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
RMSNormQuantizeDynamicScaledFloat8
struct RMSNormQuantizeDynamicScaledFloat8
Registers the mo.rms_norm_quantize_dynamic_scaled_float8 graph op: mo.reduce.rms_norm fused with a group-scaled (blockwise-FP8) mo.quantize_dynamic_scaled_float8. Same outputs and scale layout as the unfused pair; group_size_or_per_token must be a proper divisor of the row (per-token scales stay on the rms_norm + quantize composite).
Implemented traitsโ
Methodsโ
executeโ
static def execute[input_type: DType, scales_type: DType, output_type: DType, //, group_size_or_per_token: Int, multiply_before_cast: Bool, target: StringSpan[ImmStaticOrigin]](output: ManagedTensorSlice[IOSpec[_, _].Output, static_spec=output.static_spec], scales: ManagedTensorSlice[IOSpec[_, _].Output, static_spec=scales.static_spec], input: ManagedTensorSlice[IOSpec[_, _].FusedInput, static_spec=input.static_spec], gamma: ManagedTensorSlice[IOSpec[_, _].Input, static_spec=gamma.static_spec], epsilon: Float32, weight_offset: Scalar[input_type], scale_ub: Float32, ctx: DeviceContextBase)