>nobody in this threat knows how to actually use LLMs

get a chutes account, get the api key from it for deepseek, make an openrouter account, attach the api key from chutes to openrouter, then retrieve the API key from openrouter.

get sillytavern, set your connection profile API to text completion, add your openrouter API key to the tab, set your openrouter model as DeepSeek R1 0528 (free), and your model provider as Chutes - connect and authorize then test it - if it gens, everything worked correctly.

go to advanced formatting, set your context template to chatml, go to instruct template, set it to chatml as well and then set your system prompt to blank

go to ai response configuration, set your response tokens to about 600, context to 163k
temp 0.8, min p 0.02, rep penalty 1.08, freq penalty 0, top A to 0, presence penalty to 0, everything else set to 1

go to default author's note, set this below as your author's note to make your AI not generate garble:

>[SYSTEM NOTE: Avoid excessive line breaks. Keep dialogue, descriptions, and actions fluidly connected. Avoid prose; describe actions, thoughts, and emotions formally. Do not add reflective fluff or prompts to follow along at the end. Only write from the current character's perspective. Do not follow up on jokes, insults, or remarks. Avoid excessive dialogue. Use pronouns correctly. Use proper punctuation. Keep sentences concise.]

finally, get the guided generations extension for ST, add a custom guide, set it to 0 depth, and write whatever formatting you want the AI to shit out

congratulations, you're ready to goon, the hard part is finding or making chatbots that don't suck ass, try to find chatbots around 1500 in token count.