kumoai.trainer.GraphTransformerModelPlan

class kumoai.trainer.GraphTransformerModelPlan

Configuration parameters that define how the Kumo graph transformer model is architected.

Note

Graph Transformer is supported for only node prediction tasks.

Variables:

• channels – (list[int]) A list of potential dimension of layers in the graph transformer model (default: [64, 128, 256]). Supported Task Types: Node Prediction

• num_layers – (list[int]) A list of potential number of transformer layers in the Graph Transformer model (default: [4, 6, 8]). Supported Task Types: Node Prediction

• num_heads – (list[int]) A list of potential number of attention heads in the Graph Transformer model (default: [8, 16]). Supported Task Types: Node Prediction

• dropout – (list[float]) A list of potential dropout rates in the Graph Transformer model (default: [0.1, 0.5]). Supported Task Types: Node Prediction

• positional_encodings – (list[PositionalEncodingType]) A list of potential positional encodings to use in the Graph Transformer model (default: [None]). Supported Task Types: Node Prediction