Created by: Black-cup

使用的模型是ernie，模型已经训练好了，按照text-classification的demo写了一个预测demo，通过这个函数得到cls_task，之后for循环cls_task.predict(data=data)，循环预测每个句子，但是执行之后发现内存持续在增长，不预测就不会增长，请问这是哪里出了问题呢？ def loading_neg(module): dataset_neg = ChnSentiCorp_neg() inputs_neg, outputs_neg, program_neg = module.context( trainable=True, max_seq_len=128) reader = hub.reader.ClassifyReader( dataset=dataset_neg, vocab_path=module.get_vocab_path(), max_seq_len=128, use_task_id=False)

pooled_output = outputs_neg["pooled_output"]

# Setup feed list for data feeder
# Must feed all the tensor of ERNIE's module need
feed_list = [
    inputs_neg["input_ids"].name,
    inputs_neg["position_ids"].name,
    inputs_neg["segment_ids"].name,
    inputs_neg["input_mask"].name,
]

neg_config = hub.RunConfig(
    use_data_parallel=False,
    use_pyreader=True,
    use_cuda=False,
    batch_size=1,
    enable_memory_optim=False,
    checkpoint_dir=checkpoint_dir_neg,
    strategy=hub.finetune.strategy.DefaultFinetuneStrategy())

neg_cls_task = hub.TextClassifierTask(
    data_reader=reader,
    feature=pooled_output,
    feed_list=feed_list,
    num_classes=dataset_neg.num_labels,
    config=neg_config)
return neg_cls_task