File: update_differentiable_graph_requires_grad.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 (20 lines) | stat: -rw-r--r-- 740 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
#pragma once

#include <torch/csrc/jit/ir/ir.h>

namespace torch {
namespace jit {

// Because differentiable graphs detach the gradients of input Tensors,
// creating and inlining differentiable graphs changes the requires_grad
// property of tensors in the graph. This pass updates prim::profiles
// requires_grad to keep profiled properties up to date, it does not update
// grad properties of other nodes like graph inputs bc the only downstream
// user of the grad property is the profiling executor, which just uses
// the types of prim::profiles
TORCH_API void UpdateDifferentiableGraphRequiresGrad(
    std::shared_ptr<Graph>& diff_forward_graph,
    c10::optional<bool> new_requires_grad);

} // namespace jit
} // namespace torch