experts_benchmark_bs1_seq128.csv

Batch Size,Seq Length,New Tokens,Torch Compile,Implementation,Mean Generation Latency (ms),Mean Prefill Latency (ms),Mean Decode Latency (ms),Peak Mem (MB)
1,128,16,False,eager,1452.00,480.20,971.81,27425.16
1,128,16,max-autotune-no-cudagraphs,eager,1620.98,498.00,1122.99,27410.34
1,128,16,False,grouped_mm,850.87,76.51,774.35,27425.40
1,128,16,max-autotune-no-cudagraphs,grouped_mm,492.87,76.79,416.08,27425.40
1,128,16,False,batched_mm,815.11,316.56,498.56,35866.47
1,128,16,max-autotune,batched_mm,412.98,316.33,96.65,35866.48
1,128,16,False,grouped_prefill+batched_decode,588.87,77.15,511.72,27470.51
1,128,16,max-autotune,grouped_prefill+batched_decode,181.79,76.78,105.01,27424.49
1,128,64,False,eager,4524.16,486.84,4037.31,27418.44
1,128,64,max-autotune-no-cudagraphs,eager,5116.71,477.42,4639.29,27419.62
1,128,64,False,grouped_mm,3327.45,76.46,3250.98,27434.67
1,128,64,max-autotune-no-cudagraphs,grouped_mm,1824.70,76.55,1748.15,27433.49
1,128,64,False,batched_mm,2411.23,316.18,2095.05,35875.48
1,128,64,max-autotune,batched_mm,707.73,316.24,391.49,35875.48
1,128,64,False,grouped_prefill+batched_decode,2219.34,76.89,2142.45,27479.51
1,128,64,max-autotune,grouped_prefill+batched_decode,489.06,76.88,412.18,27433.50