Acacia: A Graph Foundation Model Trained on the Web Graph
What it is
Acacia is a graph foundation model trained from scratch on the Common Crawl web graph, independent of pretrained LLMs. It supports arbitrary feature dimensionalities and a wide range of tasks, including node classification, link prediction, and graph generation, without requiring additional training or task-specific heads.
The problem it solves
Existing graph foundation models typically require training additional classification heads or feature projectors for new graphs/labels and often rely on stitching together pretrained LLMs, making them less flexible and more complex to deploy.
What you could use it for
- Node classification on new graphs without retraining
- Link prediction in large-scale web-scale graphs
- Node clustering and graph generation tasks
- In-context learning for graph-structured data
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.