BestERP

Psyfighter 13B

nsfw
open router
overall
0.7
(12)
nsfw
0.8
(4)
role playing
0.8
(4)
story writing
0.5
(2)

Overview

According to AliCat's benchmark, this is their preferred model for role-play and adventure. The model may tend to loop / repeat itself, so consider using repetition penalty or other similar setting. Another suggestion is to use Generate only one line per request in the prompt.

Submit your review

2.5 Not good

Pick which version of the model you base your review on.

Reviews

4.5

One of my favorite small models for ERP (NSFW role-play). The model is very permissive in that if the {{user}} asks something, the {{char}} usually complies -- this may feel a bit unrealistic, but you get used to it.

Only tried AWQ quant.

4.5
quant: gguf q5

Mainly tested it in role play but for it, it is one of the coolest models I have yet experienced with no big obvious major flaws in it. It also tends to stick to its instructions. Can roleplay a character from just a few keywords but still generates a interesting characters. It is also not an over the top pervert like other models, giving characters a much more natural behaviour and therefor doesn't automaticly roleplay every woman as a sex hungry bimbo - You can still force them towards that behaviour, describing them as such. Also with some fetishes it behaves really subtle, which I like.

  • It has a tendency to not place or missplace asterisks for italics/empathis during chat, you are better off dismissing that and only rely on overall quotations for your chat experience.
  • It also has a tendency to be very descriptive/expressive with flowery prose. It can easily happen that it writes hundreds of words in a single reply only for it to contain one or two short sentences. If you want your characters speaking long monologues, that will likely be hard to achieve.
  • When it generates a new paragraph/line it sometimes switches the perspective of the character.
  • Like many other models, you can sometimes end up in a strange loop, where the character doesn't want to end an activity or is just repeating the same actions over and over again, even when shoved away from it - but it doesn't happen that much like in some other models.
  • Frequent use of short user replies don't shorten the replies from the AI characters over time. It really likes to generate and explain its stuff.
  • Extremely permissive, if you describe the character doing something in your reply, the AI character will most likely execute on it without any protest. Some other models can easily ignore you or sometimes even get mad at you for such behaviour. While it has its uses, this model most likely won't work as well for a more rebel type of chat experience.
4.0
quant: gguf q5

I actually prefer this to the V2 version, as I found it a bit more agreeable and helpful. That being said, I don’t feel either of these models are top tier. They’re not bad, however, and probably worth a gander.

2.5

This is absolutely garbage. The only thing that its good for is maybe basic conversation. Otherwise, I don't recommend this model.