List your company

PrefixLM Attention on Llama3.1 405B (q=128, k=8) (1k)

122.22PrefixLM Attention TFLOPS (1k)CuBridge
10.336839.383468.4397.4766May 6, 2026

Evaluation results

MethodLinks
CuBridgeImplementation=CuBridge2026.05122.22
FlexAttentionImplementation=PyTorch FlexAttention2026.05107.89
Qimeng AttnImplementation=Qimeng Attention2026.0576.01
TorchImplementation=Standard PyTorch2026.0514.64