Contents

struct

SushiAI::Train::TrainerOptions

How the loop behaves.

Declared in
include/SushiAI/train/trainer.hpp

Public attributes

int64_t epochs = 1

How many passes over the dataset fit() makes.

bool track_accuracy = true

Score accuracy, at the cost of one host read of the logits per step.

bool profile = false

Collect the runtime's per-node timings; needs a profiling RuntimeContext.

OptimizerPlacement optimizer_placement = OptimizerPlacement::IN_PLAN

Whether to fold the optimizer's update into the compiled step.

const Graph::IFusionSelector * fusion_selector = nullptr

Specifies the fusion selector for device-specific kernel variants.

const Graph::IPrecisionPolicy * precision = nullptr

Precision policy used to validate operand types before lowering.