# 【合入来源】 > **如有社区issue，请关联issue链接** > **请勿携带内部流程信息（需求链接、问题单、内部issue等）** - [x] 需求 - [ ] 问题单 - [x] issue/工单 https://gitcode.com/Ascend/pytorch/issues/2629 - [ ] 重构优化 - [ ] 资料更新 # 【修改方案】 A5 (Ascend950+) 砍掉 GE TransData 路径，私有格式（FRACTAL_NZ 系列）仅走 `aclnnNpuFormatCast`。前期只支持 ND2NZ，NZ tensor 的 D2H / 打印等 debug 场景不可用。本 PR 适配 D2H 和打印场景： 1. op_api copy_ d2h 补 NZ→ND auto-cast（CopyKernelOpApi.cpp） 2. copy_ 标记 internal_format_opapi，去除 A5 gen 守卫（npu_native_functions.yaml） 3. Tensor.__repr__ patch：私有格式先 .cpu() 再打印（tensor_repr_patches.py） 4. NZ_C0 序列化兼容（NPUSerialization.cpp、FormatCastKernelNpu.cpp） # 【资料变更】 不涉及 # 【接口变更】 不涉及 # 【功能验证】 测试场景：A5 上 fp16/bf16/int8/int32 四种 dtype NZ tensor 的 D2H、打印、序列化 round-trip、C0 compat load。A2/A3上跳过这些用例，原用例PASS。 实测输出： - A2/A3: ``` # python test_serialization_format.py [W716 10:43:26.785109430 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:43:39.330520500 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:43:52.914299730 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:44:06.520910930 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:44:18.151774430 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:44:38.550773670 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:44:50.137489630 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:44:50.138126100 TensorFactories.cpp:348] Warning: Cannot create tensor with interal format while allow_internel_format=False, tensor will be created with base format. (function operator()) [W716 10:45:16.799575350 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:45:30.578267770 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:45:30.578893090 TensorFactories.cpp:348] Warning: Cannot create tensor with interal format while allow_internel_format=False, tensor will be created with base format. (function operator()) [W716 10:45:55.367333900 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:46:08.893237850 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) [W716 10:46:08.894251630 TensorFactories.cpp:348] Warning: Cannot create tensor with interal format while allow_internel_format=False, tensor will be created with base format. (function operator()) [W716 10:46:33.330123970 NPUCachingAllocator.cpp:201] Warning: The current CANN and Soc versions require processing for 32 padding size, with memory allocation. (function operator()) .sss ---------------------------------------------------------------------- Ran 4 tests in 200.025s OK (skipped=3) ``` - A5: ``` # python test_serialization_format.py s.Warning: The current version of the file storing weights is old, and it is relanded due to internal bug of torch and compatibility issue. We will deprecate the loading support for this type of file in the future, please use newer torch to re-store the weight file. .Warning: The current version of the file storing weights is old, and it is relanded due to internal bug of torch and compatibility issue. We will deprecate the loading support for this type of file in the future, please use newer torch to re-store the weight file. . ---------------------------------------------------------------------- Ran 4 tests in 26.036s OK (skipped=1) ``` # 【CheckList】 - [x] 代码注释完备，正确记录错误日志 - [x] 代码实现进行了返回值、空指针等校验 - [x] PR标题正确使用类型标签，如：feat、fix、refactor、docs、test等 - [x] PR持续集成流水线（CI）执行通过，代码检查无异常