Struct Trainer 

pub struct Trainer<B: Backend> {
    pub executor: Executor<B>,
    pub batch_size: usize,
    /* private fields */
}

High-level training loop runner.

Reads the @training configuration from an IrProgram and orchestrates forward, backward, and optimizer steps.

Fields

executor: Executor<B>

The graph executor.

batch_size: usize

Batch size (for reference, data batching is external).

Implementations

impl<B: Backend> Trainer<B>

pub fn from_program( program: IrProgram, device: B::Device, config: RuntimeConfig, ) -> Result<Self>

Create a Trainer from an IrProgram.

Reads the @training block and fails if no training config exists.

pub fn train( &mut self, data: &[HashMap<String, Tensor<B>>], targets_key: &str, ) -> Result<TrainResult>

Run the training loop with an iterator of input batches.

Each batch is a HashMap<String, Tensor<B>> mapping input names to tensors. Returns a TrainResult with per-epoch loss logs.

targets_key is the name of the target tensor in each batch map.

pub fn infer( &self, inputs: &HashMap<String, Tensor<B>>, ) -> Result<HashMap<String, Tensor<B>>>

Run a single forward pass (inference mode).

pub fn model_graph_name(&self) -> &str

Get the model graph name.

pub fn loss_fn_name(&self) -> &str

Get the loss function name.

pub fn epochs(&self) -> usize

Get the number of epochs.