Rename LightningLite (6/n) (#15982)

Author: awaelchli

.azure/gpu-tests-fabric.yml

@@ -23,8 +23,8 @@ pr:
     include:
       - ".actions/**"
       - ".azure/gpu-tests-fabric.yml"
-      - "examples/lite/**"
-      - "examples/run_lite_examples.sh"
+      - "examples/fabric/**"
+      - "examples/run_fabric_examples.sh"
       - "tests/tests_lite/run_standalone_*.sh"
       - "tests/tests_pytorch/run_standalone_tests.sh"  # used by Lite through a symlink
       - "requirements/fabric/**"
@@ -126,9 +126,9 @@ jobs:
 
     - script: |
         # In order to run the examples, we need to substitute the meta package imports with the standalone package
-        python ../.actions/assistant.py copy_replace_imports --source_dir="./lite" --source_import="lightning.fabric" --target_import="lightning_fabric.fabric"
+        python ../.actions/assistant.py copy_replace_imports --source_dir="./fabric" --source_import="lightning.fabric" --target_import="lightning_fabric.fabric"
         set -e
-        bash run_lite_examples.sh --accelerator=cuda --devices=1
-        bash run_lite_examples.sh --accelerator=cuda --devices=2 --strategy ddp
+        bash run_fabric_examples.sh --accelerator=cuda --devices=1
+        bash run_fabric_examples.sh --accelerator=cuda --devices=2 --strategy ddp
       workingDirectory: examples
       displayName: 'Testing: Lite examples'

.github/checkgroup.yml

@@ -204,8 +204,8 @@ subprojects:
     paths:
       - ".actions/**"
       - ".azure/gpu-tests-fabric.yml"
-      - "examples/lite/**"
-      - "examples/run_lite_examples.sh"
+      - "examples/fabric/**"
+      - "examples/run_fabric_examples.sh"
       - "tests/tests_lite/run_standalone_*.sh"
       - "tests/tests_pytorch/run_standalone_tests.sh"  # used by Lite through a symlink
       - "requirements/fabric/**"

examples/README.md

@@ -11,13 +11,13 @@ ______________________________________________________________________
 
 ______________________________________________________________________
 
-## Lightning Lite Examples
+## Lightning Fabric Examples
 
-We show how to accelerate your PyTorch code with [Lightning Lite](https://pytorch-lightning.readthedocs.io/en/latest/starter/lightning_lite.html) with minimal code changes.
+We show how to accelerate your PyTorch code with [Lightning Fabric](https://pytorch-lightning.readthedocs.io/en/latest/starter/lightning_fabric.html) with minimal code changes.
 You stay in full control of the training loop.
 
-- [MNIST with vanilla PyTorch](lite/image_classifier_1_pytorch.py)
-- [MNIST with Lightning Lite](lite/image_classifier_2_lite.py)
+- [MNIST with vanilla PyTorch](fabric/image_classifier_1_pytorch.py)
+- [MNIST with Lightning Fabric](fabric/image_classifier_2_fabric.py)
 
 ______________________________________________________________________
 

examples/app_multi_node/README.md

@@ -18,15 +18,15 @@ or you can use the built-in component for it.
 lightning run app train_pytorch_spawn.py
 ```
 
-## Multi Node with raw PyTorch + Lite
+## Multi Node with raw PyTorch + Fabric
 
-You can run the multi-node raw PyTorch and Lite by running the following commands.
+You can run the multi-node raw PyTorch and Fabric by running the following commands.
 
