X：Nathan Lambert (@natolambert)· @scottgeng00·· 3 天前AI 评分
Gacha Decoding：用外部 RNG 与指令遵循提升 LM 输出多样性，样本效率超此前方法 11 倍
RT by @natolambert: Introducing Gacha Decoding 🎲: we show how to diversify LM responses, beating prior work by >11x in sample efficiency. Reroll your LM for varied ideas, data, RL envs, {your use here}! We get diversity from instruction following + external RNG, not token entropy. Here's how 🧵
Gacha Decoding 通过给 LM 配备 RNG 工具并让其"掷骰规划"，把多样性来源从 token 熵转向指令遵循，样本效率比此前方法提升逾 11 倍。
来源：X：Nathan Lambert (@natolambert) · x.lingyaoai.com