Inference Throughput on ISL 8192 OSL 16384 workload
224Max Batch SizeNano v3 12BA2A
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Nano v3 12BA2AISL/OSL=8192/16384, Hardware=H100 GPU, Precision=bfloat16, Framework=vLLM2026.05 | 224 | 2.4 | |
| Nano v3 23BA2.8AISL/OSL=8192/16384, Hardware=H100 GPU, Precision=bfloat16, Framework=vLLM2026.05 | 108 | 1.8 | |
| Nano v3 30BA3.6AISL/OSL=8192/16384, Hardware=H100 GPU, Precision=bfloat16, Framework=vLLM2026.05 | 36 | 1 |