Dataset Viewer
The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Couldn't infer the same data file format for all splits. Got {NamedSplit('train'): ('webdataset', {}), NamedSplit('validation'): (None, {}), NamedSplit('test'): ('xml', {})}
Error code:   FileFormatMismatchBetweenSplitsError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

agentic-rl-evalsets

Versioned eval sets for the slime agentic-RL setup (mini-swe-agent on Modal sandboxes). Built from harbor-framework/harbor-datasets by multinode-training-guide/scripts_agenticRL/prepare_eval_data.sh: converted to slime prompt rows, deterministically subsampled (seeds pinned in evalsets/v1/manifest.json), and oracle-checked (reference solutions pass through the exact rollout path; openthoughts-tblite is additionally screened to the tasks whose solutions pass on Modal/gVisor).

Layout mirrors the slime-data volume root: evalsets/v1/*.jsonl are the eval datasets; metadata.task_path in each row resolves against the repo root (ASYNC_RL_TASK_ROOT). Install:

EXPERIMENT_CONFIG=w_qwen3_swe_eval modal run slime/modal_train.py::download_data

(or huggingface_hub.snapshot_download straight into your data root)

Downloads last month
12