Code Generation on HumanEval+ (Accuracy, TPS, Speedup)
193.1Throughput (tok/s)PARSE + EAGLE3
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PARSE + EAGLE3Model=Qwen3-235B-A22B, Draft Model=Qwen3-8B, Decoding Strategy=PARSE with EAGLE3 on both draft and target, Precision=FP8, Inference Framework=SGLang2026.05 | 193.1 | — | 2.08 | |
| Eagle3Model=Qwen3-235B-A22B, Decoding Strategy=Speculative decoding, Precision=FP8, Inference Framework=SGLang2026.05 | 166.6 | — | 1.8 | |
| PARSEModel=Qwen3-235B-A22B, Draft Model=Qwen3-8B, Decoding Strategy=Full and partial verify, Precision=FP8, Inference Framework=SGLang2026.05 | 149.6 | 87.2 | 1.62 | |
| Qwen3-235B-A22BModel=Qwen3-235B-A22B, Decoding Strategy=Plain autoregressive, Precision=FP8, Inference Framework=SGLang2026.05 | 92.6 | 83.5 | — | |
| Qwen3-8BModel=Qwen3-8B, Decoding Strategy=Standalone2026.05 | — | 70.7 | — |