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

rms_norm_fused_quantize_dynamic_scaled_fp8

def rms_norm_fused_quantize_dynamic_scaled_fp8[in_dtype: DType, out_dtype: DType, scale_dtype: DType, rank: Int, InputFn: def[width: Int, alignment: Int, coord_rank: Int](IndexList[coord_rank]) -> SIMD[in_dtype, width] & RegisterPassable & ImplicitlyCopyable, OutputFn: def[width: SIMDLength, rank: Int, alignment: Int](IndexList[rank], SIMD[out_dtype, width]) -> None & RegisterPassable & ImplicitlyCopyable, ScaleFn: def[coord_rank: Int](IndexList[coord_rank], Scalar[scale_dtype]) -> None & RegisterPassable & ImplicitlyCopyable, AxisSizeT: CoordLike, /, target: StringSpan[ImmStaticOrigin], reduce_dim: Int = (rank - Int(1))](input_fn: InputFn, output_fn: OutputFn, scale_fn: ScaleFn, shape: Coord, axis_size: AxisSizeT, gamma: TileTensor[in_dtype, Storage=gamma.Storage, address_space=gamma.address_space, linear_idx_type=gamma.linear_idx_type], epsilon: Scalar[in_dtype], weight_offset: Scalar[in_dtype], scale_ub: Float32, context: Optional[DeviceContext] = None) where (eq InputFn.in_dtype, in_dtype) where (eq OutputFn.out_dtype, out_dtype) where (eq ScaleFn.scale_dtype, scale_dtype)