source stringclasses 13
values | gpu stringclasses 4
values | model stringclasses 13
values | kernel_family stringclasses 7
values | kernel_name stringclasses 328
values | dtype stringclasses 2
values | held_out bool 2
classes | M float64 1 16.4k ⌀ | N float64 8 248k ⌀ | K float64 2.05k 29.6k ⌀ | bs float64 1 1 ⌀ | seq float64 1 16.4k ⌀ | n_heads float64 8 64 ⌀ | head_dim float64 64 256 ⌀ | kv_heads float64 1 64 ⌀ | numel float64 64 531M ⌀ | op_type stringclasses 177
values | gpu_time_duration_ms float64 0 703 | launch_block_size float64 16 512 | launch_grid_size float64 1 1.04M | dram_bytes_sum float64 1.27 94.3B ⌀ | launch_registers_per_thread float64 16 255 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 152,064 | 8,192 | 1 | 12,288 | null | null | null | null | qwen72b_lm_head | 527.53 | 256 | 57,600 | 69,540,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 152,064 | 8,192 | 1 | 12,288 | null | null | null | null | qwen72b_lm_head | 527.57 | 256 | 57,600 | 52,350,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 2,048 | 2,048 | 1 | 12,288 | null | null | null | null | filler_sq2k | 1.96 | 256 | 768 | 202,620,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 2,048 | 2,048 | 1 | 12,288 | null | null | null | null | filler_sq2k | 1.96 | 256 | 768 | 202,630,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 2,048 | 2,048 | 1 | 12,288 | null | null | null | null | filler_sq2k | 1.96 | 256 | 768 | 202,620,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 5,120 | 5,120 | 1 | 12,288 | null | null | null | null | filler_sq5k | 11.44 | 256 | 2,304 | 1,010,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 5,120 | 5,120 | 1 | 12,288 | null | null | null | null | filler_sq5k | 11.44 | 256 | 2,304 | 1,010,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 12,288 | 5,120 | 5,120 | 1 | 12,288 | null | null | null | null | filler_sq5k | 11.44 | 256 | 2,304 | 1,010,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 4,096 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_q_or_o | 9.58 | 256 | 2,048 | 855,070,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 4,096 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_q_or_o | 9.58 | 256 | 2,048 | 854,760,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 4,096 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_q_or_o | 9.58 | 256 | 2,048 | 854,900,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 6,144 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_qkv_fused | 14.55 | 256 | 3,072 | 1,380,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 6,144 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_qkv_fused | 14.55 | 256 | 3,072 | 1,380,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 6,144 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_qkv_fused | 14.55 | 256 | 3,072 | 1,380,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s1688gemm_bf16_128x128_ldg8_f2f_stages_32x1_nn | bf16 | false | 16,384 | 1,024 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_kv | 2.47 | 128 | 1,024 | 399,900,000 | 234 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s1688gemm_bf16_128x128_ldg8_f2f_stages_32x1_nn | bf16 | false | 16,384 | 1,024 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_kv | 2.47 | 128 | 1,024 | 405,410,000 | 234 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s1688gemm_bf16_128x128_ldg8_f2f_stages_32x1_nn | bf16 | false | 16,384 | 1,024 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_kv | 2.47 | 128 | 1,024 | 388,570,000 | 234 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 14,336 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_ffn_gate_up | 33.68 | 256 | 7,168 | 3,120,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 14,336 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_ffn_gate_up | 33.68 | 256 | 7,168 | 3,120,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 14,336 | 4,096 | 1 | 16,384 | null | null | null | null | llama8b_ffn_gate_up | 33.68 | 256 | 7,168 | 3,120,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 4,096 | 14,336 | 1 | 16,384 | null | null | null | null | llama8b_ffn_down | 33.07 | 256 | 2,048 | 2,660,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 4,096 | 14,336 | 1 | 16,384 | null | null | null | null | llama8b_ffn_down | 33.07 | 256 | 2,048 | 2,660,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 4,096 | 14,336 | 1 | 16,384 | null | null | null | null | llama8b_ffn_down | 33.07 | 256 | 2,048 | 2,660,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 128,256 | 4,096 | 1 | 16,384 | null | null | null | null | llama_lm_head_d4k | 299.56 | 256 | 64,128 | 27,250,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 128,256 | 4,096 | 1 | 16,384 | null | null | null | null | llama_lm_head_d4k | 299.54 | 256 | 64,128 | 27,260,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 128,256 | 4,096 | 1 | 16,384 | null | null | null | null | llama_lm_head_d4k | 299.54 | 256 | 64,128 | 27,240,000,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_q_or_o | 37.94 | 256 | 4,096 | 3,510,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_q_or_o | 37.94 | 256 | 4,096 | 3,510,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_q_or_o | 37.94 | 256 | 4,096 | 3,510,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 10,240 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_qkv_fused | 47.79 | 256 | 5,120 | 4,400,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 10,240 