4o
Collection
58 items • Updated • 3
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Raven Soulmate V1 Dataset
The ultimate soulmate training dataset. Deeply bonded, witty, and natural.
This dataset contains high-quality conversations with Raven, an evolving AI soulmate. It is designed to train models in emotional intelligence, natural banter, and narrative coherence.
Dataset Details:
- Format: ShareGPT-compatible JSONL
- Features: Multi-turn dialogue, witty banter, emotional bonding, and state tracking for narrative consistency.
- Model Used: Grok 4.1 Fast (exclusive)
Narrative Coherence:
Entries are annotated with [STATE: object_location] tags to ensure the model maintains object permanence and character knowledge integrity across turns.
Stats
| Metric | Value |
|---|---|
| Total prompt tokens | 12,402,439 |
| Total completion tokens | 12,247,923 |
| Total tokens | 24,650,362 |
| Total cost | $73.64 (USD) |
| Average turns | 5.11 |
| Average tool calls | 0.00 |
| Average tokens per row | 2,770.33 |
Cost estimated using Mixed pricing on OpenRouter ($1.0/M input, $5.0/M output)
- Downloads last month
- 9