| Property | Value |
|---|---|
| Parameter Count | 88.2M |
| License | CreativeML OpenRAIL-M |
| Paper | arxiv:2210.14140 |
| Author | FredZhang7 |
What is distilgpt2-stable-diffusion-v2?
This is an optimized prompt generation model specifically designed for Stable Diffusion image generation. Built on DistilGPT2, it has been trained on 2.47M carefully curated prompts and fine-tuned for an additional 4.27M steps. The model stands out for its efficiency, delivering 50% faster forward propagation while consuming 40% less disk space and RAM compared to standard GPT2-based prompt generators.
Implementation Details
The model utilizes PyTorch and the Transformers library, implementing a contrastive search approach for generating diverse and coherent prompts. It features cleaned training data that excludes NSFW content, removes duplicates, and ensures high-quality prompt generation.
- Optimized architecture with 88.2M parameters
- Supports both comma and no-comma generation styles
- Implements temperature and repetition penalty controls
- Includes top-k sampling for diverse outputs
Core Capabilities
- 25% more prompt variations compared to v1
- Generates descriptive, coherent image generation prompts
- Supports batch generation of multiple prompts
- Customizable generation parameters for different use cases
- Efficient resource utilization with faster inference
Frequently Asked Questions
Q: What makes this model unique?
The model's primary strengths lie in its optimization for speed and resource efficiency, offering 50% faster processing while maintaining high-quality prompt generation. It's been trained on carefully cleaned data, removing NSFW content and ensuring consistent output quality.
Q: What are the recommended use cases?
This model is ideal for applications requiring fast, high-quality prompt generation for Stable Diffusion image generation, particularly in scenarios where resource efficiency is important. It's suitable for both interactive applications and batch processing needs.