🛠️ Sampler Parameters

Currently supported sampler parameters include:

• temperature

  Controls randomness — higher = more creative, lower = more deterministic
  Range：[0.0, 2.0]

• presence_penalty

  Penalizes tokens that have appeared before to encourage new topics
  Range：[-2.0, 2.0]

• frequency_penalty

  Penalizes frequent repetition of the same tokens to reduce verbosity
  Range：[-2.0, 2.0]

• top_p

  Aka Nucleus sampling — consider only tokens within the cumulative probability p
  Range：[0.0, 1.0]

• top_k

  Restrict sampling to the top k most probable tokens
  Range：[1, 100]

• min_p

  Minimum probability threshold — limits the token pool by cutting off low-probability tokens relative to the top token
  Range：[0.0, 1.0]

• top_a

  Adaptive nucleus sampling — a threshold for token selection based on the square of the highest token probability
  Range：[0.0, 1.0]

• repetition_penalty

  Attempts to curb repetition by penalizing tokens based on how often they occur in the context
  Range：[0.0, 2.0]