LLM 遇到題目換個說法或數字,常會重新推理一次,讓重複工作花上更多成本。
- 這篇 NVIDIA 論文讓 Claude Opus 4.5 把推理過程轉成可替換輸入的 Python 函式,之後由 CPU 執行。
- 只有像擷取數值這類小型文字工作,才交給微調過的 1B 模型處理。這套方法在 8 項數學與科學基準中的 7 項,準確率追平或高於原模型。
- 在 AIME '25 的分數是 93.3%,原模型是 83.3%。研究也指出,遇到改寫或換數字的題目時,這種做法至少能處理 90%,成本可降低 16 到 291 倍。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
這項研究把重複推理改寫成可帶入新數字的 Python 函式,並比較了多項數學與科學基準的準確率和成本。
New NVIDIA paper finds that once an LLM has reasoned through a problem, the solution can run as CPU code for rephrased and re-numbered versions of it.
that answers similar questions 16x to 291x cheaper.
LLMs re-solve every problem from scratch, even when only wording or numbers change. Agents sending the same instructions with new data pay for that repeated reasoning each time.
Their method has Claude Opus 4.5 turn a reasoning trace into small Python functions, with changeable inputs as parameters. At run time they execute on a CPU, calling a fine-tuned 1B model only for small text jobs like extracting values.
The cache matched or beat the source model's accuracy on 7 of 8 math and science benchmarks, scoring 93.3% versus 83.3% on AIME '25. It handled at least 90% of rephrased or re-numbered variants.
If your agent runs the same prompt on new inputs all day, cache the reasoning as code and use the LLM only on misses.