"SparseSoftmaxCrossEntropyWithLogits"));
  // `gradient_checker` only works with model that returns only 1 tensor.
  // Although, `ops::SparseSoftmaxCrossEntropyWithLogits` returns 2 tensors, the
  // second tensor isn't needed for computing gradient so we could safely drop
  // it.
  outputs[0] = loss;
  backprop->Unref();
  return absl::OkStatus();
}

Status BiasAddModel(AbstractContext* ctx,

void RamFileBlockCache::Trim() {
  while (!lru_list_.empty() && cache_size_ > max_bytes_) {
    RemoveBlock(block_map_.find(lru_list_.back()));
  }
}

/// Move the block to the front of the LRU list if it isn't already there.
void RamFileBlockCache::UpdateLRU(const Key& key,
                                  const std::shared_ptr<Block>& block,
                                  TF_Status* status) {
  absl::MutexLock lock(&mu_);

unless indicated otherwise.

1) Add the new version to the `VERSIONS` variable inside
   `tensorflow/tools/toolchains/python/python_repo.bzl`. \
   While this isn't necessary for running the updater, it is required for
   actually using the new version with Tensorflow.

2) In the `WORKSPACE` file, add the new version to the `python_versions`

  std::vector<AbstractTensorHandle*> temp_outputs(2);
  TF_RETURN_IF_ERROR(
      ops::IdentityN(ctx, inputs, absl::MakeSpan(temp_outputs), "IdentityN"));
  // Although, `ops::IdentityN` returns 2 tensors, the first tensor isn't needed
  // for computing gradient so we could safely drop it.
  outputs[0] = temp_outputs[1];
  temp_outputs[0]->Unref();
  return absl::OkStatus();
}

class CppGradients

        tensorflow::io::JoinPath(root_dir_, path);
    // We have already checked `GetParam().empty()` in
    // `ModularFileSystemTest()`. root_dir_ should contain `GetParam() + "://"`
    // if it isn't empty.
    return translated_name;
  }

  // Converts absolute paths to paths relative to root_dir_.
  StringPiece GetRelativePath(StringPiece absolute_path) {

    s->status = InvalidArgument("TF_WhileParams `cond_output` field isn't set");
    return false;
  }
  for (int i = 0; i < params.ninputs; ++i) {
    if (params.body_outputs[i].oper == nullptr) {
      s->status = InvalidArgument("TF_WhileParams `body_outputs[", i, "]` ",
                                  "field isn't set");
      return false;
    }
  }
  if (params.name == nullptr) {

    read = 0;
  } else if (!absl::SimpleAtoi(content_length->second, &read)) {
    TF_SetStatus(status, TF_UNKNOWN, "Could not get content-length header");
    return -1;
  }
  // `TF_OUT_OF_RANGE` isn't considered as an error. So we clear it here.
  TF_SetStatus(status, TF_OK, "");
  TF_VLog(1, "Successful read of %s @ %u of size: %u", path.c_str(), offset,
          read);
  stream.read(buffer, read);

      const std::function<void(BackwardFunction*)>& backward_function_deleter);

  // Returns true if `Accumulate` is active somewhere above on the stack and
  // there isn't an intervening PushState. This is useful for ordering
  // ForwardAccumulators, where more deeply nested accumulators should not see
  // computations from less deeply nested accumulators.

}

::testing::AssertionResult CompareSubString(int64_t offset, size_t length,
                                            absl::string_view result,
                                            size_t read) {
  // Result isn't a null-terminated string so we have to wrap it inside a
  // `string_view`
  if (length == read && content_view.substr(offset, length) ==
                            absl::string_view(result).substr(0, read))

    which determines whether the SavedModel format stores the Keras model/layer
    call functions. The traced functions allow Keras to revive custom models and
    layers without the original class definition, but if this isn't required the
    tracing can be disabled with the added option.
*   The `tf.keras.mixed_precision` API is now non-experimental. The
    non-experimental API differs from the experimental API in several ways.