Dialogue on MT-Bench (Score, TPS, and Speedup)
194TPS (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 | 194 | — | 2.03 | |
| PARSEModel=Qwen3-235B-A22B, Draft Model=Qwen3-8B, Decoding Strategy=Full and partial verify, Precision=FP8, Inference Framework=SGLang2026.05 | 167.5 | 8.88 | 1.75 | |
| Eagle3Model=Qwen3-235B-A22B, Decoding Strategy=Speculative decoding, Precision=FP8, Inference Framework=SGLang2026.05 | 153.9 | — | 1.61 | |
| Qwen3-235B-A22BModel=Qwen3-235B-A22B, Decoding Strategy=Plain autoregressive, Precision=FP8, Inference Framework=SGLang2026.05 | 95.6 | 9.19 | — | |
| Qwen3-8BModel=Qwen3-8B, Decoding Strategy=Standalone2026.05 | — | 7.93 | — |