File size: 2,200 Bytes
665ac6e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
---
language:
  - en
size_categories:
  - 100K<n<1M
task_categories:
  - text-generation
pretty_name: Identity Data
tags:
  - synthetic
  - identity
  - pretraining
  - marin
---

# 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.