KIE中的SER模型训练报错

[471417367]

win11 paddlenlp 2.3.1 paddleocr 2.5.0.3 paddlepaddle-gpu 0.0.0.post110

ser_vi_layoutxlm_xfund_zh.yml Architecture: model_type: kie algorithm: &algorithm "LayoutXLM" Transform: Backbone: name: LayoutXLMForSer pretrained: true checkpoints: # one of base or vi mode: vi num_classes: &num_classes 7

因为网络下载模型有问题，所以在paddlenlp.transformers.model_utils中直接写了本地模型路径（PaadleOCR提供的预训练模型）
'./inference/ser_vi_layoutxlm_xfund_pretrained/best_accuracy/model_state.pdparams'

Traceback (most recent call last): File "tools/train.py", line 201, in main(config, device, logger, vdl_writer) File "tools/train.py", line 174, in main program.train(config, train_dataloader, valid_dataloader, device, model, File "E:\自研OCR接口\PaddleOCR-release-2.6\tools\program.py", line 299, in train preds = model(batch) File "C:\software\Anaconda3\lib\site-packages\paddle\fluid\dygraph\layers.py", line 911, in call outputs = self.forward(*inputs, **kwargs) File "E:\自研OCR接口\PaddleOCR-release-2.6\ppocr\modeling\architectures\base_model.py", line 86, in forward x = self.backbone(x) File "C:\software\Anaconda3\lib\site-packages\paddle\fluid\dygraph\layers.py", line 911, in call outputs = self.forward(*inputs, **kwargs) File "E:\自研OCR接口\PaddleOCR-release-2.6\ppocr\modeling\backbones\vqa_layoutlm.py", line 177, in forward res.update(x[1]) IndexError: tuple index out of range

[471417367]

不确定这个修改是否正确，但是修改后训练正常 修改：

模型网络输出层修改

ppocr.modeling.backbones.vqa_layoutlm class LayoutLMv2ForSer(NLPBaseModel) if self.training: res = {"backbone_out": x[0]} # res.update(x[1]) 注释掉 return res else: return x

模型加载

paddlenlp.transformers.model_utils 模型加载的path直接path = './inference/model_state.pdparams' 模型网上提前下载好。 pretrained_resource_files_map = { "model_state": { "layoutxlm-base-uncased": "https://bj.bcebos.com/paddlenlp/models/transformers/layoutxlm_base/model_state.pdparams", } } 上面模型的下载，在win上面最后一个'/'会变成''导致无法下载。指定本地模型也可以解决无法联网的问题。

疑问：ser_vi_layoutxlm_xfund_pretrained中提供的预训练模型加载会报错。 试过checkpoints的再训练也不行，而且**_udml还不支持，不知道后续有没有相关优化。

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.