This model is intentionally trained on data that is known to contain offensive, disturbing, explicit, hateful or otherwise objectionable content. The outputs generated do not reflect the views or opinions of the developers and are intended solely for research and experimentation.
Controls randomness. Higher values produce more varied outputs.
0.05
Drops tokens below a fraction of the top token's probability.
1.1
Reduces the model's tendency to repeat the same lines.
toast
Hourly Tokens: ...
Bad Responses?
Yotsuba is a raw text completion model, not a chat model.
Unlike chat-tuned LLMs, Yotsuba generates plain text continuations that aren't meant to have a consistent length. It's trained on 4chan threads, which are inconsistent and chaotic.
Common issues you may see:
Stopping after generating only 1-2 posts.
Getting stuck in a repetitive loop.
Ending a thread early or formatting it inconsistently.
If a thread ends too soon, just keep hitting the continue button. This forces the
model to continue generating from where it stopped, but it won't necessarily work all the time. If it starts generating incoherent or one word posts, it's best to hit Reset and start a new thread.
Because this is a completion model, output quality can vary significantly depending
on the initial prompt. A strong opening usually produces much better results.
Yotsuba