Wenjie Qiang

Results 8 comments of Wenjie Qiang

你好,请问下用Python3.10之后,安装的TTS版本是0.22.0吗? 我这里TTS==0.22.0要求对应的torch>=2.1,这样的torch版本源代码就跑不了了。 辛苦看下您的TTS版本和torch版本,感谢!

您好,我试了一下8卡训练,发现刚开始loss和mae特别大,几个epoch后能收敛。但是8卡bz=16时,sha的mae只能到56左右。并且8卡bz=16训练,甚至没有单卡bz=2快。我理解是不是和DM-COUNT的loss计算有关,因为DM-COUNT的源码也是默认assert self.device_count == 1 。

作者您好,我还想请教一下,不开sparse_train的话,和原文中主要有什么区别呢?

作者您好,这几天尝试着使用你们的方法。我有个疑惑,不开sparse_train,设置split_size = [1],去掉router,那么是不是就是baseline的结构一样了?

![image](https://github.com/quic/aimet/assets/47969298/74691fde-dec6-4605-9c8c-0f89aafda500)![image](https://github.com/quic/aimet/assets/47969298/354f7e36-12e8-43a2-bf4a-ca5b4bd70e19) The front one is resnet18_torch_quant.onnx,and the behind one is resnet18_after_adaround.onnx.

Hi, I would like to add a question. After reading your code, I wonder if token reorganization is used for only three layers in the test phase of Deit-S, but...

同问,BEVDepth的推理速度如何?

Hi, I noticed that while the vehicle is moving, the camera parameters remain unchanged for each frame, and only the ego2global changes. However, in the BEVDet inference code, ego2global is...