openpi0.5 自定义数据集的训练参数配置

自定义aloha数据 代码训练参数配置

TrainConfig(# 这里在pi0模型的基础上，微调叠毛巾模型name="pi0_aloha_towel2",# 采用pi0的结构，如果是采用pi0.5，则这行改成model=pi0_config.Pi0Config(pi05=True),model=pi0_config.Pi0Config(),# 自定义数据集的位置data=LeRobotAlohaDataConfig(# 将自定义数据集放在本地的xxxx/.che/huggingface/repo_id目录下repo_id="aloha_fold_towel_lerobot",assets=AssetsConfig(assets_dir="/yonghu/openpi/assets",# 归一化文件的存放位置asset_id="pi0_aloha_towel2/aloha_fold_towel_lerobot",),default_prompt="fold the towel",repack_transforms=_transforms.Group(inputs=[_transforms.RepackTransform({"images":{"cam_high":"observation.images.cam_high","cam_left_wrist":"observation.images.cam_left_wrist","cam_right_wrist":"observation.images.cam_right_wrist",},"state":"observation.state","actions":"action",})]),),weight_loader=weight_loaders.CheckpointWeightLoader("gs://openpi-assets/checkpoints/pi0_aloha_towel/params"),num_train_steps=20_000,batch_size=64,),