Hi, thanks for open-sourcing Agent0.
I followed the README and completed the previous steps, but when I run:
bash examples/train/math_tir/train_qwen3_4b_adpo.sh
it fails with:
In 'ppo_trainer': Could not find 'data/legacy_data'
It looks like executor_train/verl_tool/trainer/config/ppo_trainer.yaml references data@data: legacy_data, but the corresponding config/data/legacy_data.yaml file is missing.
Could you please clarify:
- Is the
executor_train directory complete in the current public release?
- Is there any missing config or preprocessing file for executor training?
- Could you provide more detailed instructions for executor data preparation and the exact training script/config expected for the Qwen3-4B ADPO setup?
I would also appreciate a clearer description of how the data produced in the data curation step should be converted or placed for executor_train.
Thanks.