accelerate API reference

81 public APIs from accelerate (huggingface/accelerate) — 27 classes, 26 functions, 28 methods. Signatures extracted by static analysis of the actual source.

Repository: huggingface/accelerate

Kind	Count
Classes	27
Functions	26
Methods	28

API list

methodsrc.accelerate.accelerator.Accelerator.context_parallel_rank() -> int

Context parallelism is not supported yet.

methodsrc.accelerate.accelerator.Accelerator.is_main_process()

True for one process only.

methodsrc.accelerate.accelerator.Accelerator.lomo_backward(loss:torch.Tensor, learning_rate:float) -> None

Runs backward pass on LOMO optimizers.

methodsrc.accelerate.accelerator.Accelerator.pipeline_parallel_rank() -> int

Pipeline parallelism is not supported yet.

methodsrc.accelerate.accelerator.Accelerator.tensor_parallel_rank() -> int

Returns the local rank for tensor parallelism.

classsrc.accelerate.commands.menu.input.KeyHandler

Metaclass that adds the key handlers to the class

funcsrc.accelerate.commands.menu.keymap.get_raw_chars()

Gets raw characters from inputs

funcsrc.accelerate.hooks.add_hook_to_module(module:nn.Module, hook:ModelHook, append:bool=False)

Adds a hook to a given module.

classsrc.accelerate.local_sgd.LocalSGD

A helper class to support local SGD on top of Accelerator.

classsrc.accelerate.logging.MultiProcessAdapter

An adapter to assist with logging in multiprocess.

classsrc.accelerate.optimizer.AcceleratedOptimizer

Internal wrapper around a torch optimizer.

methodsrc.accelerate.optimizer.AcceleratedOptimizer.eval()

Sets the optimizer to "eval" mode.

classsrc.accelerate.tracking.AimTracker

A `Tracker` class that supports `aim`.

methodsrc.accelerate.tracking.AimTracker.finish()

Closes `aim` writer

methodsrc.accelerate.tracking.AimTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

classsrc.accelerate.tracking.ClearMLTracker

A `Tracker` class that supports `clearml`.

methodsrc.accelerate.tracking.ClearMLTracker.finish()

Close the ClearML task.

methodsrc.accelerate.tracking.ClearMLTracker.log_images(values:dict, step:Optional[int]=None, **kwargs)

Logs `images` to the current run.

classsrc.accelerate.tracking.CometMLTracker

A `Tracker` class that supports `comet_ml`.

methodsrc.accelerate.tracking.CometMLTracker.finish()

Flush `comet-ml` writer

methodsrc.accelerate.tracking.CometMLTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

classsrc.accelerate.tracking.DVCLiveTracker

A `Tracker` class that supports `dvclive`.

methodsrc.accelerate.tracking.DVCLiveTracker.finish()

Closes `dvclive.Live()`.

methodsrc.accelerate.tracking.DVCLiveTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

classsrc.accelerate.tracking.MLflowTracker

A `Tracker` class that supports `mlflow`.

methodsrc.accelerate.tracking.MLflowTracker.finish()

End the active MLflow run.

methodsrc.accelerate.tracking.MLflowTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

methodsrc.accelerate.tracking.MLflowTracker.log_figure(figure:Any, artifact_file:str, **save_kwargs)

Logs an figure to the current run.

classsrc.accelerate.tracking.SwanLabTracker

A `Tracker` class that supports `swanlab`.

methodsrc.accelerate.tracking.SwanLabTracker.finish()

Closes `swanlab` writer

methodsrc.accelerate.tracking.SwanLabTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

methodsrc.accelerate.tracking.SwanLabTracker.log_images(values:dict, step:Optional[int]=None, **kwargs)

Logs `images` to the current run.

classsrc.accelerate.tracking.TensorBoardTracker

A `Tracker` class that supports `tensorboard`.

methodsrc.accelerate.tracking.TensorBoardTracker.finish()

Closes `TensorBoard` writer

methodsrc.accelerate.tracking.TensorBoardTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

methodsrc.accelerate.tracking.TensorBoardTracker.log_images(values:dict, step:Optional[int]=None, **kwargs)

Logs `images` to the current run.

classsrc.accelerate.tracking.TrackioTracker

A `Tracker` class that supports `trackio`.

methodsrc.accelerate.tracking.TrackioTracker.finish()

Closes `trackio` run

methodsrc.accelerate.tracking.TrackioTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

classsrc.accelerate.tracking.WandBTracker

A `Tracker` class that supports `wandb`.

methodsrc.accelerate.tracking.WandBTracker.finish()

Closes `wandb` writer

methodsrc.accelerate.tracking.WandBTracker.log(values:dict, step:Optional[int]=None, **kwargs)

Logs `values` to the current run.

methodsrc.accelerate.tracking.WandBTracker.log_images(values:dict, step:Optional[int]=None, **kwargs)

Logs `images` to the current run.

classsrc.accelerate.utils.dataclasses.ComputeEnvironment

Represents a type of the compute environment.

