Module attention 

Prefill/decode attention kernels and related attention variants.

Structs

CsaCompressConfig

CsaLightningIndexerConfig

CsaSharedMqaConfig

CsaTopkSelectorConfig

HcaConfig

MsaBlockMaxConfig

MsaGatheredGqaConfig

MsaSelectTopkBlocksConfig

MsaSelectedTokenPositionsConfig

SlidingWindowAttentionConfig

Functions

attention_sink_decode_f32

attention_sink_prefill_f32

compressed_sparse_attention_compress_f32

compressed_sparse_attention_lightning_indexer_f32

compressed_sparse_attention_shared_mqa_f32

compressed_sparse_attention_topk_selector_f32

conv_output_len

fmha_prefill_lse_f32

fused_neighborhood_attention_f32

heavily_compressed_attention_compress_f32

heavily_compressed_attention_f32

minimax_sparse_attention_block_max_f32

minimax_sparse_attention_gathered_gqa_f32

minimax_sparse_attention_select_topk_blocks_f32

minimax_sparse_attention_selected_token_positions_i32

mla_decode_lse_f32

mla_prefill_f32

multi_token_attention_f32

paged_kv_decode_attention_f32

paged_kv_prefill_attention_f32

paged_mla_decode_attention_f32

ragged_kv_prefill_attention_f32

sliding_window_attention

Causal left-window attention with contiguous grouped-query head mapping.

sliding_window_attention_f32

softcapped_window_attention_f32

softcapped_window_decode_lse_f32

sparse_mla_prefill_f32

sparsemax_tau

splitk_reduce_f32