# Sources

RPMem data is prepared from the following sources through normalization,
segmentation, probe generation, and split construction. Original attribution
and source-specific terms apply; the bundle has no single shared license.

| Source | Terms |
| --- | --- |
| [UltraChat](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k) | MIT |
| [TaskMaster2](https://github.com/google-research-datasets/Taskmaster/tree/master/TM-2-2020) | CC-BY-4.0 |
| [Schema-Guided Dialogue](https://github.com/google-research-datasets/dstc8-schema-guided-dialogue) | CC-BY-SA-4.0 |
| [Toucan](https://huggingface.co/datasets/Agent-Ark/Toucan-1.5M) | Apache-2.0 |
| [SWE-Zero](https://huggingface.co/datasets/nvidia/SWE-Zero-openhands-trajectories) | CC-BY-4.0; underlying code licenses also apply |
| [D2L synthetic QA](https://huggingface.co/datasets/SakanaAI/self_gen_qa_d2l) | Redistribution confirmed; contact source authors for further use terms |
| [PERMA](https://huggingface.co/datasets/ustclsc/PERMA) | Apache-2.0 |
| [PersonaMem-v2](https://huggingface.co/datasets/bowen-upenn/PersonaMem-v2) | CC-BY-4.0 |
| [PrefEval](https://github.com/amazon-science/PrefEval) | CC-BY-NC-4.0 (noncommercial) |

Source revisions and processing settings are provided in the
[source recipe](https://github.com/Quark-Medical/rpmem/blob/main/experiments/phase1/formal_sources_v1.yaml)
and benchmark metadata.