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

composite_layer_norm_rope_ragged_shape

def composite_layer_norm_rope_ragged_shape[dtype: DType, freq_dtype: DType, rank: Int](input: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=input.static_spec], gamma: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=gamma.static_spec], beta: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=beta.static_spec], epsilon: Float32, input_row_offsets: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=input_row_offsets.static_spec], start_pos: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=start_pos.static_spec], freqs_cis: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=freqs_cis.static_spec]) -> IndexList[rank]

Computes the output shape for the mo.composite.layer_norm_rope_ragged graph op.

Parameters:

• ​dtype (DType): Element type of the input, gamma, and beta tensors.
• ​freq_dtype (DType): Element type of the freqs_cis RoPE table.
• ​rank (Int): Number of dimensions in the input and output tensors.

Args:

• ​input (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=input.static_spec]): Activation tensor normalized by LayerNorm then partially rotated by ragged RoPE.
• ​gamma (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=gamma.static_spec]): Per-column scale weights applied after normalization.
• ​beta (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=beta.static_spec]): Per-column shift weights applied after scaling.
• ​epsilon (Float32): Small constant added inside the normalization variance for numerical stability.
• ​input_row_offsets (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=input_row_offsets.static_spec]): Ragged batch boundaries.
• ​start_pos (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=start_pos.static_spec]): Per-sequence cache length used for the RoPE position lookup.
• ​freqs_cis (ManagedTensorSlice[IOSpec[_, _].Input, static_spec=freqs_cis.static_spec]): RoPE frequency table; its width sets the rotated prefix.

Returns:

IndexList[rank]: The output shape, which matches the input shape.