| license: cc-by-4.0 | |
| configs: | |
| - config_name: tasks | |
| default: true | |
| data_files: | |
| - split: default | |
| path: data/tasks/all.jsonl | |
| - split: dev | |
| path: data/tasks/dev.jsonl | |
| - config_name: submissions | |
| data_files: | |
| - split: default | |
| path: data/submissions/*.jsonl | |
| - config_name: task_scores | |
| data_files: | |
| - split: default | |
| path: data/task_scores/*.jsonl | |
| pretty_name: 🕺 DABstep | |
| # Data Agent Benchmark for Multi-step Reasoning (DABstep) Dataset | |
| This repository hosts a HF Dataset the supports the benchmark and leaderboard. | |
| For the main entrypoint to the benchmark, see the leaderboard here: | |
| https://huggingface.co/spaces/adyen/DABstep | |
| This Dataset has 3 splits: | |
| * tasks | |
| * submissions | |
| * task_scores | |
| Users of the benchmark would read from the `tasks` split to run the baseline. The other splits are used to support the leaderboard. | |
| The datasets are in the `data/context` directory. | |
| ## Example Usage | |
| ``` | |
| from datasets import load_dataset | |
| ds = load_dataset("adyen/DABstep", name="tasks", split="default") | |
| for task in ds: | |
| print(task) | |
| # agent solves task | |
| # write tasks answers to file in the format provided in the leaderboard | |
| # submit file to the form in the leaderboard | |
| ```` | |
| Full example here: | |
| https://colab.research.google.com/drive/1pXi5ffBFNJQ5nn1111SnIfjfKCOlunxu |