with TrainingModeManager([feature_extractor, classifier], train=False) as mgr什么意思？

请问这句代码怎么理解呢？

with TrainingModeManager([feature_extractor, classifier], train=False) as mgr,
Accumulator(['feature_test', 'predict_prob_test', 'label', 'test_shared_weight']) as target_accumulator,
torch.no_grad():
主要是mgr是干什么的？