config, etc.

fblgit changed discussion status to closed

Sign up or log in to comment