Make _GPUAccelerator.get_parallel_devices fit GPUAccelerator API (#240)

The PTL Accelerator API is expected to return a List, and not None. This PR updates our _GPUAccelerator abstraction to fit this API.

Signed-off-by: amogkam <[email protected]>

Author: amogkam

.github/workflows/test.yaml

@@ -16,7 +16,7 @@ jobs:
       run: |
         python -m pip install --upgrade pip
         python -m pip install codecov
-        python -m pip install -U yapf==0.23.0 flake8==3.9.1 flake8-comprehensions flake8-quotes
+        python -m pip install -U -r requirements-lint.txt
     - name: Run format script
       run: |
         ./format.sh --all

ray_lightning/accelerators/delayed_gpu_accelerator.py

@@ -42,7 +42,7 @@ def get_parallel_devices(devices: List[int]) -> List[torch.device]:
         if devices:
             return [torch.device("cuda", i) for i in devices]
         else:
-            return None
+            return []
 
     @staticmethod
     def is_available() -> bool:

requirements-lint.txt

@@ -0,0 +1,5 @@
+flake8==3.9.1
+flake8-comprehensions
+flake8-quotes
+importlib_metadata==4.13.0
+yapf==0.23.0

requirements-test.txt

@@ -1,14 +1,10 @@
 fairscale
-flake8==3.9.1
-flake8-comprehensions
-flake8-quotes
-yapf==0.23.0
 pytest
 pytorch-lightning==1.6.4
 lightning-bolts==0.3.3
 ray[tune]
 torch==1.12.0
-torchmetrics
+torchmetrics==0.9.3
 torchvision
 protobuf<=3.20.1
 jsonargparse>=4.13.2