# Change the value of CACHEBUSTER when upgrading the toolchain, or when testing
# different compilation methods. E.g. for a PR to test a new CUDA version, set
# the CACHEBUSTER to the PR number.
build --action_env=CACHEBUSTER=20220325

# Use Python 3.X as installed in container image
build --action_env PYTHON_BIN_PATH="/usr/local/bin/python3"
build --python_path="/usr/local/bin/python3"

    forward_op_->outputs.push_back(retvals[i]);
  }
  // TODO(b/166669239): This is needed to support AttrBuilder::Get for string
  // attributes. Number type attrs and DataType attrs work fine without this.
  // Consider getting rid of this and making the behavior between number types
  // and string consistent.
  forward_op_->attrs.BuildNodeDef();
  std::unique_ptr<GradientFunction> gradient_fn;

Instead, an independent Python toolchain is registered, ensuring the right
dependencies are always used. \
See https://github.com/bazelbuild/rules_python/ for more details.

### Specifying the Python version

Note: Only a number of minor Python versions are supported at any given time.

By default, the lowest supported version is used.

To set a different version, use the `TF_PYTHON_VERSION` environment variable,
e.g.

```

# Change the value of CACHEBUSTER when upgrading the toolchain, or when testing
# different compilation methods. E.g. for a PR to test a new CUDA version, set
# the CACHEBUSTER to the PR number.
build --action_env=CACHEBUSTER=20220325

# Use Python 3.X as installed in container image
build --action_env PYTHON_BIN_PATH="/usr/local/bin/python3"
build --python_path="/usr/local/bin/python3"

    TF_Status* status);

// ----------------------------  PJRT  -----------------------------------------
// Passes the pointer to a vector of PJRT_NamedValue and number of options to
// set options for creating a PJRT client. Passes nullptr for create_options and
// 0 for num_options if no options need to be set. You can use
// ConvertToPjRtNamedValueList in

  if (input_arg.number_attr().empty()) {
    status->status = tensorflow::errors::NotFound(
        op_name, " does not have number_attr() defined.");
    return nullptr;
  }

  // The returned string is owned by OpRegistry, so liveness is not a concern.
  return input_arg.number_attr().c_str();
}

int TF_OpIsStateful(const char* op_type, TF_Status* status) {

        InvalidArgument("Node ", node->name(), " was not found in the graph");
    return -1;
  }

  tensorflow::shape_inference::ShapeHandle shape = ic->output(output.index);

  // Unknown rank means the number of dimensions is -1.
  if (!ic->RankKnown(shape)) {
    return -1;
  }

  return ic->Rank(shape);
}

void TF_GraphGetTensorShape(TF_Graph* graph, TF_Output output, int64_t* dims,

*   Fixes a number of `CHECK`-failures in `MapStage`
    ([CVE-2022-21734](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2022-21734))
*   Fixes a division by zero in `FractionalMaxPool`
    ([CVE-2022-21735](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2022-21735))
*   Fixes a number of `CHECK`-fails when building invalid/overflowing tensor
    shapes

// executed in sequence. Assigning nodes to different executors allows executing
// nodes in parallel.
// in_flight_nodes_limit: when is_async is true, this value controls the
// maximum number of in flight async nodes. Enqueuing of additional async ops
// after the limit is reached blocks until some inflight nodes finishes.
// The effect is bounding the memory held by inflight TensorHandles that are

// into python now.
template <typename Gradient, typename BackwardFunction, typename TapeTensor>
class VSpace {
 public:
  virtual ~VSpace() {}

  // Returns the number of elements in the gradient tensor.
  virtual int64_t NumElements(Gradient* tensor) const = 0;

  // Consumes references to the tensors in the gradient_tensors list and returns
  // a tensor with the result.