Mathematical Reasoning on GSM8K (Accuracy, TPS, Speedup)
420.7Throughput (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 | 420.7 | — | 4.46 | |
| PARSEModel=Qwen3-235B-A22B, Draft Model=Qwen3-8B, Decoding Strategy=Full and partial verify, Precision=FP8, Inference Framework=SGLang2026.05 | 405 | 95.2 | 4.29 | |
| Eagle3Model=Qwen3-235B-A22B, Decoding Strategy=Speculative decoding, Precision=FP8, Inference Framework=SGLang2026.05 | 176.4 | — | 1.87 | |
| Qwen3-235B-A22BModel=Qwen3-235B-A22B, Decoding Strategy=Plain autoregressive, Precision=FP8, Inference Framework=SGLang2026.05 | 94.3 | 94.8 | — | |
| Qwen3-8BModel=Qwen3-8B, Decoding Strategy=Standalone2026.05 | — | 94 | — |