Tag
Qwen team updated recommended temperature, top_p and presence_penalty values for their new 27B model to optimize both thinking and instruct modes.