Migrate Pytorch To Accelerator0 26 1

Last update

Jun 10, 2024

Short description

This codemod converts existing PyTorch code that follows the standard naming conventions to use HuggingFace Accelerate 0.26.1.

Detailed description

This codemod converts existing PyTorch code that follows the standard naming conventions to use HuggingFace Accelerate 0.26.1 so that ML code can easily run in a distributed manner. The changes include the additional import statements and some changes when device is concerned.

https://huggingface.co/docs/accelerate/v0.26.1/en/basic_tutorials/migration

Examples

Before

import torch

device = "cuda"
model.to(device)

for batch in training_dataloader:
    optimizer.zero_grad()
    inputs, targets = batch
    inputs = inputs.to(device)
    targets = targets.to(device)
    outputs = model(inputs)
    loss = loss_function(outputs, targets)
    loss.backward()
    optimizer.step()
    scheduler.step()

After

import torch
import accelerator

device = accelerator.device
model, optimizer, training_dataloader, scheduler = accelerator.prepare(
  model, optimizer, training_dataloader, scheduler
)

for batch in training_dataloader:
    optimizer.zero_grad()
    inputs, targets = batch
    outputs = model(inputs)
    loss = loss_function(outputs, targets)
    accelerator.backward(loss)
    optimizer.step()
    scheduler.step()