Text Generation
Transformers
Safetensors
llama
Not-For-All-Audiences
nsfw
text-generation-inference
Inference Endpoints
My Tuned Settings For Best Dawn Results
#4
by
Delcos
- opened
After testing on multiple conversations and characters, these are the settings that seem to give the best results and keep in the same style of speaking as the character given without causing repetition. Everything was done with full context to make sure that doesn't become an issue later.
temperature: 0.9
top_p: 0.99
top_k: 85
repetition_penalty: 1.2
presence_penalty: 5.8
frequency_penalty: 5.5
typical_p: 0.68
tfs: 0.68
min_p is also good but it seems kinda on and off with Dawn.