Code Generation on HumanEval Hard
97.78Base Pass RateTextBFGS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TextBFGSKnowledge Base (KB) Source=MBPP KB, Retrieval Strategy=gradient-based retrieval2026.01 | 97.78 | 93.33 | |
| TextBFGS-REMOKnowledge Base (KB) Source=MBPP KB, Retrieval Strategy=input-based retrieval2026.01 | 95.56 | 91.11 | |
| TextGradKnowledge Base (KB) Source=None2026.01 | 91.11 | 82.22 | |
| TextGrad-MomentumKnowledge Base (KB) Source=None, Momentum=true2026.01 | 91.11 | 86.67 | |
| TextBFGS (w/o KB)Knowledge Base (KB) Source=None2026.01 | 91.11 | 82.22 |