llama.cpp
OGA Hybrid
OGA NPU
OGA CPU
FastFlowLM NPU
Use Lemonade with your favorite app
Controls randomness in responses (0 = deterministic, 2 = very random)
Limits token selection to top K most likely tokens
Nucleus sampling - considers tokens with cumulative probability up to P
Penalty for repeating tokens (1 = no penalty, >1 = less repetition)
Determines whether hybrid reasoning models, such as Qwen3, will use thinking.