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 struct

CausalConv1DVarlenFwd

struct CausalConv1DVarlenFwd[activation: StringSlice[ImmStaticOrigin], channels_last: Bool = False]

Varlen causal 1D convolution forward pass.

Performs causal 1D convolution on variable-length sequences that are concatenated together. Uses cumulative sequence lengths to identify sequence boundaries.

The registration lives here in the built-in kernel library (mirroring the gated_delta_conv1d_fwd precedent) so the graph compiler / serve path can resolve the op with no out-of-tree custom_extensions. The kernel math lives in state_space.varlen_causal_conv1d.

The underlying kernels index x/output purely through runtime dim/seqlen strides, so the token-axis memory layout is a free parameter. With channels_last=True the op consumes and produces tokens-major (total_seqlen, dim) tensors — the layout the surrounding graph naturally carries — eliminating the materialized (dim, total_seqlen) transposes on both sides of the op. Only the stride/extent bookkeeping below changes; the per-element compute is identical in both layouts.

Tensor Shapes: - output: (dim, total_seqlen) - Output tensor ((total_seqlen, dim) when channels_last) - x: (dim, total_seqlen) - Input tensor (concatenated sequences) ((total_seqlen, dim) when channels_last) - weight: (dim, width) - Convolution weights per channel - bias: (dim,) - Per-channel bias - query_start_loc: (batch + 1,) - Cumulative sequence lengths - cache_indices: (batch,) - Indices into conv_states (optional) - has_initial_state: (batch,) - Whether to use initial state (optional) - conv_states: (batch, dim, width - 1) - Conv states (optional, in/out)

Parameters

• ​activation (StringSlice[ImmStaticOrigin]): Activation function - "none" or "silu".
• ​channels_last (Bool): If True, x and output are tokens-major (total_seqlen, dim) instead of (dim, total_seqlen).

Implemented traits

AnyType, ImplicitlyDeletable

Methods

execute

static def execute[dtype: DType, target: StringSlice[ImmStaticOrigin]](output: ManagedTensorSlice[IOSpec[_, _].Output, static_spec=output.static_spec], x: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=x.static_spec], weight: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=weight.static_spec], bias: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=bias.static_spec], conv_states: ManagedTensorSlice[IOSpec[_, _].MutableInput, static_spec=conv_states.static_spec], query_start_loc: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=query_start_loc.static_spec], cache_indices: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=cache_indices.static_spec], has_initial_state: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=has_initial_state.static_spec], ctx: DeviceContext)