def linear(g, q_input, q_weight, bias, op_scale, op_zero_point): input, input_scale, _ = sym_help.dequantize_helper(g, q_input) weight, weight_scale, _ = sym_help.dequantize_helper(g, q_weight) q_bias = sym_help.requantize_bias_helper(g, bias, input_scale, weight_scale) bias, _, _ = sym_help.dequantize_helper(g, q_bias) output = linear(g, input, weight, bias) return sym_help.quantize_helper(g, output, op_scale, op_zero_point)
def linear(g, q_input, q_weight, bias, op_scale, op_zero_point): input, input_scale, _, _ = symbolic_helper.dequantize_helper(g, q_input) weight, weight_scale, _, axis = symbolic_helper.dequantize_helper(g, q_weight) q_bias = symbolic_helper.requantize_bias_helper( g, bias, input_scale, weight_scale, axis ) bias, _, _, _ = symbolic_helper.dequantize_helper(g, q_bias) output = opset9.linear(g, input, weight, bias) return symbolic_helper.quantize_helper(g, output, op_scale, op_zero_point)