 ```bash
-lightning run app train_lite.py
+lightning run app train_fabric.py
 ```
 
-Using Lite, you retain control over your loops while accessing in a minimal way all Lightning distributed strategies.
+Using Fabric, you retain control over your loops while accessing in a minimal way all Lightning distributed strategies.
 
 ## Multi Node with Lightning Trainer
 

examples/app_multi_node/train_lite.py renamed to examples/app_multi_node/train_fabric.py

@@ -15,19 +15,19 @@ def run(self):
         )
 
         # 2. Create Fabric.
-        lite = Fabric(strategy="ddp", precision=16)
-        model, optimizer = lite.setup(model, torch.optim.SGD(model.parameters(), lr=0.01))
+        fabric = Fabric(strategy="ddp", precision=16)
+        model, optimizer = fabric.setup(model, torch.optim.SGD(model.parameters(), lr=0.01))
         criterion = torch.nn.MSELoss()
 
         # 3. Train the model for 1000 steps.
         for step in range(1000):
             model.zero_grad()
-            x = torch.tensor([0.8]).to(lite.device)
-            target = torch.tensor([1.0]).to(lite.device)
+            x = torch.tensor([0.8]).to(fabric.device)
+            target = torch.tensor([1.0]).to(fabric.device)
             output = model(x)
             loss = criterion(output, target)
-            print(f"global_rank: {lite.global_rank} step: {step} loss: {loss}")
-            lite.backward(loss)
+            print(f"global_rank: {fabric.global_rank} step: {step} loss: {loss}")
+            fabric.backward(loss)
             optimizer.step()
 
 

examples/fabric/README.md

@@ -0,0 +1,31 @@
+## MNIST Examples
+
+Here are two MNIST classifiers implemented in PyTorch.
+The first one is implemented in pure PyTorch, but isn't easy to scale.
+The second one is using [Lightning Fabric](https://pytorch-lightning.readthedocs.io/en/stable/starter/lightning_fabric.html) to accelerate and scale the model.
+
+#### 1. Image Classifier with Vanilla PyTorch
+
+Trains a simple CNN over MNIST using vanilla PyTorch. It only supports singe GPU training.
+
+```bash
+# CPU
+python image_classifier_1_pytorch.py
+```
+
+______________________________________________________________________
+
+#### 2. Image Classifier with Lightning Fabric
+
+This script shows you how to scale the pure PyTorch code to enable GPU and multi-GPU training using [Lightning Fabric](https://pytorch-lightning.readthedocs.io/en/stable/starter/lightning_fabric.html).
+
+```bash
+# CPU
+lightning run model image_classifier_2_fabric.py
+
+# GPU (CUDA or M1 Mac)
+lightning run model image_classifier_2_fabric.py --accelerator=gpu
+
+# Multiple GPUs
+lightning run model image_classifier_2_fabric.py --accelerator=gpu --devices=4
+```

examples/lite/image_classifier_1_pytorch.py renamed to examples/fabric/image_classifier_1_pytorch.py

@@ -14,7 +14,7 @@
 
 """Here are 4 easy steps to use Fabric in your PyTorch code.
 
-1. Create the Lightning Lite object at the beginning of your script.
+1. Create the Lightning Fabric object at the beginning of your script.
 
 2. Remove all ``.to`` and ``.cuda`` calls since Fabric will take care of it.
 
@@ -25,7 +25,7 @@
 
 Accelerate your training loop by setting the ``--accelerator``, ``--strategy``, ``--devices`` options directly from
 the command line. See ``lightning run model --help`` or learn more from the documentation:
-https://pytorch-lightning.readthedocs.io/en/latest/starter/lightning_lite.lite.html.
+https://pytorch-lightning.readthedocs.io/en/latest/starter/lightning_fabric.html.
 """
 
 import argparse
@@ -47,18 +47,18 @@
 
 
 def run(hparams):
-    # Create the Lightning Lite object. The parameters like accelerator, strategy, devices etc. will be proided
+    # Create the Lightning Fabric object. The parameters like accelerator, strategy, devices etc. will be proided
     # by the command line. See all options: `lightning run model --help`
-    lite = Fabric()
+    fabric = Fabric()
 
-    lite.hparams = hparams
+    fabric.hparams = hparams
     seed_everything(hparams.seed)  # instead of torch.manual_seed(...)
 
     transform = T.Compose([T.ToTensor(), T.Normalize((0.1307,), (0.3081,))])
     # This is meant to ensure the data are download only by 1 process.
-    if lite.is_global_zero:
+    if fabric.is_global_zero:
         MNIST(DATASETS_PATH, download=True)
-    lite.barrier()
+    fabric.barrier()
     train_dataset = MNIST(DATASETS_PATH, train=True, transform=transform)
     test_dataset = MNIST(DATASETS_PATH, train=False, transform=transform)
     train_loader = torch.utils.data.DataLoader(
@@ -68,19 +68,19 @@ def run(hparams):
     test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=hparams.batch_size)
 
     # don't forget to call `setup_dataloaders` to prepare for dataloaders for distributed training.
-    train_loader, test_loader = lite.setup_dataloaders(train_loader, test_loader)
+    train_loader, test_loader = fabric.setup_dataloaders(train_loader, test_loader)
 
     model = Net()  # remove call to .to(device)
     optimizer = optim.Adadelta(model.parameters(), lr=hparams.lr)
 
     # don't forget to call `setup` to prepare for model / optimizer for distributed training.
     # the model is moved automatically to the right device.
-    model, optimizer = lite.setup(model, optimizer)
+    model, optimizer = fabric.setup(model, optimizer)
 
     scheduler = StepLR(optimizer, step_size=1, gamma=hparams.gamma)
 
     # use torchmetrics instead of manually computing the accuracy
-    test_acc = Accuracy().to(lite.device)
+    test_acc = Accuracy().to(fabric.device)
 
     # EPOCH LOOP
     for epoch in range(1, hparams.epochs + 1):
@@ -92,7 +92,7 @@ def run(hparams):
             optimizer.zero_grad()
             output = model(data)
             loss = F.nll_loss(output, target)
-            lite.backward(loss)  # instead of loss.backward()
+            fabric.backward(loss)  # instead of loss.backward()
 
             optimizer.step()
             if (batch_idx == 0) or ((batch_idx + 1) % hparams.log_interval == 0):
@@ -130,18 +130,18 @@ def run(hparams):
                     break
 
         # all_gather is used to aggregated the value across processes
-        test_loss = lite.all_gather(test_loss).sum() / len(test_loader.dataset)
+        test_loss = fabric.all_gather(test_loss).sum() / len(test_loader.dataset)
 
         print(f"\nTest set: Average loss: {test_loss:.4f}, Accuracy: ({100 * test_acc.compute():.0f}%)\n")
         test_acc.reset()
 
         if hparams.dry_run:
             break
 
-    # When using distributed training, use `lite.save`
+    # When using distributed training, use `fabric.save`
     # to ensure the current process is allowed to save a checkpoint
     if hparams.save_model:
-        lite.save(model.state_dict(), "mnist_cnn.pt")
+        fabric.save(model.state_dict(), "mnist_cnn.pt")
 
 
 if __name__ == "__main__":