Dataset Registry
All candidate datasets evaluated for PAUL Open. Only a highly curated subset of 245 examples was selected for the SFT and DPO stages. The table below represents the license audit and provenance of datasets considered for this research.
| Dataset Candidate | Provenance | License | Status |
|---|---|---|---|
| PAUL Indic SFT v1 | Curated by PAUL Foundry | Apache 2.0 | Built |
| PAUL DPO Corrective v2 | Curated by PAUL Foundry | Apache 2.0 | Built |
| OpenHermes 2.5 (Subset) | teknium | Apache 2.0 | Experimental |
| OASST2 Indic (Filtered) | LAION | Apache 2.0 | Planned |
| NCERT Biology QA (Candidate) | Various sources | CC-BY-NC-4.0 | Pending |