QUEST-RL-Data

# QUEST RL Data Training split for the **QUEST / DeepResearch** RL recipe. Each row includes `prompt`, `reward_model`, `extra_info`, and related fields. ## Usage examples ### Load with `datasets` and filter by category ```python from datasets import load_dataset # Streaming (lower memory) ds = load_dataset("osunlp/QUEST-RL-Data", split="train", streaming=True) # Objective only obj = ds.filter(lambda x: x["rl_task_category"] == "objective") # Open-ended only oe = ds.filter(lambda x: x["rl_task_category"] == "open-ended") # First objective row row = next(iter(obj)) print(row["rl_task_category"]) print(row["data_source"]) print(len(row["prompt"])) # number of messages ``` ### Non-streaming (full table in memory) ```python from datasets import load_dataset ds = load_dataset("osunlp/QUEST-RL-Data", split="train") obj_ds = ds.filter(lambda x: x["rl_task_category"] == "objective") oe_ds = ds.filter(lambda x: x["rl_task_category"] == "open-ended") print(len(obj_ds), len(oe_ds)) ``` ### Read Parquet directly (no Hub dataset metadata required) ```python from datasets import load_dataset url = "https://huggingface.co/datasets/osunlp/QUEST-RL-Data/resolve/main/data/train.parquet" ds = load_dataset("parquet", data_files=url, split="train") ``` ## Other columns (short reference) - **`data_source`**: Source tag (e.g. `deepresearch_tasks`). - **`prompt`**: `list[{"role", "content"}]` chat-style input. - **`ability`**: High-level capability / task label. - **`reward_model`**: Reward configuration (Python literal string; some rows embed numpy-like `array(...)` and need custom parsing). - **`extra_info`**: Extra metadata (usually a Python dict literal as a string). ## License This dataset is made available under the [**Open Data Commons Attribution License (ODC-BY) v1.0**](https://opendatacommons.org/licenses/by/1-0/) (Open Data Commons **Attribution** family). See the license text at the link above for attribution and reuse terms. For questions about this dataset, open a Discussion or Issue on this dataset repo.