model_name_or_path: /path/to/model
sampling:
  max_new_tokens: 1024
  repetition_penalty: 1.0
  temperature: 1.0
  top_k: null
  top_p: 1.0
server:
  host: 0.0.0.0
  max_image_buffer_size: 2
  port: 8341
  ttl: 1800
w8a8:
  mode: 'off'