Chat

Settings

System prompt Background instructions that steer how the model responds. Max new tokens: Maximum length of the reply, in tokens. Image token budget Higher values preserve more visual detail (OCR/documents). Lower is faster. Temperature: Higher is more random and creative; 0 is deterministic. Top-p: Keep the smallest set of tokens whose probabilities sum to this. Top-k: Consider only the k most likely tokens (0 turns it off). Repetition penalty: Higher discourages repeating words; 1 turns it off.