guodaosun/tale-frame
TinyStories Dataset README Overview This dataset is based on TinyStories and includes structured JSON data with corresponding annotations, designed for research in controllable story generation and related tasks. Dataset Structure Each data item contains the following fields: 1. conversations Type: List Purpose: Contains the JSON of the story from: Always set to "human". value: Structured data containing entities, events, story… See the full description on the dataset page: https://huggingface.co/datasets/guodaosun/tale-frame.
TinyStories Dataset README
Overview
This dataset is based on TinyStories and includes structured JSON data with corresponding annotations, designed for research in controllable story generation and related tasks.
Dataset Structure
Each data item contains the following fields:
1. conversations
- Type: List
- Purpose: Contains the JSON of the story
- `from`: Always set to
"human". - `value`: Structured data containing entities, events, story structures and relationships described in JSON format.
2. chosen
- Type: Object
- Purpose: Contains the final selected story.
- `from`: Always set to
"gpt". - `value`: The accepted story text.
3. rejected
- Type: Object
- Purpose: Contains a rejected version of the story.
- `from`: Always set to
"gpt". - `value`: The rejected story text.
Structure of value Field
Units Attributes Definition
Entity
Event
Relationship
Story
Nullability
- Required fields: Must contain a valid value.
- Nullable fields: Can be left empty.
Dataset Statistical Features
Entities
- Total entities: 24,154
- Unique entity names: 4,560
- Total entity traits: 4,770
- Entity trait distribution:
- Negative
- Neutral
- Positive
Stories
- Total stories: 9,851
- Emotional distribution:
- Negative: 1,243
- Neutral: 2,433
- Positive: 17,727
Events
- Total events: 68,553
- Emotional distribution:
- Negative: 9,226
- Neutral: 2,213
- Positive: 12,715
Relationships
- Total relationships: 69,481
- Emotional distribution:
- Negative: 12,213
- Neutral: 15,458
- Positive: 40,882
Annotations
Stories are evaluated using models like GPT-4 and annotated as "Chosen" or "Rejected" based on quality.
