List your company

Fused Matmul and Sampling on Synthetic D=4096, V=151k

1.98Speedup vs Multinomial SamplingFlashSampling
1.13761.35631.5751.7937Mar 16, 2026

Evaluation results

MethodLinks
FlashSamplingBatch size (B)=128, GPU=B3002026.031.982.281.33
FlashSamplingBatch size (B)=128, GPU=B2002026.031.891.551.27
FlashSamplingBatch size (B)=64, GPU=B2002026.031.841.671.39
FlashSamplingBatch size (B)=64, GPU=B3002026.031.812.521.37
FlashSamplingBatch size (B)=32, GPU=B2002026.031.681.681.38
FlashSamplingBatch size (B)=256, GPU=B3002026.031.652.041.09
FlashSamplingBatch size (B)=32, GPU=B3002026.031.642.471.34
FlashSamplingBatch size (B)=128, GPU=H1002026.031.641.351.12
FlashSamplingBatch size (B)=64, GPU=H1002026.031.61.461.26
FlashSamplingBatch size (B)=64, GPU=H2002026.031.61.471.21
FlashSamplingBatch size (B)=256, GPU=B2002026.031.581.321.07
FlashSamplingBatch size (B)=16, GPU=B2002026.031.571.651.36
FlashSamplingBatch size (B)=16, GPU=B3002026.031.542.471.33
FlashSamplingBatch size (B)=8, GPU=B2002026.031.531.611.33
FlashSamplingBatch size (B)=8, GPU=B3002026.031.52.461.3
FlashSamplingBatch size (B)=4, GPU=B2002026.031.471.611.32
FlashSamplingBatch size (B)=32, GPU=H2002026.031.471.471.22
FlashSamplingBatch size (B)=1, GPU=B2002026.031.461.511.32
FlashSamplingBatch size (B)=1, GPU=B3002026.031.462.451.31
FlashSamplingBatch size (B)=2, GPU=B2002026.031.461.561.3
FlashSamplingBatch size (B)=4, GPU=B3002026.031.442.371.3
FlashSamplingBatch size (B)=2, GPU=B3002026.031.432.181.28
FlashSamplingBatch size (B)=32, GPU=H1002026.031.381.371.2
FlashSamplingBatch size (B)=16, GPU=H2002026.031.371.451.21
FlashSamplingBatch size (B)=128, GPU=H2002026.031.371.120.91
FlashSamplingBatch size (B)=8, GPU=H2002026.031.311.441.19
FlashSamplingBatch size (B)=16, GPU=H1002026.031.31.351.18
FlashSamplingBatch size (B)=256, GPU=H1002026.031.291.040.81
FlashSamplingBatch size (B)=1, GPU=H2002026.031.281.351.2
FlashSamplingBatch size (B)=4, GPU=H2002026.031.271.411.18
FlashSamplingBatch size (B)=8, GPU=H1002026.031.271.331.17
FlashSamplingBatch size (B)=2, GPU=H2002026.031.261.381.17
FlashSamplingBatch size (B)=1, GPU=H1002026.031.251.31.18
FlashSamplingBatch size (B)=2, GPU=H1002026.031.241.31.16
FlashSamplingBatch size (B)=4, GPU=H1002026.031.241.321.16
FlashSamplingBatch size (B)=256, GPU=H2002026.031.170.980.78