Searched defs:max_seqlen_k (Results 1 – 6 of 6) sorted by relevance
/aosp_15_r20/external/pytorch/aten/src/ATen/native/transformers/cuda/flash_attn/ |
H A D | mask.h | 39 const int max_seqlen_k, const int row_idx_offset, in apply_mask_local() 76 const int max_seqlen_k, const int row_idx_offset, in apply_mask_causal() 86 const int col_idx_offset_, const int max_seqlen_k, const int row_idx_offset) in apply_mask_causal_w_idx() 113 const int max_seqlen_k, max_seqlen_q; member 120 : max_seqlen_k(max_seqlen_k) in max_seqlen_k() function
|
H A D | flash_api.cpp | 298 const int num_heads, const int head_size, const int max_seqlen_k, const int max_seqlen_q, in set_params_splitkv() 553 const int max_seqlen_k, in mha_varlen_fwd() 1031 const int max_seqlen_k, // max sequence length to choose the kernel in mha_varlen_bwd()
|
H A D | alibi.h | 20 const int max_seqlen_k, max_seqlen_q; member
|
/aosp_15_r20/external/pytorch/torch/utils/ |
H A D | flop_counter.py | 333 max_seqlen_k, argument 413 max_seqlen_k, argument 512 max_seqlen_k, argument
|
/aosp_15_r20/external/pytorch/aten/src/ATen/native/transformers/cuda/ |
H A D | attention_backward.cu | 264 int64_t max_seqlen_k, in _efficient_attention_backward() 818 int64_t max_seqlen_k = k_t.size(1); in _scaled_dot_product_efficient_attention_backward_cuda() local
|
H A D | attention.cu | 995 int64_t max_seqlen_q = 0, max_seqlen_k = 0; in _efficient_attention_forward() local
|