missing language instructions to covert mobile aloha dataset

[fayefw]

Hi, thank you for your great work! I tried to covert the mobile aloha dataset from hdf5 to tfrecords but the output was empty. I inspected the hdf5 file and found there is no instruction in it. could you tell me how can I find or set the dataset with instructions? thanks a lot!

[thkkk]

You can check the expanded_instruction_gpt-4-turbo.json in the dataset, which is the instructions of the tasks. During training, one of the three prompts will be uniformly randomly sampled. If the long prompt is selected, the algorithm will also uniformly randomly select a prompt from the long prompt.