DeepSpeedPrecisionPlugin

class lightning.pytorch.plugins.precision.DeepSpeedPrecisionPlugin(precision)

Bases: PrecisionPlugin

Precision plugin for DeepSpeed integration.

Warning

This is an experimental feature.

Parameters:

precision (Literal['32-true', '16-mixed', 'bf16-mixed']) – Full precision (32), half precision (16) or bfloat16 precision (bf16).

Raises:

ValueError – If unsupported precision is provided.

backward(tensor, model, optimizer, *args, **kwargs)

Performs back-propagation using DeepSpeed’s engine.

Parameters:

• tensor (Tensor) – the loss tensor

• model (LightningModule) – the model to be optimized

• optimizer (Optional[Steppable]) – ignored for DeepSpeed

• *args (Any) – additional positional arguments for the deepspeed.DeepSpeedEngine.backward() call

• **kwargs (Any) – additional keyword arguments for the deepspeed.DeepSpeedEngine.backward() call

Return type:

None

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

DeepSpeed handles gradient clipping internally.

Return type:

None

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

Hook to run the optimizer step.

Return type:

Any