yolov5训练多GPU yolov5训练为什么用不了gpu

转载

新新人类 2024-08-30 14:06:56

文章标签 yolov5训练多GPU pytorch YOLO 人工智能系统修复 文章分类 游戏开发

Yolov5 + 从CPU转GPU + Python多版本切换 + Conda包处理

文章目录

Yolov5 + 从CPU转GPU + Python多版本切换 + Conda包处理
1.Pytorch套件中存在版本不匹配
2.numpy停留在3.8没跟上pytorch2.2.2
3.ModuleNotFoundError: No module named 'pandas._libs.interval'
4.ImportError: cannot import name '_c_internal_utils' from partially initialized module 'matplotlib' (most likely due to a circular import)
5. 单升级matplotlib导致依赖缺失未升级
6.ImportError: The scipy install you are using seems to be broken, (extension modules cannot be imported)
7.If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management 炸空间
8.NotImplementedError: Could not run ‘torchvision::nms' with arguments from the 'cuDA' backend.

1.Pytorch套件中存在版本不匹配

这是后续一系列惨烈报错的起点，包括但不限于pytorch与torch，torch与torchvision，numpy与python，升级python及numpy导致matplotlib多版本残留，处理matplotlib又导致scipy包损坏，最后一切修好后出现炸掉空间无法启动。。。。。。

ERROR: pip’s dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
torchvision 0.17.2 requires torch==2.2.2, but you have torch 1.8.0 which is incompatible.

解决办法
与此前自动升级了torch而落下torchvision相反，这次是torch被落下（
发生原因是用conda create -n yolov5 python=3.8后，考虑原项目使用的是

torch==1.80+python3.8.16

把torch，torchvision，python都从低版本升级到高版本后，各种问题开始浮上水面。

2.numpy停留在3.8没跟上pytorch2.2.2

torch OSError: [WinError 126] 找不到指定的模块

升级numpy

pip install --upgrade numpy

3.ModuleNotFoundError: No module named ‘pandas._libs.interval’

(yolo5) C:\Users\ASUS\Desktop\yolo\yolov5>python train.py --img 640 --batch 32 --epoch 3 --data data/horse.yaml --cfg models/yolov5s.yaml --weights weights/yolov5s.pt
Traceback (most recent call last):
File “C:\Users\ASUS\Desktop\yolo\yolov5\train.py”, line 49, in
import val as validate # for end-of-epoch mAP
^^^^^^^^^^^^^^^^^^^^^^
File “C:\Users\ASUS\Desktop\yolo\yolov5\val.py”, line 39, in
from models.common import DetectMultiBackend
File “C:\Users\ASUS\Desktop\yolo\yolov5\models\common.py”, line 18, in
import pandas as pd
File “E:\anaconda3\envs\yolo5\Lib\site-packages\pandas_init_.py”, line 22, in
from pandas.compat import is_numpy_dev as is_numpy_dev # pyright: ignore # noqa:F401
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "E:\anaconda3\envs\yolo5\Lib\site-packages\pandas\compat_init.py", line 25, in
from pandas.compat.numpy import (
File “E:\anaconda3\envs\yolo5\Lib\site-packages\pandas\compat\numpy_init_.py”, line 4, in
from pandas.util.version import Version
File “E:\anaconda3\envs\yolo5\Lib\site-packages\pandas\util_init_.py”, line 2, in
from pandas.util._decorators import ( # noqa:F401
File “E:\anaconda3\envs\yolo5\Lib\site-packages\pandas\util_decorators.py”, line 14, in
from pandas.libs.properties import cache_readonly
File "E:\anaconda3\envs\yolo5\Lib\site-packages\pandas_libs_init.py", line 13, in
from pandas._libs.interval import Interval
ModuleNotFoundError: No module named ‘pandas._libs.interval’

缺了pandas小小的库依赖。。。

pip install --force-reinstall pandas

4.ImportError: cannot import name ‘_c_internal_utils’ from partially initialized module ‘matplotlib’ (most likely due to a circular import)