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_qkv_fused | 47.79 | 256 | 5,120 | 4,400,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 10,240 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_qkv_fused | 47.79 | 256 | 5,120 | 4,400,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 1,024 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_kv | 5.33 | 256 | 512 | 424,690,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 1,024 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_kv | 5.33 | 256 | 512 | 424,770,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | ampere_bf16_s16816gemm_bf16_256x128_ldg8_f2f_stages_32x3_nn | bf16 | false | 16,384 | 1,024 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_kv | 5.33 | 256 | 512 | 424,590,000 | 218 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 28,672 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_ffn_gate_up | 132.65 | 256 | 14,336 | 11,340,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 28,672 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_ffn_gate_up | 132.66 | 256 | 14,336 | 11,400,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 28,672 | 8,192 | 1 | 16,384 | null | null | null | null | llama70b_ffn_gate_up | 132.65 | 256 | 14,336 | 11,250,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 28,672 | 1 | 16,384 | null | null | null | null | llama70b_ffn_down | 132.02 | 256 | 4,096 | 11,640,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 28,672 | 1 | 16,384 | null | null | null | null | llama70b_ffn_down | 132.02 | 256 | 4,096 | 11,640,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 28,672 | 1 | 16,384 | null | null | null | null | llama70b_ffn_down | 132.01 | 256 | 4,096 | 11,630,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 128,256 | 8,192 | 1 | 16,384 | null | null | null | null | llama_lm_head_d8k | 593.47 | 256 | 64,512 | 79,740,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 128,256 | 8,192 | 1 | 16,384 | null | null | null | null | llama_lm_head_d8k | 593.49 | 256 | 64,512 | 82,840,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 128,256 | 8,192 | 1 | 16,384 | null | null | null | null | llama_lm_head_d8k | 593.45 | 256 | 64,512 | 88,150,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 29,568 | 8,192 | 1 | 16,384 | null | null | null | null | qwen72b_ffn_gate_up | 137.96 | 256 | 15,360 | 11,880,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 29,568 | 8,192 | 1 | 16,384 | null | null | null | null | qwen72b_ffn_gate_up | 137.96 | 256 | 15,360 | 11,970,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 29,568 | 8,192 | 1 | 16,384 | null | null | null | null | qwen72b_ffn_gate_up | 137.96 | 256 | 15,360 | 11,760,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 29,568 | 1 | 16,384 | null | null | null | null | qwen72b_ffn_down | 136.14 | 256 | 4,096 | 12,000,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 29,568 | 1 | 16,384 | null | null | null | null | qwen72b_ffn_down | 136.14 | 256 | 4,096 | 12,010,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 8,192 | 29,568 | 1 | 16,384 | null | null | null | null | qwen72b_ffn_down | 136.13 | 256 | 4,096 | 11,990,000,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 152,064 | 8,192 | 1 | 16,384 | null | null | null | null | qwen72b_lm_head | 703.39 | 256 | 76,800 | 94,300,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 152,064 | 8,192 | 1 | 16,384 | null | null | null | null | qwen72b_lm_head | 703.39 | 256 | 76,800 | 88,830,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 152,064 | 8,192 | 1 | 16,384 | null | null | null | null | qwen72b_lm_head | 703.4 | 256 | 76,800 | 85,200,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 2,048 | 2,048 | 1 | 16,384 | null | null | null | null | filler_sq2k | 2.54 | 256 | 1,024 | 267,190,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 2,048 | 2,048 | 1 | 16,384 | null | null | null | null | filler_sq2k | 2.54 | 256 | 1,024 | 267,190,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_128x256_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 2,048 | 2,048 | 1 | 16,384 | null | null | null | null | filler_sq2k | 2.54 | 256 | 1,024 | 267,180,000 | 222 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 5,120 | 5,120 | 1 | 16,384 | null | null | null | null | filler_sq5k | 15.25 | 256 | 3,072 | 1,320,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 5,120 | 5,120 | 1 | 16,384 | null | null | null | null | filler_sq5k | 15.25 | 256 | 3,072 | 1,320,000,000 | 215 |
roofline_sweep | RTX3090 | roofline | gemm | void cutlass::Kernel2<cutlass_80_tensorop_bf16_s16816gemm_relu_bf16_256x128_32x3_nn_align8>(T1::Params) | bf16 | false | 16,384 | 5,120 | 5,120 | 1 | 16,384 | null | null | null | null | filler_sq5k | 15.25 | 256 | 3,072 | 1,320,000,000 | 215 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 4,096 | null | 1 | 1 | null | null | null | 4,096 | reduce | 0.00576 | 512 | 1 | 64,380 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 5,120 | null | 1 | 1 | null | null | null | 5,120 | reduce | 0.00506 | 512 | 1 | 64,130 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 6,144 | null | 1 | 1 | null | null | null | 6,144 | reduce | 0.00538 | 512 | 1 | 66,180 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 8,192 | null | 1 | 1 | null | null | null | 8,192 | reduce | 0.0056 | 512 | 1 | 66,300 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 10,240 | null | 1 | 1 | null | null | null | 10,240 | reduce | 0.0056 | 512 | 1 | 68,480 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 12,288 | null | 1 | 1 | null | null | null | 12,288 | reduce | 0.00541 | 512 | 1 | 68,350 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 1 | 16,384 | null | 1 | 1 | null | null | null | 16,384 | reduce | 0.0057 | 512 | 1 | 72,320 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 4,096 | null | 1 | 2 | null | null | null | 8,192 | reduce | 0.00573 | 512 | 1 | 72,320 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 5,120 | null | 1 | 2 | null | null | null | 10,240 | reduce | 0.00611 | 512 | 1 | 76,420 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 6,144 | null | 1 | 2 | null | null | null | 12,288 | reduce | 0.00589 | 512 | 1 | 78,460 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 8,192 | null | 1 | 2 | null | null | null | 16,384 | reduce | 0.00595 | 512 | 1 | 80,510 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 10,240 | null | 1 | 2 | null | null | null | 20,480 | reduce | 0.00611 | 512 | 1 | 80,510 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 12,288 | null | 1 | 2 | null | null | null | 24,576 | reduce | 0.00653 | 512 | 1 | 88,700 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 2 | 16,384 | null | 1 | 2 | null | null | null | 32,768 | reduce | 0.00646 | 512 | 1 | 88,830 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 4,096 | null | 1 | 4 | null | null | null | 16,384 | reduce | 0.00582 | 512 | 1 | 72,320 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 5,120 | null | 1 | 4 | null | null | null | 20,480 | reduce | 0.00595 | 512 | 1 | 72,450 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 6,144 | null | 1 | 4 | null | null | null | 24,576 | reduce | 0.0063 | 512 | 1 | 76,420 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 8,192 | null | 1 | 4 | null | null | null | 32,768 | reduce | 0.00611 | 512 | 1 | 76,420 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 10,240 | null | 1 | 4 | null | null | null | 40,960 | reduce | 0.00614 | 512 | 1 | 80,640 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 12,288 | null | 1 | 4 | null | null | null | 49,152 | reduce | 0.00634 | 512 | 1 | 80,640 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 4 | 16,384 | null | 1 | 4 | null | null | null | 65,536 | reduce | 0.00602 | 512 | 2 | 88,700 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 4,096 | null | 1 | 8 | null | null | null | 32,768 | reduce | 0.00586 | 512 | 2 | 88,700 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 5,120 | null | 1 | 8 | null | null | null | 40,960 | reduce | 0.00624 | 512 | 2 | 96,900 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 6,144 | null | 1 | 8 | null | null | null | 49,152 | reduce | 0.0064 | 512 | 2 | 96,900 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 8,192 | null | 1 | 8 | null | null | null | 65,536 | reduce | 0.0065 | 512 | 2 | 105,340 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 10,240 | null | 1 | 8 | null | null | null | 81,920 | reduce | 0.00627 | 512 | 2 | 105,090 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 12,288 | null | 1 | 8 | null | null | null | 98,304 | reduce | 0.00669 | 512 | 2 | 121,470 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 8 | 16,384 | null | 1 | 8 | null | null | null | 131,072 | reduce | 0.00688 | 512 | 2 | 121,470 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 4,096 | null | 1 | 16 | null | null | null | 65,536 | reduce | 0.00675 | 512 | 1 | 88,700 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 5,120 | null | 1 | 16 | null | null | null | 81,920 | reduce | 0.00656 | 512 | 1 | 88,700 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 6,144 | null | 1 | 16 | null | null | null | 98,304 | reduce | 0.00688 | 512 | 1 | 97,020 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 8,192 | null | 1 | 16 | null | null | null | 131,072 | reduce | 0.0071 | 512 | 1 | 97,020 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 10,240 | null | 1 | 16 | null | null | null | 163,840 | reduce | 0.00752 | 512 | 1 | 105,090 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 12,288 | null | 1 | 16 | null | null | null | 196,608 | reduce | 0.00746 | 512 | 1 | 105,220 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 16 | 16,384 | null | 1 | 16 | null | null | null | 262,144 | reduce | 0.00589 | 512 | 4 | 121,470 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 32 | 4,096 | null | 1 | 32 | null | null | null | 131,072 | reduce | 0.00608 | 512 | 4 | 121,600 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 32 | 5,120 | null | 1 | 32 | null | null | null | 163,840 | reduce | 0.0064 | 512 | 4 | 137,860 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 32 | 6,144 | null | 1 | 32 | null | null | null | 196,608 | reduce | 0.00624 | 512 | 4 | 137,980 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 32 | 8,192 | null | 1 | 32 | null | null | null | 262,144 | reduce | 0.00637 | 512 | 4 | 154,240 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 32 | 10,240 | null | 1 | 32 | null | null | null | 327,680 | reduce | 0.00646 | 512 | 4 | 154,370 | 32 |
misc_sweep | RTX3090 | misc_sweep | reduce | void reduce_kernel<512, 1, ReduceOp<c10::BFloat16, func_wrapper_t<c10::BFloat16, sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | 32 | 12,288 | null | 1 | 32 | null | null | null | 393,216 | reduce | 0.00691 | 512 | 4 | 187,010 | 32 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.