Memory-Augmented Compression:将可复用推理注入提示
This paper shows a different way to make chain-of-thought cheaper: move reusable…
提供了一条无需微调即可显著加速推理并维持精度的新路径,对 Agent 场景极具参考价值,建议关注其工程落地潜力。
This paper shows a different way to make chain-of-thought cheaper: move reusable reasoning from generation into the prompt.
本文展示了一种降低思维链成本的不同方法:将可复用的推理从生成阶段移至提示词中。
Instead of asking an LLM to regenerate every reasoning step, give it the relevant reasoning pattern upfront and let it think shorter.
与其让大语言模型重新生成每一个推理步骤,不如提前提供相关的推理模式,让它进行更简短的思考。
Memory-Augmented Compression is training-free: it distills solved examples into reusable reasoning memories, retrieves relevant ones for each query, and injects them before compressed reasoning.
记忆增强压缩是一种无需训练的方法:它将已解决的示例提炼为可复用的推理记忆,为每个查询检索相关记忆,并在压缩推理之前将其注入。
That shifts some work from slow autoregressive decoding to the more parallel prefill stage.
这将部分工作从缓慢的自回归解码转移到了并行度更高的预填充阶段。
With Qwen2.5-7B, adding memory to Chain-of-Draft recovered 21.4 accuracy points on GSM8K and 28.0 on MATH, while model latency stayed 1.49× and 1.14× faster than standard CoT.
在 Qwen2.5-7B 上,为草稿链添加记忆后,在 GSM8K 上恢复了 21.4 个准确率点,在 MATH 上恢复了 28.0 个准确率点,而模型延迟仍比标准思维链快 1.49 倍和 1.14 倍。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力