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
blockwise_scaled_fp8_with_epilogue
def blockwise_scaled_fp8_with_epilogue[c_type: DType, a_type: DType, b_type: DType, a_scales_type: DType, b_scales_type: DType, EpilogueFnType: ElementwiseEpilogueFn = inflated$def[_no_epilogue_body[_, _, _, _, _]], //, *, scales_granularity_mnk: IndexList[Int(3)], transpose_b: Bool = False, has_epilogue_fn: Bool = False](c: TileTensor[c_type, Engine=c.Engine, linear_idx_type=c.linear_idx_type], a: TileTensor[a_type, Engine=a.Engine, linear_idx_type=a.linear_idx_type], b: TileTensor[b_type, Engine=b.Engine, linear_idx_type=b.linear_idx_type], a_scales: TileTensor[a_scales_type, Engine=a_scales.Engine, linear_idx_type=a_scales.linear_idx_type], b_scales: TileTensor[b_scales_type, Engine=b_scales.Engine, linear_idx_type=b_scales.linear_idx_type], ctx: DeviceContextBase, epilogue_fn: EpilogueFnType = no_epilogue_fn)
Our sm100 blockwise scaled fp8 matmul kernel still does not support fusion of elementwise operations. This is a temporary implementation that uses our sm100 blockwise scaled fp8 matmul kernel and dispatch a separate epilogue kernel to apply the elementwise operations. For non B200 GPUs, we use the naive blockwise scaled fp8 matmul which support normal epilogue natively. Callers must allocate c; with has_epilogue_fn the matmul result is written into c and then read back by epilogue_fn.