从变压器加载 model 时出现“无法从 pytorch 检查点文件加载权重”

Question

尝试加载 model 时：

from transformers import TrOCRProcessor,VisionEncoderDecoderModel

model=VisionEncoderDecoderModel.from_pretrained('microsoft/trocr-base-handwritten')

得到：

OSError: Unable to load weights from pytorch checkpoint file for '/home/user/.cache/huggingface/hub/models--microsoft--trocr-base-handwritten/snapshots/69659a277424eb381574e4952f3b3fa3440a419b/pytorch_model.bin' at '/home/user/.cache/huggingface/hub/models--microsoft--trocr-base-handwritten/snapshots/69659a277424eb381574e4952f3b3fa3440a419b/pytorch_model.bin'. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True.

尝试降级 torch 但没有成功，设置 from_tf=True 以及此处的所有内容： https://github.com/huggingface/transformers/issues/4336

Answer 1

我设法找到了解决方案，我需要做的就是从 hugging face trocr 中删除模型的文件夹（在我的例子中为 69659a277424eb381574e4952f3b3fa3440a419b）：“/home/user/.cache/huggingface/hub/models--microsoft--trocr-base -handwritten/snapshots/69659a277424eb381574e4952f3b3fa3440a419b”，当我重新启动该过程时，再次下载 model 没有问题。