For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo function
ssd_chunk_cumsum
def ssd_chunk_cumsum[kernel_dtype: DType, A_LT: TensorLayout, D_LT: TensorLayout, dt_bias_LT: TensorLayout, Engine: TensorEngine, SLT: TensorLayout, SEngine: TensorEngine, DtFn: def[width: Int, alignment: Int](Int, Int) -> SIMD[kernel_dtype, width] & RegisterPassable & ImplicitlyCopyable, dt_softplus: Bool](slots_i: Int32, A: TileTensor[kernel_dtype, A_LT, ImmUntrackedOrigin, Engine=Engine], D: TileTensor[kernel_dtype, D_LT, ImmUntrackedOrigin, Engine=Engine], dt_bias: TileTensor[kernel_dtype, dt_bias_LT, ImmUntrackedOrigin, Engine=Engine], info: TileTensor[.int32, SLT, MutUntrackedOrigin, Engine=SEngine], cs: TileTensor[.float32, SLT, MutUntrackedOrigin, Engine=SEngine], delta_out: TileTensor[.float32, SLT, MutUntrackedOrigin, Engine=SEngine], dt_fn: DtFn)
Writes dt and the inclusive cumulative log2 decay of every chunk.
Grid (slots, nheads), block CHUNK threads, one token each.