Extreme repetition
Great model, but it repeats phrases, paragraphs, and sometimes whole responses very often.
I increased repetition_penalty but not much luck.
Are there any recommended settings to fix this issue?
I am using q5.
Try to neutralize sampler, and only use Temperature and min_p ! (try 0.5 temp 0.05 min_p)
Try to neutralize sampler, and only use Temperature and min_p ! (try 0.5 temp 0.05 min_p)
I use WebUI not Sillytavern, so I don't think there is a neutralize sampler option.
I reset all my settings and only set 0.5 temp 0.05 min_p, no luck.
dry could possibly fix this, but it doesn't work with GGUF.
Hmmmm... I use Kobold + ST so maybe the sampler don't act exactly the same.
@itsnottme DRY Sampler absolutely does work with GGUF files, just here to correct it. It's also available in Ooba.
@MarinaraSpaghetti
It doesn't show up when I am using llama.cpp. Am I missing something?
Shows up fine on ExLlamav2_HF
Edit:
For anyone trying to use Dry with GGUF, convert your GGUF file and use llamacpp_HF.