Traceback (most recent call last):
File “C:\Users\ASUS\Desktop\yolo\yolov5\models\common.py”, line 27, in
import ultralytics
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics_init_.py”, line 5, in
from ultralytics.data.explorer.explorer import Explorer
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\data_init_.py”, line 3, in
from .base import BaseDataset
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\data\base.py”, line 17, in
from ultralytics.utils import DEFAULT_CFG, LOCAL_RANK, LOGGER, NUM_THREADS, TQDM
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\utils_init_.py”, line 21, in
import matplotlib.pyplot as plt
File “E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib_init_.py”, line 157, in
from . import _api, _version, cbook, docstring, rcsetup
File "E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib\cbook_init.py", line 35, in
from matplotlib import _api, _c_internal_utils
ImportError: cannot import name ‘_c_internal_utils’ from partially initialized module ‘matplotlib’ (mos

Traceback (most recent call last):
File “C:\Users\ASUS\Desktop\yolo\yolov5\train.py”, line 49, in
import val as validate # for end-of-epoch mAP
^^^^^^^^^^^^^^^^^^^^^^
File “C:\Users\ASUS\Desktop\yolo\yolov5\val.py”, line 39, in
from models.common import DetectMultiBackend
File “C:\Users\ASUS\Desktop\yolo\yolov5\models\common.py”, line 34, in
import ultralytics
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics_init_.py”, line 5, in
from ultralytics.data.explorer.explorer import Explorer
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\data_init_.py”, line 3, in
from .base import BaseDataset
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\data\base.py”, line 17, in
from ultralytics.utils import DEFAULT_CFG, LOCAL_RANK, LOGGER, NUM_THREADS, TQDM
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\utils_init_.py”, line 21, in
import matplotlib.pyplot as plt
File “E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib_init_.py”, line 157, in
from . import _api, _version, cbook, docstring, rcsetup
File "E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib\cbook_init.py", line 35, in
from matplotlib import _api, _c_internal_utils
ImportError: cannot import name 'c_internal_utils’ from partially initialized module ‘matplotlib’ (most likely due to a circular import) (E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib_init.py)

比较复杂，问题分析：

1.项目中存在与matplotlib同名的文件，修改同名的文件（因为matplotlib是库名不能修改，即只要项目中没有同名的文件即可），再三检查后，没有同名的文件，果断放弃这一方法。

2.因多次安装卸载Python的不同版本，可能存在多个matplotlib，重新卸载matplotlib库，使用：pip uninstall matplotlib,相关的也建议删除，然后再重新安装:pip install matplotlib.
如果因为相关库删除后无法找到库，要记得重新安装（下面马上就要考）
用这个方法解决了。

另外更新库是没有用的。

pip uninstall matplotlib
pip install matplotlib

5. 单升级matplotlib导致依赖缺失未升级

During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File “C:\Users\ASUS\Desktop\yolo\yolov5\train.py”, line 49, in
import val as validate # for end-of-epoch mAP
^^^^^^^^^^^^^^^^^^^^^^
File “C:\Users\ASUS\Desktop\yolo\yolov5\val.py”, line 39, in
from models.common import DetectMultiBackend
File “C:\Users\ASUS\Desktop\yolo\yolov5\models\common.py”, line 34, in
import ultralytics
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics_init_.py”, line 5, in
from ultralytics.data.explorer.explorer import Explorer
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\data_init_.py”, line 3, in
from .base import BaseDataset
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\data\base.py”, line 17, in
from ultralytics.utils import DEFAULT_CFG, LOCAL_RANK, LOGGER, NUM_THREADS, TQDM
File “E:\anaconda3\envs\yolo5\Lib\site-packages\ultralytics\utils_init_.py”, line 21, in
import matplotlib.pyplot as plt
File “E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib_init_.py”, line 272, in
check_versions()
File "E:\anaconda3\envs\yolo5\Lib\site-packages\matplotlib_init.py", line 266, in check_versions
module = importlib.import_module(modname)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "E:\anaconda3\envs\yolo5\Lib\importlib_init.py", line 90, in import_module
return _bootstrap.gcd_import(name[level:], package, level)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "E:\anaconda3\envs\yolo5\Lib\site-packages\kiwisolver_init.py", line 8, in
from ._cext import (
ModuleNotFoundError: No module named ‘kiwisolver._cext’

