(args)
| 242 | |
| 243 | |
| 244 | def get_checkpoint_iteration(args): |
| 245 | # Read the tracker file and set the iteration. |
| 246 | tracker_filename = get_checkpoint_tracker_filename(args.load) |
| 247 | if not os.path.isfile(tracker_filename): |
| 248 | print_rank_0('WARNING: could not find the metadata file {} '.format( |
| 249 | tracker_filename)) |
| 250 | print_rank_0(' will not load any checkpoints and will start from ' |
| 251 | 'random') |
| 252 | return 0, False, False |
| 253 | iteration = 0 |
| 254 | release = False |
| 255 | with open(tracker_filename, 'r') as f: |
| 256 | metastring = f.read().strip() |
| 257 | try: |
| 258 | iteration = int(metastring) |
| 259 | except ValueError: |
| 260 | release = metastring == 'release' |
| 261 | if not release: |
| 262 | print_rank_0('ERROR: Invalid metadata file {}. Exiting'.format( |
| 263 | tracker_filename)) |
| 264 | exit() |
| 265 | |
| 266 | assert iteration > 0 or release, 'error parsing metadata file {}'.format( |
| 267 | tracker_filename) |
| 268 | |
| 269 | return iteration, release, True |
| 270 | |
| 271 | def load_checkpoint(model, optimizer, lr_scheduler, args): |
| 272 | """Load a model checkpoint.""" |
no test coverage detected