For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

selective_scan_fwd_minimal_shape

def selective_scan_fwd_minimal_shape[dtype: DType](u: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=u.static_spec], delta: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=delta.static_spec], A: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=A.static_spec], B: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=B.static_spec], C: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=C.static_spec]) -> IndexList[Int(3)]

Returns the output shape for the selective_scan_fwd_minimal op.

The output of the minimal selective scan forward pass has the same shape as the input sequence u: (batch, dim, seqlen).

Args:

u (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=u.static_spec]): Input sequence tensor with shape (batch, dim, seqlen).
delta (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=delta.static_spec]): Time-step tensor with shape (batch, dim, seqlen).
A (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=A.static_spec]): State transition matrix with shape (dim, dstate).
B (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=B.static_spec]): Input projection with shape (batch, n_groups, dstate, seqlen).
C (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=C.static_spec]): Output projection with shape (batch, n_groups, dstate, seqlen).

Returns:

IndexList[Int(3)]: The shape of the output tensor (batch, dim, seqlen).