Struct GraphFoundationModel 

pub struct GraphFoundationModel {
    pub config: FoundationModelConfig,
    pub encoder_layers: Vec<String>,
    pub pretraining_head: PretrainingHead,
    pub task_heads: HashMap<String, String>,
    pub tokenizer: GraphTokenizer,
    pub parameters: FoundationModelParameters,
}

Graph foundation model architecture

Fields

config: FoundationModelConfig

Model configuration

encoder_layers: Vec<String>

Encoder layers (stored as indices/configs instead of trait objects for clonability)

pretraining_head: PretrainingHead

Pre-training head

task_heads: HashMap<String, String>

Fine-tuning heads (stored as type names for reconstruction)

tokenizer: GraphTokenizer

Tokenizer for graph elements

parameters: FoundationModelParameters

Model parameters

Implementations

impl GraphFoundationModel

pub fn new(config: FoundationModelConfig) -> Result<Self, FoundationModelError>

Create a new foundation model

pub fn pretrain( &mut self, graphs: &[GraphData], num_epochs: usize, ) -> Result<PretrainingStats, FoundationModelError>

Pre-train the foundation model

pub fn finetune( &mut self, task_name: &str, train_data: &[(GraphData, Tensor)], val_data: &[(GraphData, Tensor)], task_config: TaskConfig, ) -> Result<FinetuningStats, FoundationModelError>

Fine-tune on downstream task

Trait Implementations

impl Debug for GraphFoundationModel

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.