"SlimOrca Dedup" is a deduplicated, unfiltered subset of the SlimOrca dataset, excluding RLHF instances, resulting in 363k unique examples.
Removal of RLHF instances.
Deduplication using minhash and Jaccard similarity techniques.
Demo Models
Note: These models were trained on the full SlimOrca dataset, not the deduplicated, unfiltered version.
