WPAIGPT-fse-patterns-1 is a fine-tuned version of the Qwen 2.5 32B Instruct model, specialized for generating structured WordPress Gutenberg Patterns.
The model was fine-tuned on a dataset derived from the WordPress.org Pattern Repository, containing:
The dataset includes patterns in multiple languages and covers all available WordPress blocks as of September 2024.
Fine-tuning was performed on A100 GPUs (80GB each) using a conversational format. The process included:
Formal evaluations have not been conducted. Performance is assessed through:
Users should be aware of potential biases in the training data that may affect pattern suggestions. Care should be taken to review and customize generated patterns for accessibility and inclusivity.
It's crucial to note that while the model generates structurally accurate patterns, full implementation within a WordPress development ecosystem requires additional tooling. Specifically, the model may reference non-existent images and URLs, necessitating a system to replace these with appropriate assets from the user's WordPress site. This functionality is being developed for integration with AgentWP (
https://agentwp.com).
For more information on using WordPress block patterns, refer to the official WordPress documentation:
https://wordpress.org/documentation/article/block-pattern/
