| 71 | self.checkpoint_state = [] |
| 72 | |
| 73 | def inject_default_parser(self): |
| 74 | p = self.parser |
| 75 | p.add_argument("-d", "--devices", default="", help="set data parallel training") |
| 76 | p.add_argument( |
| 77 | "-c", |
| 78 | "--continue", |
| 79 | type=extant_file, |
| 80 | metavar="FILE", |
| 81 | dest="continue_fpath", |
| 82 | help="continue from one certain checkpoint", |
| 83 | ) |
| 84 | p.add_argument("--local_rank", default=0, type=int, help="process rank on node") |
| 85 | p.add_argument( |
| 86 | "-p", |
| 87 | "--port", |
| 88 | type=str, |
| 89 | default="16005", |
| 90 | dest="port", |
| 91 | help="port for init_process_group", |
| 92 | ) |
| 93 | |
| 94 | def register_state(self, **kwargs): |
| 95 | self.state.register(**kwargs) |