()
| 251 | |
| 252 | |
| 253 | def main(): |
| 254 | |
| 255 | parser = HfArgumentParser((ModelArguments, DataTrainingArguments, MyTrainingArguments)) |
| 256 | if len(sys.argv) == 2 and sys.argv[1].endswith(".json"): |
| 257 | # If we pass only one argument to the script and it's the path to a json file, |
| 258 | # let's parse it to get our arguments. |
| 259 | model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1])) |
| 260 | else: |
| 261 | model_args, data_args, training_args = parser.parse_args_into_dataclasses() |
| 262 | |
| 263 | send_example_telemetry("run_clm", model_args, data_args) |
| 264 | |
| 265 | # Setup logging |
| 266 | logging.basicConfig(format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",datefmt="%m/%d/%Y %H:%M:%S", |
| 267 | level=logging.INFO, # if training_args.local_rank in [-1, 0] else logging.WARN, |
| 268 | handlers=[logging.StreamHandler(sys.stdout)],) |
| 269 | |
| 270 | |
| 271 | if training_args.should_log: |
| 272 | # The default of training_args.log_level is passive, so we set log level at info here to have that default. |
| 273 | transformers.utils.logging.set_verbosity_info() |
| 274 | |
| 275 | log_level = training_args.get_process_log_level() |
| 276 | logger.setLevel(log_level) |
| 277 | datasets.utils.logging.set_verbosity(log_level) |
| 278 | transformers.utils.logging.set_verbosity(log_level) |
| 279 | transformers.utils.logging.enable_default_handler() |
| 280 | transformers.utils.logging.enable_explicit_format() |
| 281 | # transformers.tokenization_utils.logging.set_verbosity_warning() |
| 282 | |
| 283 | # Log on each process the small summary: |
| 284 | logger.warning( |
| 285 | f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}" |
| 286 | + f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16 or training_args.bf16}" |
| 287 | ) |
| 288 | |
| 289 | # Detecting last checkpoint. |
| 290 | last_checkpoint = None |
| 291 | if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir: |
| 292 | last_checkpoint = get_last_checkpoint(training_args.output_dir) |
| 293 | if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0: |
| 294 | raise ValueError( |
| 295 | f"Output directory ({training_args.output_dir}) already exists and is not empty. " |
| 296 | "Use --overwrite_output_dir to overcome." |
| 297 | ) |
| 298 | elif last_checkpoint is not None and training_args.resume_from_checkpoint is None: |
| 299 | logger.info( |
| 300 | f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change " |
| 301 | "the `--output_dir` or add `--overwrite_output_dir` to train from scratch." |
| 302 | ) |
| 303 | |
| 304 | # Set seed before initializing model. |
| 305 | set_seed(training_args.seed) |
| 306 | |
| 307 | config_kwargs = { |
| 308 | "cache_dir": model_args.cache_dir, |
| 309 | "revision": model_args.model_revision, |
| 310 | "use_auth_token": True if model_args.use_auth_token else None, |
no test coverage detected