fp8_gemm_blockwise

paddle.incubate.nn.functional. fp8_gemm_blockwise ( a, a_decode_scale, b, b_decode_scale, out_dtype, out: Optional[Tensor] = None, bias: Optional[Tensor] = None, accumulate: bool = False, use_split_accumulator: bool = True, is_a_1d_scaled: bool = True, is_b_1d_scaled: bool = True ) [source]