wxsdiscover

使用官方镜像
cr.metax-tech.com/public-ai-release/maca/onnxruntime maca.ai3.1.0.5-torch2.4-py310-ubuntu22.04-amd64 2fac80e009f7 6 months ago 28.4GB

安装ultralytics,修改了源码使其在调用onnx格式yolo模型，
2026-03-24 19:08:43,090 - root - INFO - -----准备加载目标识别模型：/file_manage/models/train_weights/49b34b4427214d2bba0dd27dd7b01490.onnx
2026-03-24 19:08:43,093 - root - INFO - -----加载目标识别模型成功
Loading /file_manage/models/train_weights/49b34b4427214d2bba0dd27dd7b01490.onnx for ONNX Runtime inference...
Using ONNX Runtime 1.12.0+git-a811f5a3 with MACAExecutionProvider

使用CPU可以推理结果，同环境下-使用GPU推理的推理不出来
2026-03-24 19:08:43,090 - root - INFO - -----准备加载目标识别模型：/file_manage/models/train_weights/49b34b4427214d2bba0dd27dd7b01490.onnx
2026-03-24 19:08:43,093 - root - INFO - -----加载目标识别模型成功
Loading /file_manage/models/train_weights/49b34b4427214d2bba0dd27dd7b01490.onnx for ONNX Runtime inference...
Using ONNX Runtime 1.12.0+git-a811f5a3 with MACAExecutionProvider
2026-03-24 19:08:55,020 - root - INFO - 第0张图片：未检测到任何目标
2026-03-24 19:08:55,048 - werkzeug - INFO - 172.18.0.5 - - [24/Mar/2026 19:08:55] "POST /api/predict HTTP/1.1" 200 -

使用5060也能推理出来
配置如下
root@a6fee036f708:/predict# mx-smi
mx-smi version: 2.2.8

=================== MetaX System Management Interface Log ===================
Timestamp : Tue Mar 24 19:13:48 2026

Attached GPUs : 16
+---------------------------------------------------------------------------------+
| MX-SMI 2.2.8 Kernel Mode Driver Version: 3.3.12 |
| MACA Version: 3.1.0.14 BIOS Version: 1.29.1.0 |
|------------------------------------+---------------------+----------------------+
| GPU NAME Persistence-M | Bus-id | GPU-Util sGPU-M |
| Temp Pwr:Usage/Cap Perf | Memory-Usage | GPU-State |
|====================================+=====================+======================|
| 0 MetaX C500 Off | 0000:05:00.0 | 0% Native |
| 47C 67W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 1 MetaX C500 Off | 0000:06:00.0 | 0% Native |
| 47C 62W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 2 MetaX C500 Off | 0000:09:00.0 | 0% Native |
| 47C 58W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 3 MetaX C500 Off | 0000:0f:00.0 | 0% Native |
| 45C 56W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 4 MetaX C500 Off | 0000:27:00.0 | 0% Native |
| 48C 60W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 5 MetaX C500 Off | 0000:28:00.0 | 0% Native |
| 50C 64W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 6 MetaX C500 Off | 0000:2b:00.0 | 0% Native |
| 49C 58W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 7 MetaX C500 Off | 0000:31:00.0 | 0% Native |
| 51C 62W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 8 MetaX C500 Off | 0000:85:00.0 | 0% Native |
| 47C 59W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 9 MetaX C500 Off | 0000:86:00.0 | 0% Native |
| 47C 62W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 10 MetaX C500 Off | 0000:89:00.0 | 0% Native |
| 47C 62W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 11 MetaX C500 Off | 0000:8f:00.0 | 0% Native |
| 46C 58W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 12 MetaX C500 Off | 0000:a5:00.0 | 0% Native |
| 50C 64W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 13 MetaX C500 Off | 0000:a6:00.0 | 0% Native |
| 50C 65W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 14 MetaX C500 Off | 0000:a9:00.0 | 0% Native |
| 49C 63W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+
| 15 MetaX C500 Off | 0000:af:00.0 | 0% Native |
| 50C 62W / 350W P0 | 858/65536 MiB | Available |
+------------------------------------+---------------------+----------------------+

+---------------------------------------------------------------------------------+
| Process: |
| GPU PID Process Name GPU Memory |
| Usage(MiB) |
|=================================================================================|
| no process found |
+---------------------------------------------------------------------------------+

End of Log

我的操作系统是openeuler 22.05, 准备跑yolo识别
驱动相关如下
mx-smi version: 2.2.9

=================== MetaX System Management Interface Log ===================
Timestamp : Thu Feb 26 17:37:00 2026

Attached GPUs : 16
+---------------------------------------------------------------------------------+
| MX-SMI 2.2.9 Kernel Mode Driver Version: 3.3.12 |
| MACA Version: 3.2.1.10 BIOS Version: 1.29.1.0 |
|------------------+-----------------+---------------------+----------------------|
| Board Name | GPU Persist-M | Bus-id | GPU-Util sGPU-M |
| Pwr:Usage/Cap | Temp Perf | Memory-Usage | GPU-State |
|==================+=================+=====================+======================|
| 0 MetaX C500 | 0 Off | 0000:05:00.0 | 0% Disabled |
| 65W / 350W | 43C P0 | 858/65536 MiB | Available |
+------------------+-----------------+---------------------+----------------------+

安装torch版本如下，使用的是maca-pytorch2.4-py312-3.2.1.3-x86_64.tar.xz
Package Version

flash_attn 2.6.3+metax3.2.1.3torch2.4
torch 2.4.0+metax3.2.1.3
torchaudio 2.4.1+metax3.2.1.3
torchvision 0.15.1+metax3

运行python -c "import torch; print(torch.ones(2).cuda())"，出现错误
(yolo_semantic) [root@standalone-amd64-mxc500 download]# python -c "import torch; print(torch.ones(2).cuda())"
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "/opt/miniconda/envs/yolo_semantic/lib/python3.12/site-packages/torch/init.py", line 290, in <module>
from torch._C import * # noqa: F403
^^^^^^^^^^^^^^^^^^^^^^
ImportError: /opt/maca//lib/libmcFlashAttn.so: undefined symbol: nv_arch_type_internal

找不到问题，求解