我正在尝试让 pytorch 模型在句子分类任务上运行。在处理医学笔记时,我正在使用 ClinicalBert ( https://github.com/kexinhuang12345/clinicalBERT ) 并希望使用其预先训练的权重。不幸的是,ClinicalBert 模型仅将文本分类为 1 个二进制标签,而我有 281 个二进制标签。因此,我正在尝试实现此代码https://github.com/kaushaltrivedi/bert-toxic-comments-multilabel/blob/master/toxic-bert-multilabel-classification.ipynb,其中 bert 之后的最终分类器为 281 长。
如何在不加载分类权重的情况下从 ClinicalBert 模型加载预训练的 Bert 权重?
天真地尝试从预训练的 ClinicalBert 权重中加载权重,我收到以下错误:
size mismatch for classifier.weight: copying a param with shape torch.Size([2, 768]) from checkpoint, the shape in current model is torch.Size([281, 768]).
size mismatch for classifier.bias: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([281]).
我目前尝试从 pytorch_pretrained_bert 包中替换 from_pretrained 函数,并像这样弹出分类器权重和偏差:
def from_pretrained(cls, pretrained_model_name, state_dict=None, cache_dir=None, *inputs, **kwargs):
...
if state_dict is None:
weights_path = os.path.join(serialization_dir, WEIGHTS_NAME)
state_dict = torch.load(weights_path, map_location='cpu')
state_dict.pop('classifier.weight')
state_dict.pop('classifier.bias')
old_keys = []
new_keys = []
...
我收到以下错误消息: INFO -modeling_diagnosis - BertForMultiLabelSequenceClassification 的权重未从预训练模型初始化:['classifier.weight','classifier.bias']
最后,我想从clinicalBert预训练权重中加载bert嵌入,并随机初始化顶级分类器权重。