mlir::cast<quant::QuantizedType>(non_bias_type.getElementType());
      non_bias_types.push_back(non_bias_ele_type);
    } else {
      // The non-bias hasn't been quantized, let's skip this bias.
      break;
    }
  }
  // The non-bias hasn't been quantized, let's skip this bias.
  if (non_bias_types.size() != non_biases.size()) return {};

  return func(/*op_types=*/non_bias_types, /*adjusted_quant_dim=*/-1,

    }
    const int64_t cost_per_col = 2 * weight_type.getNumElements();

    *count = cost_per_col * cols;

    auto bias = op->getOperand(2);
    if (bias) {
      auto bias_type =
          mlir::dyn_cast_or_null<mlir::RankedTensorType>(bias.getType());
      if (bias_type && bias_type.hasStaticShape()) {
        *count += output_type.getNumElements();
      }
    }

    return true;
  }

def FoldQuantWeightsIntoTposeConv : Pat<
  (TFL_TransposeConvOp
    $output_shape,
    (TFL_DequantizeOp $quant_weights),
    $quant_input,
    $bias, $padding, $stride_h, $stride_w, $faf),
  (TFL_TransposeConvOp $output_shape, $quant_weights,
    $quant_input, $bias, $padding, $stride_h, $stride_w, $faf), 

  %weight = arith.constant dense_resource<__elided__> : tensor<2x3x3x2xf32>
  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

  %q_input= "quantfork.qcast"(%input) : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3x!quant.uniform<i8:f32, 0.58810077742034317:-128>>

// RUN: json_to_flatbuffer %p/test_schema.fbs %s | flatbuffer_translate --tflite-flatbuffer-to-mlir -o - | FileCheck %s

// This test is to test that if the flatbuffer omits the last optional input `bias` of tfl.conv_2d op, the flatbuffer_importer will automatically adds `none` value to tfl.conv_2d.

// CHECK: %[[CST:.*]] = "tfl.no_value"() <{value}> : () -> none

    }

    // If all quantized or floating point then types are consistent.
    // Int is valid in combination with both quantized and floating point.
    // This occurs when doing qi16 convolution, as bias is passed as a
    // non-quantized int64
    if (allTypesFp || allTypesQuantizedOrInt) return failure();

    Location loc = op->getLoc();
    SmallVector<Value> newOperands;
    newOperands.reserve(op->getNumOperands());

                                         ArrayRef<Value> results);

// Add the second argument to the first argument, which is expected to be an
// argument list.
// Used to attach bias to einsum argument list.
SmallVector<Value> AppendToVector(ArrayRef<Value> arguments, Value append);

// Checks if the `Method` attatched to the given `tf.XlaCallModule` op has
// `WeightOnlyPtq`.

      }

      // Rewire the outputs.
      result.replaceAllUsesWith(new_result);
    }

    // Remove the old op.
    op->erase();
  });
}

// Fold quantized i32 (normally bias) into their float values.
struct FoldQuantizedI32ToFloat : public OpRewritePattern<TFL::DequantizeOp> {
  using OpRewritePattern<TFL::DequantizeOp>::OpRewritePattern;

    auto fc_op = rewriter.create<TFL::FullyConnectedOp>(
        bmm_op->getLoc(), ArrayRef<Type>{output_type},
        /*input=*/output_lhs, /*filter=*/output_rhs, /*bias=*/no_input,
        /*fused_activation_function=*/rewriter.getStringAttr("NONE"),
        /*weights_format=*/rewriter.getStringAttr("DEFAULT"),
        /*keep_num_dims=*/rewriter.getBoolAttr(true),