| Property | Value |
|---|---|
| Model Base | GPT-2 |
| Training Data | ~12k horoscopes |
| Categories | general, career, love, wellness, birthday |
| Source | Horoscopes.com |
What is gpt2-horoscopes?
gpt2-horoscopes is a specialized language model built on GPT-2 architecture, fine-tuned specifically for generating horoscope predictions across five distinct categories. The model leverages a dataset of approximately 12,000 horoscopes scraped from Horoscopes.com, enabling it to generate contextually relevant predictions based on specified categories.
Implementation Details
The model implements a category-based generation system using special tokens for different horoscope types. It was trained for 5 epochs with a learning rate of 5e-4, warmup steps of 1e2, and epsilon of 1e-8. The sequence length was set to 300 tokens, achieving a final loss of 2.77.
- Specialized tokenization with category markers (<|category|>)
- Temperature and top-k sampling for generation diversity
- Maximum sequence length of 300 tokens
- Supports five distinct horoscope categories
Core Capabilities
- Category-specific horoscope generation
- Flexible text generation parameters
- Integration with HuggingFace pipeline API
- Custom prompt formatting with category specification
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness lies in its specialized fine-tuning for horoscope generation with category-specific outputs, utilizing a structured approach with special tokens for different horoscope types.
Q: What are the recommended use cases?
The model is designed for educational and learning purposes, specifically for generating category-based horoscope text. It's important to note that it doesn't attempt to provide actual astrological predictions and should be used primarily for experimental or educational contexts.