Bolano

@duohaoxue you can get true output in TensorRT7.2.1 but can't in Xavier ?which TensorRT version in your Xavier?if TensorRT

@JimXu1989 mmdet

> In `demo.py`, what argument should be passed for `plugin`? What role does this play here? Thanks. When using versions earlier than Tensorrt 8.4.1.5, you should compile the MultiScaleDeformableAttention layer...

感谢您的帮助，可以正常加载模型权重了

> > 感谢您的帮助，可以正常加载模型权重了 > > @talebolano 您好，请问是怎么解决的呀，可以贴下解决代码吗？谢谢！ @HonestyBrave 参照 https://github.com/ultralytics/ultralytics/pull/11787/files 对代码进行修改，随后将训练代码修改为如下： ```python from ultralytics import YOLOv10 model = YOLOv10('./data/yolov10x.yaml') model.info() model.train(data="data/coco.yaml",imgsz=1280,epochs=250,device="0,1,2,3,4,5,6,7",batch=32,lr0=0.005,pretrained="yolov10x.pt") ```

same question

@HandsLing 我尝试了将attention_mask_l直接置为None，对最终输出结果没有任何影响，我怀疑一开始这里便写错了