Very Horny, Settings Recommendations?
I tried out a Q5_K_M quant.
This model is much, much hornier than Euryale 2.1 and 2.2. The only settings I used are Temp 1.2 and min-p 0.1. Everythings else was either disabled or on 1 (such as rep pen).
Do you have any other recommended settings (temp, DRY, XTC, etc.)?
Otherwise, this model is very good, remembering small details and even bringing them up directly or indirectly. But it really goes hard on the flirty side, ignoring sexual preferences, even when reminded. (Though this seems to be an issue with virtually any model I tried.)
It works amazing for me, try these settings. I'm using a paid llm host though and i'm not sure which quant. Also use the context and instruct templates from https://huggingface.co/Virt-io/SillyTavern-Presets if you want to try to have same settings as me. These work across the board in my experience with Euryale verions and Hanami.
It works amazing for me, try these settings. I'm using a paid llm host though and i'm not sure which quant. Also use the context and instruct templates from https://huggingface.co/Virt-io/SillyTavern-Presets if you want to try to have same settings as me. These work across the board in my experience with Euryale verions and Hanami.
I am using KoboldCPP, I'll try your settings as far as it allows. Some settings are not available on Kobold that I see here. We'll see, thanks.
It works amazing for me, try these settings. I'm using a paid llm host though and i'm not sure which quant. Also use the context and instruct templates from https://huggingface.co/Virt-io/SillyTavern-Presets if you want to try to have same settings as me. These work across the board in my experience with Euryale verions and Hanami.
I am using KoboldCPP, I'll try your settings as far as it allows. Some settings are not available on Kobold that I see here. We'll see, thanks.
What settings are you missing? i can set them to default and give a shot on getting it working with the settings you have available
Also, i use this as a post-history message, it might be what makes it behave:
[Prioritize realism and accuracy in character behavior and dialogue.
Ensure characters remain clothed when possible, only removing clothing when absolutely necessary.
Maintain a conversational tone, avoiding flowery language or over-embellishment.
{{char}} will avoid fast-forwarding the plot, ensuring the user retains control over pacing and direction.
All responses will be appropriately detailed, balancing narrative flow and user engagement. Phrasing will remain fresh and non-repetitive across responses.
Adhere to {{char}}'s established dialogue and internal monologue style.
Narrative elements will use minimalist prose, emphasizing “show, don’t tell” through clear sensory descriptions.
Always end every message from {{char}} with dialogue or actions to maintain immersion.]
It works amazing for me, try these settings. I'm using a paid llm host though and i'm not sure which quant. Also use the context and instruct templates from https://huggingface.co/Virt-io/SillyTavern-Presets if you want to try to have same settings as me. These work across the board in my experience with Euryale verions and Hanami.
I am using KoboldCPP, I'll try your settings as far as it allows. Some settings are not available on Kobold that I see here. We'll see, thanks.
What settings are you missing? i can set them to default and give a shot on getting it working with the settings you have available
Those are the available sampler settings in Kobold.