The error in the pi 0.5 model's reasoning is quite significant. #29
Description
Hello, thank you very much for sharing the open-source script for accelerating the real-time pi0.5 model. Here are my questions:
The correct pkl format conversion is achieved through the model conversion script. However, during the inference process, there are significant differences in the robotic arms of the AJX inference. Below is the comparison chart of the output curves of my left-arm inference. Could you please help me solve this problem? Is there any detail that needs adjustment during the conversion or the inference process? Thank you very much. I'm looking forward to your reply. Thank you.
