primeqa.ir.dense.dpr_top.torch_util.transformer_optimize.TransformerOptimize

class primeqa.ir.dense.dpr_top.torch_util.transformer_optimize.TransformerOptimize(hypers: primeqa.ir.dense.dpr_top.torch_util.hypers_base.HypersBase, num_instances_to_train_over: int, model)

Bases: object

Collects standard steps to train transformer call step_loss after computing each loss

Methods

backward_on_loss
optimizer_report
optimizer_step
should_continue
step_loss	param loss

step_loss(loss: torch.Tensor, **moving_averages)

Parameters

• loss –

• moving_averages –

Returns

the value of the loss (float)