Class RnnDescriptor

tensorflow/stream_executor/dnn.h:205–216

Specifies the descriptor for a RNN model. An example use case: The user first creates a model through createRnnDescriptor. The user queries the size of the underlying opaque parameter buffer. The user creates and initializes a parameter buffer of the proper size. The user runs forward and backward operations using this RNN descriptor. Once a while, user queries maintainable weights and bias regio

//       compatible and should used in saving and restoring a model.
//   * The user releases the RNN descriptor when the model is no longer in use.
class RnnDescriptor {
 public:
  struct ParamsRegion {
    int64 offset;
    int64 size;
  };
  typedef std::vector<ParamsRegion> ParamsRegions;
  virtual ~RnnDescriptor() {}
  virtual int64 ParamsSizeInBytes() const { return -1; }
  virtual ParamsRegions ParamsWeightRegions() const { return ParamsRegions(); }
  virtual ParamsRegions ParamsBiasRegions() const { return ParamsRegions(); }
};

// Specifies the CTC Loss computation.
//