Code Generation on HumanEval+ (Accuracy, TPS (tok/s), Speedup)
80.5AccuracyGLM-4.7-FP8
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GLM-4.7-FP8Mode=Standalone2026.05 | 80.5 | — | 66.4 | — | |
| PARSETarget Model=GLM-4.7-FP8, Draft Model=Qwen3.5-9B2026.05 | 80.5 | — | 147 | 2.21 | |
| Qwen3.5-9BMode=Standalone2026.05 | 73.2 | — | — | — |