Great Model!!
This model is spectacular. Thank you so much. He does not usually repeat answers or phrases like other models. So far, it is the best Rolepay model based on Llama 3. a spectacle
Thank you this model is incredible so far. I'm having one problem - When I use the model for roleplay it often types 'assistant' or sometimes 'user assistant' then writes the next line on behalf of the 'user'. I've tried instructing the model not to do this in the prompt but to no avail! any ideas?
Edit: sorry i've just realised i've posted this on Andersons thread accidentally, also i'm using the GGUF file so I will repost this there.
Thank you this model is incredible so far. I'm having one problem - When I use the model for roleplay it often types 'assistant' or sometimes 'user assistant' then writes the next line on behalf of the 'user'. I've tried instructing the model not to do this in the prompt but to no avail! any ideas?
Edit: sorry i've just realised i've posted this on Andersons thread accidentally, also i'm using the GGUF file so I will repost this there.
You need to set the EOS token <|eot_id|> (ID 12009) correctly in your inference program (which is Faraday). I don't know how it's handled there. If there's no way to do this, LM Studio, Koboldcpp or Ooba should work.