This dataset contains 500 synthetic RL tasks for the Altimate-centered DataAgentBench sandbox harness.
The rows are compiled for the dab_sandbox_altimate_noctx data source and are intended to plug into VERL-style RL training with the same Altimate no-context harness used during generation and verification.
train.parquet: VERL-readable RL rows, 500 examples.
train.jsonl: JSONL mirror of the train split, 500 examples.… See the full description on the dataset page:
https://huggingface.co/datasets/forseasons/dataagent-dabench-altimate-noctx-rl-500.