openLiBing可接管PR, PR详情链接:
https://www.openlibing.com/apps/transitionPrDetail?owner=Ascend&repo=pytorch&number=32693&platform=gitcode&codeHostingPlatformFlag=gitcode

Thanks for your pull-request.
The full list of commands accepted by me can be found at here。
You can get sig-info at here

PR Approval Progress

✅ Congratulations! All modules have met the lgtm and approve requirements.

Module Approval Details

💡 Tip:

• Committer can comment /approve or /lgtm
• Commenting /approve implies both code review (lgtm) and intent to merge (approve)

CLA Signature Pass

wangzhenwu, thanks for your pull request. All authors of the commits have signed the CLA. 👍

compile

问题/功能描述

本次PR主要包含三部分内容，旨在提升NPU算子ONNX导出功能的正确性和接口一致性。首先，激活了五个NPU自定义算子（多头注意力、格式转换、LSTM单元、LSTM网络和GRU网络）的ONNX导出测试用例，以验证其导出路径的有效性。其次，对torch_npu.npu_format_cast函数的API签名进行了优化，将可选参数customize_dtype明确为仅关键字参数，以提升调用清晰度和安全性。最后，修复了_NPUFormatCastOP类forward方法的参数定义，使其与相关符号图方法和包装函数保持一致，确保算子在前向传播、符号图构建及用户调用层面的接口统一。

修改方案描述

修改方案涉及三个文件。在测试文件中，移除了五个测试方法（test_wrapper_npu_multi_head_attention等）的@unittest.skip装饰器，使其能够正常执行并验证ONNX模型文件的生成。在JSON配置文件中，将torch_npu.npu_format_cast的函数签名从(self, acl_format, customize_dtype=None)调整为(self, acl_format, *, customize_dtype=None)，强制customize_dtype必须以关键字参数形式传递。在算子实现文件中，将_NPUFormatCastOP.forward方法的参数从通用的*args, **kwargs明确指定为(self, acl_format, customize_dtype=None)，并与对应的symbolic方法及包装函数对齐，同时将包装函数中的customize_dtype也改为仅关键字参数。

/lgtm

/approve

Review Guide

This pull-request passes review.
Committers who wrote a comment of /approve are: chuboning.
Reviewers who wrote a comment of /lgtm are: chuboning, liangsongwei.