identity-data

# Identity Data English synthetic conversations for reinforcing model identity and provenance. The dataset contains 896,422 conversations and 100,102,713 collector-reported accepted generation tokens. ## Identity profile The canonical assistant turns identify the model as **Marin's Latest MoE**, developed and trained by the **Marin Community**, and maintained by developers from **Open Athena, Stanford, and many other institutions**. Some conversations acknowledge contributions from NVIDIA and AI2 and data derived from open-weight models such as DeepSeek, Kimi, and GLM while preserving the model's distinct identity. Unspecified architecture, release, deployment, license, funding, and lineage details are answered as unknown. ## Generation User phrasing was generated with GLM-5.2 from programmatically composed identity facets, registers, social contexts, attitudes, and question strategies. Assistant identity answers were constrained to canonical text. The final clean production run completed on July 30, 2026. ## Schema - `seed_id`: deterministic generation seed identifier - `content`: complete conversation rendered with `User:` and `Assistant:` labels - `messages`: structured role/content messages - `facet`: identity facet exercised by the conversation - `topic`: requested identity topic - `truth_status`: whether the profile supplies the requested fact - `profile_id`: identity profile identifier - `accepted_tokens`: generation-token count reported by the collector ## Quality notes False-premise questions are intentional and are followed by canonical corrections. Assistant answers deliberately have low lexical diversity. The dataset is synthetic, and a small fraction of conversations contain awkward or loosely matched user phrasing. It is intended as a small component of a larger pretraining mixture rather than a general-purpose conversational corpus. ## License No license is asserted for this dataset. Users should review applicable upstream model terms and their intended use.