[Bug]: ont-hot 编码为int 算loss报错

[chenchao222]

软件环境

- paddlepaddle:
- paddlepaddle-gpu: 2.4.2.post117
- paddlenlp: 2.5.2

重复问题

• I have searched the existing issues

错误描述

File "D:\Python Python310lib site-packagespaddlelfluiddygraphllayers.py", line 1012， in --callreturn self.forward(*inputs， **kwargs)File "p:\Python\Python310\libsite-packages\paddlelnn layerlloss.py", line 130, in forwardout = paddle.nn.functional.binary_cross_entropy_with_logits(File "p: Python ython310lib site-packages paddle nnifunctional losspy", line 824, in binary_cross-entropy_with_logitsout = _C_ops.sigmoid_cross_entropy_with_logits(kernel with key (6PU,MCHM,int64) of kemnel 'sigmoid_cross_entropy-with_logits’is not registered and fail to fallback to CPU one.RuntimeError: (NotFound) The
[Hint: Expected kemnel_iter != iter->second.end), but received kemnel_iter == iter->second.end().] (at .. paddlephilcorelkemnel_factory.c:168)

稳定复现步骤 & 代码

pretrained_model = AutoModelForSequenceClassification.from_pretrained('ernie-health-chinese')
tokenizer = AutoTokenizer.from_pretrained('ernie-health-chinese')

criterion = paddle.nn.BCEWithLogitsLoss()
多标签 labels 做ont-hot编码为int，我改为float后可以了
在算loss时报的

[w5688414]

请参考API文档，文档显示只支持float类型。

https://www.paddlepaddle.org.cn/documentation/docs/zh/api/paddle/nn/BCEWithLogitsLoss_cn.html#bcewithlogitsloss