运行train_whiten(nli).py出现内存错误

[Anery]

你好，我想用roberta训练均值，但出现了内存错误：

Traceback (most recent call last): File "train_whiten(nli).py", line 79, in main() File "train_whiten(nli).py", line 63, in main a_vecs_train = sents_to_vecs(a_sents_train, tokenizer, model, POOLING, MAX_LENGTH) File "/data/yuany/BERT-whitening-pytorch/all_utils.py", line 65, in sents_to_vecs vecs = np.array(vecs) numpy.core._exceptions.MemoryError: Unable to allocate 2.81 GiB for an array with shape (981382, 768) and data type float32

请问你碰到过这个问题吗？

[autoliuweijie]

应该是你用的来计算均值的语料太大了，而内存却不足够。 你可以减小语料试一试，我们实验表明用于计算均值的语料不必太大。