|
|
|
|
|
## **DATASET_README.md** |
|
|
```markdown |
|
|
--- |
|
|
language: |
|
|
- en |
|
|
task_categories: |
|
|
- text-generation |
|
|
- summarization |
|
|
- question-answering |
|
|
- conversational |
|
|
tags: |
|
|
- multitask |
|
|
- email |
|
|
- stories |
|
|
- qa |
|
|
- summarization |
|
|
- chat |
|
|
license: |
|
|
- cc-by-4.0 |
|
|
- apache-2.0 |
|
|
- mit |
|
|
--- |
|
|
|
|
|
# Gilbert-Multitask-Mix |
|
|
|
|
|
A diverse multitask dataset for text generation training, combining samples from 5 different domains with structured prompt formatting. |
|
|
|
|
|
## Dataset Description |
|
|
|
|
|
This dataset contains 6,500+ examples across multiple text generation tasks, formatted for instruction fine-tuning of language models. |
|
|
|
|
|
### Domains Included: |
|
|
|
|
|
1. **Email Drafting** (1,200 examples) |
|
|
- Source: Enron AESLC emails |
|
|
- Task: Generate email replies from subject and body |
|
|
|
|
|
2. **Story Continuation** (1,500 examples) |
|
|
- Source: STORIES dataset |
|
|
- Task: Continue fictional narratives |
|
|
|
|
|
3. **Technical Q&A** (1,300 examples) |
|
|
- Source: Stack Overflow and technical forums |
|
|
- Task: Answer programming and technical questions |
|
|
|
|
|
4. **News Summarization** (1,300 examples) |
|
|
- Source: CNN/DailyMail |
|
|
- Task: Summarize news articles |
|
|
|
|
|
5. **Chat Responses** (1,200 examples) |
|
|
- Source: Everyday conversations |
|
|
- Task: Generate conversational replies |
|
|
|
|
|
## Dataset Structure |
|
|
|
|
|
### Data Format |
|
|
Each example follows this structured format: |
|
|
```json |
|
|
{ |
|
|
"text": "### Task: <task_description>\n\n### Input:\n<context>\n\n### Output:\n<target>" |
|
|
} |
