Mathematical Reasoning on GSM8K (tok/s, GSM8K strict)
138Throughput (tok/s)XFP V2 (Qwen 122B)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| XFP V2 (Qwen 122B)Eff. bits=~3.97, Platform=RTX PRO 6000 Blackwell, TP=2, Output tokens=1500, 5-shot=true2026.05 | 138 | 94.49 | |
| XFP V2 (Qwen 122B)Eff. bits=~3.97, Platform=RTX PRO 6000 Blackwell, TP=1, Output tokens=1500, 5-shot=true2026.05 | 110 | 94.62 | |
| XFP V2 (Qwen 35B)Eff. bits=~4.0, Platform=RTX PRO 6000 Blackwell, TP=1, Output tokens=1500, 5-shot=true2026.05 | 37 | 77.18 |