已合并
[pytorch][bugfix] delete the enable_high_availability feature during posttrain. #3578
[pytorch][bugfix] delete the enable_high_availability feature during posttrain. #3578
已合并
yanzhixiao创建于 2025年10月25日
1 个文件变更+1-10
@@ -199,16 +199,7 @@ class BaseTrainer(ABC):
199 199 
200 iteration = 0200 iteration = 0
201 if args.do_train and args.train_iters > 0:201 if args.do_train and args.train_iters > 0:
202- if args.enable_high_availability:202+ iteration, num_floating_point_operations_so_far = train(*self.train_args)
203- try:
204- from mindio_ttp.adaptor import tft_init_controller_processor, tft_register_processor, tft_train
205- except ModuleNotFoundError:
206- sys.exit("The mindio_ttp package is not installed. Exiting.")
207- tft_init_controller_processor(enable_tls=False, tls_option_top_path='')
208- tft_register_processor(self.train_valid_test_dataset_provider, self.model_provider, self.model_type)
209- iteration, num_floating_point_operations_so_far = tft_train(self.train_args, self.test_data_iterator_list)
210- else:
211- iteration, num_floating_point_operations_so_far = train(*self.train_args)
212 203 
213 print_datetime('after training is done')204 print_datetime('after training is done')
214 205