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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00672 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 ... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.0072 | 512 | 256 | 2,370,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 ... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00707 | 512 | 256 | 2,370,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00682 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00701 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00682 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00682 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00678 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00694 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 ... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00714 | 512 | 256 | 2,370,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 ... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.0071 | 512 | 256 | 2,370,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00717 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00688 | 512 | 16 | 671,620 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00701 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00682 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00685 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, float, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00691 | 512 | 16 | 671,490 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 ... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.0071 | 512 | 256 | 2,370,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | splitk_reduce | void cublasLt::splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, 0, __nv_bfloat16, __nv_bfloat16, __nv_bfloat16, 1, 0, 0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 ... | bf16 | false | 128 | 852 | 5,120 | null | null | null | null | null | null | null | 0.00704 | 512 | 256 | 2,370,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01645 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01306 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01626 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01341 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01626 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01344 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01638 | 128 | 96 | 2,170,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01318 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01648 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01347 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01674 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01325 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01648 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01331 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01654 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01309 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01622 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01334 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01632 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.0136 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01664 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01299 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01638 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01302 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01661 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01309 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01645 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01366 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01645 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01341 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 1, 0, 0, 0, 1, 0, 1, 0>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.0169 | 128 | 96 | 2,160,000 | 255 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | flash_attn | void pytorch_flash::flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<256, 64, 64, 4, 0, 0, cutlass::bfloat16_t, Flash_kernel_traits<256, 64, 64, 4, cutlass::bfloat16_t>>, 4, 1, 1>(pytorch_flash::Flash_fwd_params) | bf16 | false | null | null | null | 1 | 128 | 24 | 213 | 4 | null | flash_attn | 0.01312 | 128 | 768 | 6,340,000 | 54 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | copy | void at::vectorized_gather_kernel<16, long>(char *, char *, T2 *, int, long, long, long, long, bool) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 654,336 | other | 0.01066 | 256 | 384 | 1,330,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<2, at::CUDAFunctorOnSelf_add<long>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | residual | 0.00416 | 128 | 1 | 7,040 | 36 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::unrolled_elementwise_kernel<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 11)]::operator ()() const::[lambda(bool) (instance 1)], std::array<char *, 2>, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned in... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | cast | 0.00451 | 128 | 1 | 34,940 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | reduce | void at::reduce_kernel<512, 1, at::ReduceOp<bool, at::func_wrapper_t<bool, at::and_kernel_cuda(at::TensorIterator &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 12)]::operator ()() const::[lambda(bool, bool) (instance 1)]>, unsigned int, bool, 4, 4>>(T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 128 | other | 0.00608 | 32 | 1 | 54,140 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::AUnaryFunctor<long, long, bool, at::<unnamed>::CompareEqFunctor<long>>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | other | 0.00355 | 128 | 1 | 7,550 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | reduce | void at::reduce_kernel<512, 1, at::ReduceOp<bool, at::func_wrapper_t<bool, at::and_kernel_cuda(at::TensorIterator &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 12)]::operator ()() const::[lambda(bool, bool) (instance 1)]>, unsigned int, bool, 4, 4>>(T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 128 | other | 0.00589 | 32 | 1 | 54,020 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(long) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 1,024 | cast | 0.00746 | 128 | 2 | 11,390 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::unrolled_elementwise_kernel<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned in... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | cast | 0.00678 | 128 | 1 | 35,460 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | cast | 0.00506 | 128 | 6 | 15,740 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 2,560 | cast | 0.00512 | 128 | 5 | 15,230 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | copy | void at::<unnamed>::CatArrayBatchedCopy<at::<unnamed>::OpaqueType<4>, unsigned int, 3, 64, 64>(T1 *, at::<unnamed>::CatArrInputTensorMetadata<T1, T2, T4, T5>, at::<unnamed>::TensorSizeStride<T2, 4>, int, T2) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 655,360 | other | 0.00758 | 512 | 432 | 22,140 | 30 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::cos_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 4,096 | rope | 0.00557 | 128 | 8 | 72,830 | 30 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::AUnaryFunctor<float, float, float, binary_internal::MulFunctor<float>>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 4,096 | mul | 0.00355 | 128 | 8 | 37,630 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::sin_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 4,096 | rope | 0.00496 | 128 | 8 | 72,450 | 30 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::AUnaryFunctor<float, float, float, binary_internal::MulFunctor<float>>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 4,096 | mul | 0.00355 | 128 | 8 | 37,500 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::vectorized_elementwise_kernel<4, at::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 4,096 | cast | 0.00352 | 128 | 8 | 37,760 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::vectorized_elementwise_kernel<4, at::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 4,096 | cast | 0.00349 | 128 | 8 | 37,890 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::unrolled_elementwise_kernel<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned in... