Convert Lerobot dataset format to rlds

[Ke-Wang1017]

Hi I have encountered this wonderful repo for fine-tuning openVLA https://github.com/moojink/openvla-oft. My data is store in lerobot dataset format and I tried to convert this into rlds format. However this repo requires python<=3.9 while lerobot requires python>=3.10. Is there a way to easily convert it? Thanks!

[TobiasVoigt20]

Hello,

I am running this repo in a environment with python3.10 setup for Octo. I works just fine.
Maybe this works for you as well. Just try to run it with python3.10 and install the packages being used.

Best

[Ke-Wang1017]

Hello,

I am running this repo in a environment with python3.10 setup for Octo. I works just fine. Maybe this works for you as well. Just try to run it with python3.10 and install the packages being used.

Best

Hi thank you, when I install with conda env create -f environment_ubuntu.yml it asks python<=3.9, do you mean I just install package needed?
Thanks

[1786707378]

你好 请问你转换成功了吗

[TobiasVoigt01]

@Ke-Wang1017

I don’t use Conda and didn’t set up my environment with the environment_ubuntu.yml.

What I did: I created a virtual environment with Python 3.10 (using the command: python3.10 -m venv my_venv_name). Then, I installed everything needed for Octo (you can check the Octo Git repository for more details).

I then realized that the rlds_dataset_builder repo might be useful, but I didn’t want to create a separate environment with a Python version < 3.9. Ideally, I wanted to run rlds_dataset_builder using the existing environment I already set up for Octo (I assume you're facing the same issue?).

So, I simply cloned the rlds_dataset_builder repo and tried running it with the activated Octo environment. It worked on the first try, and I didn’t even have to install any additional packages.

This means: The rlds_dataset_builder repo and its dependencies run just fine in an environment with Python 3.10. You can clone the repo and use the existing Python 3.10 environment you’ve already set up to run the code and see if it works. You might need to install some additional packages if it asks for them.

I hope this is clear and helps you out.

Best regards

[Ke-Wang1017]

@Ke-Wang1017

I don’t use Conda and didn’t set up my environment with the environment_ubuntu.yml.

What I did: I created a virtual environment with Python 3.10 (using the command: python3.10 -m venv my_venv_name). Then, I installed everything needed for Octo (you can check the Octo Git repository for more details).

I then realized that the rlds_dataset_builder repo might be useful, but I didn’t want to create a separate environment with a Python version < 3.9. Ideally, I wanted to run rlds_dataset_builder using the existing environment I already set up for Octo (I assume you're facing the same issue?).

So, I simply cloned the rlds_dataset_builder repo and tried running it with the activated Octo environment. It worked on the first try, and I didn’t even have to install any additional packages.

This means: The rlds_dataset_builder repo and its dependencies run just fine in an environment with Python 3.10. You can clone the repo and use the existing Python 3.10 environment you’ve already set up to run the code and see if it works. You might need to install some additional packages if it asks for them.

I hope this is clear and helps you out.

Best regards

Hi @TobiasVoigt01 thank you, I use lerobot which uses pytorch and I need to install quite a lot of packages but you give me a good idea on how to do it. I already did the conversion of lerobot dataset. Thanks

[Ke-Wang1017]

你好 请问你转换成功了吗

chenggong le