Nlp Flair SequenceTagger PyTorch模型量化中的属性错误

我正在处理一个从继承FlairEmbedding的自定义类。在这个类中，我想使用PyTorch的torch.quantization模块实现模型量化。为了做到这一点，我需要对模型进行几批训练，以收集统计数据并选择适当的量化参数。该模型将用于序列标记器下游，因此我使用Flair的SequenceTagger类，其参数与我在下游任务中使用的参数相同。这是该类的外观：

类CustomEmbeddingsFlarembeddings： 定义初始化__ self、tag\u字典、tag\u类型、语料库、mini\u批量大小、train\u with\u dev、用于培训 模型，微调，每块字符，带空格，标记化的基本FlairEmbeddings参数 : 超级.\uuuuu初始\uuuu模型，微调，每块字符，带\u空格，标记化\u lm self.lm.qconfig=torch.quantization.default\u config torch.quantization.prepareself.lm.qconfig，就地=真 收集统计数据的小型培训 tagger=SequenceTaggerhidden\u size=256，embeddings=self，tag\u dictionary=tag\u dictionary，tag\u type=tag\u type 培训师=模型培训师，语料库 --->培训师。培训“模型”，小批量大小=小批量大小，最大时间=10，带开发的培训=带开发的培训 torch.quantization.convertself.lm，就地=真 此代码失败，出现以下错误：

  File "/home/pie3636/project/main.py", line 28, in __init__
    embeddings = CustomEmbeddings(name, **params)
  File "/home/pie3636/project/custom_embeddings.py", line 35, in __init__
    trainer.train('model', mini_batch_size=mini_batch_size, max_epochs=10, train_with_dev=train_with_dev)
  File "/usr/local/lib/python3.6/dist-packages/flair/trainers/trainer.py", line 371, in train
    loss = self.model.forward_loss(batch_step)
  File "/usr/local/lib/python3.6/dist-packages/flair/models/sequence_tagger_model.py", line 603, in forward_loss
    features = self.forward(data_points)
  File "/usr/local/lib/python3.6/dist-packages/flair/models/sequence_tagger_model.py", line 608, in forward
    self.embeddings.embed(sentences)
  File "/usr/local/lib/python3.6/dist-packages/flair/embeddings/base.py", line 60, in embed
    self._add_embeddings_internal(sentences)
  File "/usr/local/lib/python3.6/dist-packages/flair/embeddings/token.py", line 610, in _add_embeddings_internal
    text_sentences, start_marker, end_marker, self.chars_per_chunk
  File "/usr/local/lib/python3.6/dist-packages/flair/models/language_model.py", line 157, in get_representation
    _, rnn_output, hidden = self.forward(batch, hidden)
  File "/usr/local/lib/python3.6/dist-packages/flair/models/language_model.py", line 80, in forward
    output, hidden = self.rnn(emb, hidden)
  File "/home/m.meloux/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 726, in _call_impl
    hook_result = hook(self, input, result)
  File "/home/m.meloux/.local/lib/python3.6/site-packages/torch/quantization/quantize.py", line 74, in _observer_forward_hook
    return self.activation_post_process(output)
  File "/home/m.meloux/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/m.meloux/.local/lib/python3.6/site-packages/torch/quantization/observer.py", line 276, in forward
    x = x_orig.detach()  # avoid keeping autograd tape
AttributeError: 'tuple' object has no attribute 'detach'

---

我不确定问题是否来自我的代码，或者是否需要向PyTorch或Flair的问题跟踪者报告。stacktrace让我觉得是这两个库之间的交互失败了，而不是我的代码失败了，特别是因为PyTorch的量化模块仍处于测试阶段，但我可能弄错了。任何关于错误可能是什么的输入都将受到欢迎。

当模型返回的不是张量，而是列表（例如，或元组）时，就会出现这个问题