NumPy 2.0.0 compatibility issue for torch_npu and add NPU deployment guide · Pull Request !2 · openEuler/ServerlessLLM

## Commit 1

**Fix:**  NumPy 2.0.0 compatibility issue with torch_npu versions 2.4.0 and 2.5.1. The fix constrains NumPy to a version less than 2.0.0 to prevent a compilation failure caused by the removal of np.float_.

```
(sllm-worker) [root@devserver-bms-6c1da567 PROJECT-JACK]# python test_sllm_store_load.py 
torch_npu imported successfully - PrivateUse1HooksInterface should be registered
Warning: Failed to initialize NPU device 0: SetPrecisionMode:build/CMakeFiles/torch_npu.dir/compiler_depend.ts:155 NPU function error: at_npu::native::AclSetCompileopt(aclCompileOpt::ACL_PRECISION_MODE, precision_mode), error code is 500001
[ERROR] 2025-08-15-18:08:26 (PID:334737, Device:0, RankID:-1) ERR00100 PTA call acl api failed
[Error]: The internal ACL of the system is incorrect.
        Rectify the fault based on the error information in the ascend log.
EC0010: [PID: 334737] 2025-08-15-18:08:26.388.105 Failed to import Python module [AttributeError: `np.float_` was removed in the NumPy 2.0 release. Use `np.float64` instead..].
        Solution: Check that all required components are properly installed and the specified Python path matches the Python installation directory. (If the path does not match the directory, run set_env.sh in the installation package.)
```

## Commit 2

-  **Documentation:**  Adds a new deployment guide for Ascend NPU in both English and Chinese. This documentation provides clear instructions for setting up and running the project on NPU devices.
-  **Announcement:**  Adds a news update to formally announce support for Ascend NPU.

类型	指派人员	状态
审查	openeuler-ci-bot 栾建海 sinever juntian qinbinjuan	已完成 (0/0人)
测试	openeuler-ci-bot	已完成 (0/0人)

openEuler/ServerlessLLM .gitee-modal { width: 500px !important; }

搜索帮助

openEuler/ServerlessLLM