classsrc.accelerate.utils.dataclasses.DeepSpeedPlugin

This plugin is used to integrate DeepSpeed.

classsrc.accelerate.utils.dataclasses.DistributedType

Represents a type of distributed environment.

classsrc.accelerate.utils.dataclasses.FP8BackendType

Represents the backend used for FP8.

classsrc.accelerate.utils.dataclasses.FP8RecipeKwargs

Deprecated.

classsrc.accelerate.utils.dataclasses.SageMakerDistributedType

Represents a type of distributed environment.

classsrc.accelerate.utils.deepspeed.DeepSpeedOptimizerWrapper

Internal wrapper around a deepspeed optimizer.

classsrc.accelerate.utils.deepspeed.DeepSpeedSchedulerWrapper

Internal wrapper around a deepspeed scheduler.

funcsrc.accelerate.utils.environment.set_numa_affinity(local_process_index:int, verbose:Optional[bool]=None) -> None

Assigns the current process to a specific NUMA node.

funcsrc.accelerate.utils.fsdp_utils.fsdp2_prepare_model(accelerator, model:torch.nn.Module) -> torch.nn.Module

Prepares the model for FSDP2 in-place.

funcsrc.accelerate.utils.imports.is_fp16_available()

Checks if fp16 is supported

funcsrc.accelerate.utils.imports.is_fp8_available()

Checks if fp8 is supported

funcsrc.accelerate.utils.launch.setup_fp8_env(args:argparse.Namespace, current_env:dict[str, str])

Setup the FP8 environment variables.

classsrc.accelerate.utils.megatron_lm.AbstractTrainStep

Abstract class for batching, forward pass and loss handler.

classsrc.accelerate.utils.megatron_lm.BertTrainStep

Bert train step class.

funcsrc.accelerate.utils.megatron_lm.BertTrainStep.forward_step(data_iterator, model)

Forward step.

funcsrc.accelerate.utils.megatron_lm.BertTrainStep.get_batch_megatron(data_iterator)

Build the batch.

funcsrc.accelerate.utils.megatron_lm.BertTrainStep.get_batch_transformer(data_iterator)

Build the batch.

classsrc.accelerate.utils.megatron_lm.GPTTrainStep

GPT train step class.

funcsrc.accelerate.utils.megatron_lm.GPTTrainStep.forward_step(data_iterator, model)

Forward step.

funcsrc.accelerate.utils.megatron_lm.GPTTrainStep.get_batch_megatron(data_iterator)

Generate a batch

classsrc.accelerate.utils.megatron_lm.T5TrainStep

T5 train step class.

funcsrc.accelerate.utils.megatron_lm.T5TrainStep.forward_step(data_iterator, model)

Forward step.

funcsrc.accelerate.utils.megatron_lm.T5TrainStep.get_batch_megatron(data_iterator)

Build the batch.

funcsrc.accelerate.utils.megatron_lm.T5TrainStep.get_batch_transformer(data_iterator)

Build the batch.

funcsrc.accelerate.utils.modeling.find_tied_parameters(model:torch.nn.Module, **kwargs) -> list[list[str]]

Find the tied parameters in a given model.

funcsrc.accelerate.utils.modeling.id_tensor_storage(tensor:torch.Tensor) -> tuple[torch.device, int, int]

Unique identifier to a tensor storage.

classsrc.accelerate.utils.offload.PrefixedDataset

Will access keys in a given dataset by adding a prefix.

classsrc.accelerate.utils.operations.DistributedOperationException

An exception class for distributed operations.

funcsrc.accelerate.utils.other.get_free_port() -> int

Gets a free port on `localhost`.

funcsrc.accelerate.utils.other.get_pretty_name(obj)

Gets a pretty name from `obj`.

funcsrc.accelerate.utils.other.has_repeated_blocks(module:torch.nn.Module) -> bool

Check whether the module has repeated blocks, i.e.

funcsrc.accelerate.utils.other.is_compiled_module(module:torch.nn.Module) -> bool

Check whether the module was compiled with torch.compile()

funcsrc.accelerate.utils.other.is_port_in_use(port:Optional[int]=None) -> bool

Checks if a port is in use on `localhost`.

funcsrc.accelerate.utils.other.is_repeated_blocks(module:torch.nn.Module) -> bool

Check whether the module is a repeated block, i.e.

funcsrc.accelerate.utils.other.model_has_dtensor(model:torch.nn.Module) -> bool

Check if the model has DTensor parameters.

funcsrc.accelerate.utils.other.recursive_getattr(obj, attr:str)

Recursive `getattr`.

funcsrc.accelerate.utils.other.save(obj, f, save_on_each_node:bool=False, safe_serialization:bool=False)

Save the data to disk.

About this data

These signatures were extracted from the public source of huggingface/accelerate using Python's ast module. Argument names, default values, type annotations and return types are taken verbatim from the code. Implementation bodies are never stored. See how it works for details.

Back to all 805 libraries