Code Optimization on MBPP-Hard In-Domain
7.1Average Calls per TaskTextBFGS-REMO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TextBFGS-REMOFramework=One-Pass2026.01 | 7.1 | 2,262 | 16,100 | |
| TextBFGSKnowledge Retrieval=In-Domain MBPP KB2026.01 | 8.2 | 2,103.9 | 17,200 | |
| TextBFGSKnowledge Retrieval=w/o KB2026.01 | 18.2 | 1,330.2 | 24,200 | |
| TextGrad-MomentumStrategy=Stateless Momentum2026.01 | 34 | 1,114.1 | 37,900 | |
| TextGrad2026.01 | 36.6 | 727.3 | 26,600 |