重新安装依赖

华丽的归来——缺失了相关的小依赖，重新按一遍吧。。。

pip uninstall matplotlib kiwisolver
pip install matplotlib

6.ImportError: The scipy install you are using seems to be broken, (extension modules cannot be imported)

Traceback (most recent call last):
File “E:\anaconda3\envs\yolo5\Lib\site-packages\scipy_init_.py”, line 184, in
from scipy._lib._ccallback import LowLevelCallable
File “E:\anaconda3\envs\yolo5\Lib\site-packages\scipy_lib_ccallback.py”, line 1, in
from . import _ccallback_c
ImportError: cannot import name ‘_ccallback_c’ from 'scipy.lib’ (E:\anaconda3\envs\yolo5\Lib\site-packages\scipy_lib_init.py)

The above exception was the direct cause of the following exception:

scipy是重要的包，损坏要到全局层面去修

这次不在conda或者pip来处理，需要直接到cmd再开个黑框框终端

打开Windows控制台命令窗口:
Win + R 打开Windows命令运行框 输入 cmd 
修复对应环境缺失的包，输入：
conda install -n 环境名xxx scipy

7.If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management 炸空间

Traceback (most recent call last):
File “C:\Users\ASUS\Desktop\yolo\yolov5\train.py”, line 850, in
main(opt)
File “C:\Users\ASUS\Desktop\yolo\yolov5\train.py”, line 625, in main
train(opt.hyp, opt, device, callbacks)
File “C:\Users\ASUS\Desktop\yolo\yolov5\train.py”, line 384, in train
pred = model(imgs) # forward
^^^^^^^^^^^
File “E:\anaconda3\envs\yolo5\Lib\site-packages\torch\nn\modules\module.py”, line 1511, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “E:\anaconda3\envs\yolo5\Lib\site-packages\torch\nn\modules\module.py”, line 1520, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\Users\ASUS\Desktop\yolo\yolov5\models\yolo.py”, line 263, in forward
return self._forward_once(x, profile, visualize) # single-scale inference, train
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\Users\ASUS\Desktop\yolo\yolov5\models\yolo.py”, line 167, in _forward_once
x = m(x) # run
^^^^
File “E:\anaconda3\envs\yolo5\Lib\site-packages\torch\nn\modules\module.py”, line 1511, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “E:\anaconda3\envs\yolo5\Lib\site-packages\torch\nn\modules\module.py”, line 1520, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “E:\anaconda3\envs\yolo5\Lib\site-packages\torch\nn\modules\upsampling.py”, line 157, in forward
return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “E:\anaconda3\envs\yolo5\Lib\site-packages\torch\nn\functional.py”, line 4001, in interpolate
return torch._C._nn.upsample_nearest2d(input, output_size, scale_factors)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.00 MiB. GPU 0 has a total capacity of 4.00 GiB of which 0 bytes is free. Of the allocated memory 3.55 GiB is allocated by PyTorch, and 37.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

思路分析：

显卡上空间不够，实在是绷不住，此前在4090上跑GLM3-32k-6B也遇到了这种情况，还有离谱的报错，尚待解决——

yolov5训练多GPU yolov5训练为什么用不了gpu_YOLO

8.NotImplementedError: Could not run ‘torchvision::nms’ with arguments from the ‘cuDA’ backend.

yolov5训练多GPU yolov5训练为什么用不了gpu_YOLO_02

NotImplementedError: Could not run ‘torchvision::nms’ with arguments from the ‘cuDA’ backend.This could be because the operator doesn’t exist for this backend，or was omitted during the selective/custom build process (if using custom build). If you are a Facebook employee using PyTorch on mobile，please visit…

嘛，如果对您有帮助的话就开心的复制吧，整理不易转载请注明qwq！
如果有更好的建议或意见欢迎补充！
我是亓云鹏（亓Qí），努力与大家一同分享算法的快乐！

每博一图（1/1）↓

yolov5训练多GPU yolov5训练为什么用不了gpu_yolov5训练多GPU_03