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 655,360 | cast | 0.00992 | 128 | 1,280 | 1,340,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, void at::<unnamed>::pow_tensor_scalar_kernel_impl<float, float>(at::TensorIteratorBase &, T2)::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 218,112 | rmsnorm | 0.00614 | 128 | 640 | 2,630,000 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | reduce | void at::reduce_kernel<512, 1, at::ReduceOp<float, at::MeanOps<float, float, float, float>, unsigned int, float, 4, 4>>(T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 16,384 | rmsnorm | 0.03203 | 512 | 8 | 2,670,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::CUDAFunctorOnSelf_add<float>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | residual | 0.00333 | 128 | 1 | 5,250 | 30 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::rsqrt_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | rmsnorm | 0.0041 | 128 | 1 | 7,040 | 22 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast<at::BinaryFunctor<float, float, float, binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 1,310,720 | mul | 0.00835 | 128 | 2,560 | 2,630,000 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::unrolled_elementwise_kernel<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned in... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 5,120 | cast | 0.00781 | 128 | 10 | 42,370 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::CUDAFunctorOnSelf_add<float>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 2,560 | residual | 0.00349 | 128 | 5 | 25,220 | 30 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast<at::BinaryFunctor<float, float, float, binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 1,310,720 | mul | 0.00854 | 128 | 2,560 | 2,650,000 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::vectorized_elementwise_kernel<4, at::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 218,112 | cast | 0.00582 | 128 | 640 | 2,630,000 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::FillFunctor<c10::BFloat16>, std::array<char *, 1>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 20,480 | fill | 0.00371 | 128 | 40 | 4,480 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::elementwise_kernel<128, 4, void at::gpu_kernel_impl_nocast<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 12)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 1,310,720 | cast | 0.02048 | 128 | 2,560 | 2,640,000 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::elementwise_kernel<128, 4, void at::gpu_kernel_impl_nocast<at::<unnamed>::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(i... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 1,310,720 | silu | 0.01254 | 128 | 2,560 | 2,700,000 | 19 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::sigmoid_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | other | 0.00509 | 128 | 6 | 25,220 | 30 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::unrolled_elementwise_kernel<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned in... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | cast | 0.00432 | 128 | 1 | 32,260 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::exp_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | other | 0.00378 | 128 | 1 | 7,940 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 512 | neg | 0.00384 | 128 | 1 | 5,250 | 28 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::unrolled_elementwise_kernel<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>, 4, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned in... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 6,144 | cast | 0.00771 | 128 | 12 | 44,420 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::elementwise_kernel<128, 4, void at::gpu_kernel_impl<at::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 6,144 | residual | 0.00896 | 128 | 12 | 86,660 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::<unnamed>::softplus_kernel(at::TensorIteratorBase &, const c10::Scalar &, const c10::Scalar &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | other | 0.00595 | 128 | 6 | 46,720 | 24 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::elementwise_kernel<128, 2, void at::gpu_kernel_impl_nocast<at::BinaryFunctor<float, float, float, binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 12,288 | mul | 0.00512 | 128 | 24 | 35,840 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::elementwise_kernel<128, 4, void at::gpu_kernel_impl_nocast<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 12)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 655,360 | cast | 0.01437 | 128 | 1,536 | 542,080 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | cast | void at::elementwise_kernel<128, 4, void at::gpu_kernel_impl_nocast<at::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 12)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>... | bf16 | false | null | null | null | 1 | 128 | null | null | null | 655,360 | cast | 0.01389 | 128 | 1,536 | 542,080 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::BinaryFunctor<c10::BFloat16, c10::BFloat16, c10::BFloat16, binary_internal::MulFunctor<float>>, std::array<char *, 3>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 654,336 | mul | 0.00528 | 128 | 768 | 1,580,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | reduce | void at::reduce_kernel<512, 1, at::ReduceOp<c10::BFloat16, at::func_wrapper_t<c10::BFloat16, at::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(at::TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 654,336 | other | 0.00845 | 512 | 192 | 1,630,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::CUDAFunctorOnSelf_add<c10::BFloat16>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | residual | 0.00352 | 128 | 6 | 18,050 | 26 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::rsqrt_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 3)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | rmsnorm | 0.00358 | 128 | 6 | 19,070 | 22 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::elementwise_kernel<128, 4, void at::gpu_kernel_impl_nocast<at::BinaryFunctor<c10::BFloat16, c10::BFloat16, c10::BFloat16, binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 655,360 | mul | 0.00842 | 128 | 1,536 | 1,600,000 | 16 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::BinaryFunctor<c10::BFloat16, c10::BFloat16, c10::BFloat16, binary_internal::MulFunctor<float>>, std::array<char *, 3>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 654,336 | mul | 0.00528 | 128 | 768 | 1,580,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | reduce | void at::reduce_kernel<512, 1, at::ReduceOp<c10::BFloat16, at::func_wrapper_t<c10::BFloat16, at::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator ()(at::TensorIterator &)::[lambda(float, float) (instance 1)]>, unsigned int, c10::BFloat16, 4, 8>>(T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 654,336 | other | 0.00787 | 512 | 192 | 1,630,000 | 32 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::CUDAFunctorOnSelf_add<c10::BFloat16>, std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | residual | 0.00346 | 128 | 6 | 18,050 | 26 |
Qwen3.5-27B_prefill | A100 | Qwen3.5-27B | elementwise | void at::vectorized_elementwise_kernel<4, at::rsqrt_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 3)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array<char *, 2>>(int, T2, T3) | bf16 | false | null | null | null | 1 | 128 | null | null | null | 3,072 | rmsnorm | 0.00352 | 128 | 6 | 19,070 | 22 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.