IPUPrecisionPlugin

class lightning.pytorch.plugins.precision.IPUPrecisionPlugin(precision)[source]

Bases: PrecisionPlugin

Precision plugin for IPU integration.

Warning

This is an experimental feature.

Raises:

ValueError – If the precision is neither 16-mixed nor 32-true.

backward(tensor, model, *args, **kwargs)[source]

Performs the actual backpropagation.

Parameters:

• tensor (Tensor) – the loss value obtained from the closure

• model (LightningModule) – the model to be optimized

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

• *args (Any) – Positional arguments intended for the actual function that performs the backward, like backward().

• **kwargs (Any) – Keyword arguments for the same purpose as *args.

Return type:

None

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

Clips the gradients.

Return type:

None

optimizer_step(optimizer, model, closure, **kwargs)[source]

IPUs handle the optimizer step internally.

Return type:

Any