IPUPrecisionPlugin

class pytorch_lightning.plugins.precision.IPUPrecisionPlugin(precision)

Bases: pytorch_lightning.plugins.precision.precision_plugin.PrecisionPlugin

Precision plugin for IPU integration.

Raises:

ValueError – If the precision is neither 16 nor 32.

backward(model, *_, **__)

Performs the actual backpropagation.

Parameters:

• model (LightningModule) – the model to be optimized

• closure_loss – the loss value obtained from the closure

• optimizer – current optimizer being used. None if using manual optimization

Return type:

None

clip_gradients(optimizer, clip_val=0.0, gradient_clip_algorithm=GradClipAlgorithmType.NORM)

Clips the gradients.

Return type:

None

optimizer_step(model, optimizer, optimizer_idx, closure, **kwargs)

IPUs handle the optimizer step internally.

Return type:

Any