vllm_omni.attention.fish_kvcache_triton ¶

fish_decode_kvcache_attn_triton(
    query: Tensor,
    key_cache: Tensor,
    value_cache: Tensor,
    block_table: Tensor,
    seq_lens: Tensor,
    out: Tensor,
    *,
    scale: float,
    max_seq_len: int,
    small_path_max_seq_len: int,
    long_split_tokens: int,
    partial_m: Tensor,
    partial_l: Tensor,
    partial_acc: Tensor,
) -> Tensor

is_available() -> bool

load_error() -> Exception | None