报错详情：

在ultralytics项目文件夹内运行/home/xxx/ultralytics/train.py进行单GPU训练的时候训练可以正常进行

from ultralytics import YOLO

# Load a model
model = YOLO("/home/xxx/ultralytics/ultralytics/cfg/models/v8/yolov8s-FASFF.yaml")  # 

# Train the model
results = model.train(data="/home/xxx/ultralytics/ultralytics/cfg/datasets/VisDrone.yaml", device='1', batch=5, epochs=200, imgsz=640)

但是启动多GPU后训练就会发生KeyEroor报错

# Train the model
results = model.train(data="/home/xxx/ultralytics/ultralytics/cfg/datasets/VisDrone.yaml", device='0,1,2', batch=15, epochs=200, imgsz=640)

解决方案：

将项目文件夹ultralytics内的task.py和虚拟环境模块包内的task.py同时做类似如下的代码配置

• /home/xxx/ultralytics/ultralytics/nn/tasks.py
• /home/xxx/anaconda3/envs/openmmlab/lib/python3.8/site-packages/ultralytics/nn/tasks.py

问题即可迎刃而解！