cmu-lti/osim-mid-training
ODYSSIM Midtraining Corpus This repository contains the midtraining corpus used for ODYSSIM behavioral foundation model experiments. The corpus is stored as parquet shards grouped by dataset/source, with train shards and held-out test files for the corresponding sources. The release mirrors the previously staged dataset Xuhui/sft_processed_large_split_v3 into the CMU-LTI organization for the paper release. Summary statistics from the audit pass: 21.4M train rows, approximately… See the full description on the dataset page: https://huggingface.co/datasets/cmu-lti/osim-mid-training.
| Author / organization | cmu-lti |
|---|---|
| License (as declared) | other |
| Downloads | 424 (as reported by Hugging Face) |
| Likes | 5 |
| Created | 9 June 2026 |
| Last updated at source | 9 June 2026 |
Popularity, last 3 days
424 now (+7 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
More from cmu-lti
- cmu-lti/sotopia-piDataset
- cmu-lti/sotopiaDataset
- cmu-lti/machine-translation-for-visionDataset
- cmu-lti/osim-post-trainingDataset
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.