Instructions to use Presto-Design/llm_adapter_vectorizer_qwen7b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Presto-Design/llm_adapter_vectorizer_qwen7b with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Presto-Design/llm_adapter_vectorizer_qwen7b", dtype="auto") - Notebooks
- Google Colab
- Kaggle
| library_name: transformers | |
| tags: [] | |
| # Presto SVG Vectorizer | |
| A powerful SVG generation model that converts images into high-quality, editable SVG code. [In our benchmark testing against leading LLMs](https://davidmack.medium.com/a-new-benchmark-for-llm-svg-performance-8d4906918960) like GPT-4 and Claude, our model achieved: | |
| - ๐ฏ 100% Generation Success Rate | |
| - ๐ 0.899 BLEU Score for code accuracy (compared to ~0.35 for other LLMs) | |
| - ๐ผ๏ธ 0.965 Structural Similarity Score | |
| - ๐ 0.745 Pixel-wise Similarity | |
| The model excels at: | |
| - Generating clean, maintainable SVG code | |
| - Preserving visual fidelity of the original image | |
| - Handling complex design elements including colors, gradients, shapes, and masks | |
| - Suggesting relevant stock photography alternatives through alt text | |
| Try it now: [Presto Vectorizer on Hugging Face](https://huggingface.co/Presto-Design/llm_adapter_vectorizer_qwen7b) | |
| For more details on our benchmarking methodology and results, visit our [SVG Benchmark Suite](https://github.com/Presto-design/svg-benchmark). |