LLM Inference Throughput on NuminaMath
12.8Throughput (RPS)EB(k^*)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EB(k^*)GPU=H200, Model=Gemma-3-1B-IT2026.05 | 12.8 | 2.5 | |
| v1GPU=H200, Model=Gemma-3-1B-IT2026.05 | 12.49 | — | |
| v0GPU=H200, Model=Gemma-3-1B-IT2026.05 | 11.25 | — | |
| EB(k^*)GPU=RTX PRO 6000, Model=Gemma-3-1B-IT2026.05 | 8.87 | 6.1 | |
| v1GPU=RTX PRO 6000, Model=Gemma-3-1B-IT2026.05 | 8.36 | — | |
| v0GPU=RTX PRO 6000, Model=Gemma-3-1B-IT2026.05 | 7.01 | — |