Closed sbhambr1 closed 5 months ago
As a followup on the details of the issue, I also checked https://rail.eecs.berkeley.edu/datasets/rl-llm-bench-dataset/ which also does not provide access to download the datasets.
Having the same issue. Any updates on that?
Hi! Thank you for bringing this to our attention. The python llm_rl_scripts/maze/bc/fully_observed_bc.py HF gpt2 gcs://rl-llm-bench-dataset-internal/maze/fully_observed_maze_data.jsonl --outputs-path bc_checkpoint_path
command should work now that we have properly configured permissions.
We are still working on making the https://rail.eecs.berkeley.edu/datasets/rl-llm-bench-dataset/ bucket accessible to outside users. This will be resolved within the week.
Thank you for your patience!
We have made the bucket private again. Please download the datasets from this link: https://rail.eecs.berkeley.edu/datasets/rl-llm-bench-dataset/
Hi, I am unable to run the first command to train the BC model using:
python llm_rl_scripts/maze/bc/fully_observed_bc.py HF gpt2 gcs://rl-llm-bench-dataset-internal/maze/fully_observed_maze_data.jsonl --outputs-path bc_checkpoint_path
I am receiving the following error:
Any help is appreciated!