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
Struct_chunked_qkv_rope_store_ragged_paged
struct Struct_chunked_qkv_rope_store_ragged_paged[qk_head_dim: Int, v_head_dim: Int, chunk_padding: Int = Int(0)]
Registers chunked QKV unpacking, partial NeoX RoPE, and paged stores.
The execute method receives value_scale as a CPU Float32 operand and applies it to V before rounding to BF16.
Parameters
- qk_head_dim (
Int): Unpadded query and key head width. - v_head_dim (
Int): Unpadded value head width. - chunk_padding (
Int): Trailing projection rows to ignore in each chunk.
Implemented traits
Methods
execute
static def execute[cache_dtype: DType, //, target: StaticString](output: ManagedTensorSlice[IOSpec[_, _].Output, static_spec=output.static_spec], qkv: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=qkv.static_spec], input_row_offsets: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=input_row_offsets.static_spec], freqs_cis: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=freqs_cis.static_spec], kv_blocks: ManagedTensorSlice[IOSpec[_, _].MutableInput, static_spec=kv_blocks.static_spec], page_stride: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=page_stride.static_spec], cache_lengths: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=cache_lengths.static_spec], kv_lookup_table: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=kv_lookup_table.static_spec], max_prompt_length: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=max_prompt_length.static_spec], max_cache_length: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=max_cache_length.static_spec], layer_idx: UInt32, value_scale: Float32, ctx: DeviceContext)