List your company

End-to-end decode throughput on Context Length 32K

183Decode Throughput (tok/s)Llama-3.1-8B
89.712113.931138.15162.369May 13, 2026

Evaluation results

MethodLinks
Llama-3.1-8BKV Representation=SphKV2026.05183
GPT-ossKV Representation=SphKV2026.05172.5
Qwen2.5-14BKV Representation=SphKV2026.05153.9
Llama-3.1-8BKV Representation=Dense2026.05110
GPT-ossKV Representation=Dense2026.05104.9
Qwen2.5-14BKV Representation=Dense2026.0593.3