if needHash {
						sum, err := dirhash.HashZip(zipfile, dirhash.Hash1)
						if err != nil {
							t.Errorf("repo.Zip(%q): %v", tt.version, err)
						} else if sum != tt.zipSum {
							t.Errorf("repo.Zip(%q): got file with sum %q, want %q", tt.version, sum, tt.zipSum)
						} else if zipFileHash := hex.EncodeToString(h.Sum(nil)); zipFileHash != tt.zipFileHash {

// RUN: tf-opt -split-input-file -verify-diagnostics -tf-einsum %s | FileCheck %s

func.func @unary_einsum_reduce_sum_transpose(%arg0: tensor<3x4x5x6xf32>) -> tensor<3x5x4xf32> {
  %0 = "tf.Einsum"(%arg0) {T = "tfdtype$DT_FLOAT", equation = "...gse->...sg"}: (tensor<3x4x5x6xf32>) -> tensor<3x5x4xf32>
  func.return %0 : tensor<3x5x4xf32>
  // CHECK-LABEL: unary_einsum_reduce_sum_transpose
  // CHECK-DAG: %[[cst:.*]] = arith.constant dense<3> : tensor<1xi32>

}

struct EinsumDimensionNumbers {
  // Each field contains the list of dimensions appearing only in the specifed
  // arguments of the einsum op with natural ordering. For example `rhs_out`
  // contains the dimensions appearing in the RHS and the OUTPUT of the einsum
  // but not in the LHS.
  std::vector<int64_t> lhs;
  std::vector<int64_t> rhs;
  std::vector<std::tuple<int64_t, int64_t>> lhs_rhs;

  // Claim: For the einsum operation z = einsum("{eq_x},{eq_y}->{eq_z}", x, y),
  //   where the equation involves only Tensor contractions, generalized traces
  //   and transposes, the input gradients are given by the vector-jacobian
  //   products (VJPs):
  //
  //     grad_wrt_x = einsum("{eq_y},{eq_z}->{eq_x}", y, grad_wrt_z)
  //     grad_wrt_y = einsum("{eq_x},{eq_z}->{eq_y}", x, grad_wrt_z}
  //

  [], (addBenefit 10)>;

// Converts inlined Einsum pattern to TF XlaDotV2 op.
def ConvertTFEinsumToXLADotV2Op : Pat<
  (TF_EinsumOp:$einsum
    $args, $equation),
  (CreateXlaDotV2OpFromTfEinsumOp
    $equation, $args, $einsum),
  [(IsInt32ElementType $einsum),
   // Constraint to check:
   // 1. The einsum has two inputs and one output.
   // 2. The einsum is not created by the convert function itself.

  if (!value_type.hasRank()) return false;
  if (!value_type.getElementType().isInteger(integer_width)) return false;

  return true;
}

// Constraint to check:
// 1. The einsum has two inputs and one output.
// 2. The einsum is not created by the convert function itself.
// 3. Both inputs are int32 tensor.
// 4. Both inputs have the graph ancestor of either const-(sub), or cast-sub.

        // Check that each directory in hierarchy is present
        file.withInputStream { InputStream instr ->
            def dirs = [] as Set
            def zipStr = new ZipInputStream(instr)
            def entry
            while (entry = zipStr.getNextEntry()) {
                if (entry.directory) {
                    assertTrue("Duplicate directory '$entry.name'", dirs.add(entry.name))
                }

        settingsFile """
            include "dist"
            include "lib"
        """

        file("dist/build.gradle").text = """
            abstract class ZipSrc extends DefaultTask {
                @Internal
                int countResolved

                @Internal
                abstract DirectoryProperty getSources()

                @InputFiles

            tensor_spec.TensorSpec(
                shape=shape, dtype=dtypes.float32, name='input_tensor'
            )
        ),
    )
    return model

  # Prepares sample einsum input data shapes.
  # This function returns:
  # 1. Shape for input 1
  # 2. Shape for input 2
  # 3. Shape for bias
  # 4. Signature for input 1 (Could contain None dimension)

      }

      if (!is_weight_constant) {
        if (!function_name.contains("matmul") &&
            !function_name.contains("einsum")) {
          return absl::InternalError(
              "Non-constant weights are not supported at the moment,"
              " except matmul and einsum.");
        } else if (!quant_options_.enable_two_input_tensors() &&
                   !is_unitwise_quantization_enabled) {