For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

batched_matmul_dynamic_scaled_fp8

def batched_matmul_dynamic_scaled_fp8[c_type: DType, a_type: DType, b_type: DType, a_scales_type: DType, b_scales_type: DType, //, input_scale_granularity: StringSlice[StaticConstantOrigin], weight_scale_granularity: StringSlice[StaticConstantOrigin], m_scale_granularity: Int, n_scale_granularity: Int, k_scale_granularity: Int, transpose_b: Bool = False, target: StringSlice[StaticConstantOrigin] = StringSlice("cpu")](c: TileTensor[c_type, address_space=c.address_space, linear_idx_type=c.linear_idx_type, element_size=c.element_size], a: TileTensor[a_type, address_space=a.address_space, linear_idx_type=a.linear_idx_type, element_size=a.element_size], b: TileTensor[b_type, address_space=b.address_space, linear_idx_type=b.linear_idx_type, element_size=b.element_size], a_scales: TileTensor[a_scales_type, address_space=a_scales.address_space, linear_idx_type=a_scales.linear_idx_type, element_size=a_scales.element_size], b_scales: TileTensor[b_scales_type, address_space=b_scales.address_space, linear_idx_type=b_scales.linear_idx_type, element_size=b_scales.element_size], ctx: DeviceContext)