• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1functorch.compile (experimental)
2================================
3
4AOT Autograd is an experimental feature that allows ahead of time capture of
5forward and backward graphs, and allows easy integration with compilers. This
6creates an easy to hack Python-based development environment to speedup training
7of PyTorch models. AOT Autograd currently lives inside ``functorch.compile``
8namespace.
9
10.. warning::
11    AOT Autograd is experimental and the APIs are likely to change. We are looking
12    for feedback. If you are interested in using AOT Autograd and need help or have
13    suggestions, please feel free to open an issue. We will be happy to help.
14
15.. currentmodule:: functorch.compile
16
17Compilation APIs (experimental)
18-------------------------------
19.. autosummary::
20    :toctree: generated
21    :nosignatures:
22
23    aot_function
24    aot_module
25    memory_efficient_fusion
26
27Partitioners (experimental)
28---------------------------
29.. autosummary::
30    :toctree: generated
31    :nosignatures:
32
33    default_partition
34    min_cut_rematerialization_partition
35
36Compilers (experimental)
37------------------------
38.. autosummary::
39    :toctree: generated
40    :nosignatures:
41
42    nop
43    ts_compile
44