File: InferenceGraph.h

package info (click to toggle)
pytorch 1.13.1%2Bdfsg-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 139,252 kB
  • sloc: cpp: 1,100,274; python: 706,454; ansic: 83,052; asm: 7,618; java: 3,273; sh: 2,841; javascript: 612; makefile: 323; xml: 269; ruby: 185; yacc: 144; objc: 68; lex: 44
file content (26 lines) | stat: -rw-r--r-- 833 bytes parent folder | download | duplicates (2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
#pragma once

#include "caffe2/core/workspace.h"

namespace caffe2 {

/**
 * This struct stores information about the inference graph which defines
 * underlying math of BlackBoxPredictor. Other parts of it such as various
 * threading optimizations don't belong here.
 */
struct InferenceGraph {
  std::unique_ptr<NetDef> predict_init_net_def;
  // shared_ptr allows to share NetDef with its operators on each of the threads
  // without memory replication. Note that predict_init_net_def_ could be stored
  // by value as its operators are discarded immidiatly after use (via
  // RunNetOnce)
  std::shared_ptr<NetDef> predict_net_def;

  std::vector<std::string> input_names;
  std::vector<std::string> output_names;
  std::vector<std::string> parameter_names;

  bool predictor_net_ssa_rewritten{false};
};
} // namespace caffe2