LayoutLM如何读取不同类型个数的checkpoints进行模型初始化

[dinglei0719]

我之前使用LayoutXLM在某个73类的SER任务上训练了模型，现在想用这个模型初始化，训练另一个21个类型的SER任务，但是在计算Loss时会提示维度不匹配，请问如何不适用原来checkpoint的输出层？

Traceback (most recent call last):
File "/ns_group/dinglei/ocr/code/PaddleOCR/tools/train.py", line 227, in
main(config, device, logger, vdl_writer)
File "/ns_group/dinglei/ocr/code/PaddleOCR/tools/train.py", line 198, in main
program.train(config, train_dataloader, valid_dataloader, device, model,
File "/ns_group/dinglei/ocr/code/PaddleOCR/tools/program.py", line 308, in train
loss = loss_class(preds, batch)
File "/ns_group/lihongyu/miniconda3/envs/paddle_39_dl/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1254, in call
return self.forward(*inputs, **kwargs)
File "/ns_group/dinglei/ocr/code/PaddleOCR/ppocr/losses/vqa_token_layoutlm_loss.py", line 38, in forward
active_output = predicts.reshape(
File "/ns_group/lihongyu/miniconda3/envs/paddle_39_dl/lib/python3.9/site-packages/paddle/tensor/manipulation.py", line 3588, in reshape
out = _C_ops.reshape(x, new_shape)
ValueError: (InvalidArgument) The 'shape' attribute in ReshapeOp is invalid. The input tensor X'size must be divisible by known capacity of 'shape'. But received X's shape = [2, 512, 73], X's size = 74752, 'shape' is [-1, 21], known capacity of 'shape' is 21.
[Hint: Expected output_shape[unk_dim_idx] * capacity == in_size, but received output_shape[unk_dim_idx] * capacity:74739 != in_size:74752.] (at ../paddle/phi/infermeta/unary.cc:1762)