schneiderkamplab/dfm10-alexandra-multi-zebra-logic
dfm10-alexandra-multi-zebra-logic Selected Danish and English Multi-Zebra train configurations in chat form. Contents Format: gzip-compressed JSON Lines under data/train-*.jsonl.gz Schema: chat messages, optional condition and tools, plus provenance Shards: 1 Rows: 768 Category: Reasoning Upstream material alexandrainst/multi-zebra-logic Processing Six selected train configurations are combined; validation and test are excluded.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm10-alexandra-multi-zebra-logic.
dfm10-alexandra-multi-zebra-logic
Selected Danish and English Multi-Zebra train configurations in chat form.
Contents
- Format: gzip-compressed JSON Lines under
data/train-*.jsonl.gz - Schema: chat
messages, optionalconditionandtools, plus provenance - Shards: 1
- Rows: 768
- Category: Reasoning
Upstream material
alexandrainst/multi-zebra-logic
Processing
Six selected train configurations are combined; validation and test are excluded.
Every packaged row is taken from the accepted source tree used by the final DFM10 build. Tokenized arrays and epoch sampling indices are not included.
License and release review
This package does not assert a new license over upstream material. Preserve the upstream licenses, notices, and attribution requirements. Review upstream terms and dataset-card attribution before upload.
Validate
python recreate_dataset.py