Code Generation on HumanEvalPlus (Accuracy)
90.85HumanEval+ AccuracyParallel-Synthesis
Evaluation results
| Method | Links | |
|---|---|---|
| Parallel-SynthesisBackbone=Qwen3-14B, Workers=3, Merging=Weighted averaging (lambda=0.5)2026.06 | 90.85 | |
| Text-SerializationBackbone=Qwen3-14B, Workers=32026.06 | 90.24 | |
| VotingBackbone=Qwen3-14B, Workers=32026.06 | 89.63 | |
| KVLINKBackbone=Qwen3-14B, Workers=32026.06 | 87.8 | |
| SingleBackbone=Qwen3-14B, Workers=12026.06 | 84.15 | |
| CacheBlendBackbone=Qwen3-14B, Workers=32026.06 | 51.22 | |
| APEBackbone=Qwen3-14B, Workers=32026.06 | 36.59 |