Update standalone tests (#12472)

Author: daniellepintz

tests/callbacks/test_tqdm_progress_bar.py

@@ -582,7 +582,8 @@ def test_progress_bar_max_val_check_interval(
         max_epochs=1,
         enable_model_summary=False,
         val_check_interval=val_check_interval,
-        gpus=world_size,
+        accelerator="gpu",
+        devices=world_size,
         strategy="ddp",
     )
     trainer.fit(model, train_dataloaders=train_data, val_dataloaders=val_data)

tests/strategies/test_ddp_fully_sharded_with_full_state_dict.py

@@ -96,7 +96,9 @@ def test_fully_sharded_strategy_checkpoint(tmpdir):
     """Test to ensure that checkpoint is saved correctly when using a single GPU, and all stages can be run."""
 
     model = TestFSDPModel()
-    trainer = Trainer(default_root_dir=tmpdir, gpus=1, strategy="fsdp", precision=16, max_epochs=1)
+    trainer = Trainer(
+        default_root_dir=tmpdir, accelerator="gpu", devices=1, strategy="fsdp", precision=16, max_epochs=1
+    )
     _run_multiple_stages(trainer, model, os.path.join(tmpdir, "last.ckpt"))
 
 
@@ -106,7 +108,15 @@ def test_fully_sharded_strategy_checkpoint_multi_gpus(tmpdir):
 
     model = TestFSDPModel()
     ck = ModelCheckpoint(save_last=True)
-    trainer = Trainer(default_root_dir=tmpdir, gpus=2, strategy="fsdp", precision=16, max_epochs=1, callbacks=[ck])
+    trainer = Trainer(
+        default_root_dir=tmpdir,
+        accelerator="gpu",
+        devices=2,
+        strategy="fsdp",
+        precision=16,
+        max_epochs=1,
+        callbacks=[ck],
+    )
     _run_multiple_stages(trainer, model)
 
 
@@ -146,7 +156,8 @@ def test_fsdp_gradient_clipping_raises(tmpdir):
         default_root_dir=tmpdir,
         strategy="fsdp",
         fast_dev_run=True,
-        gpus=1,
+        accelerator="gpu",
+        devices=1,
         precision=16,
         gradient_clip_val=1,
         gradient_clip_algorithm="norm",

tests/strategies/test_ddp_strategy.py

@@ -68,7 +68,7 @@ def test_ddp_barrier_non_consecutive_device_ids(barrier_mock, tmpdir):
     """Test correct usage of barriers when device ids do not start at 0 or are not consecutive."""
     model = BoringModel()
     gpus = [1, 3]
-    trainer = Trainer(default_root_dir=tmpdir, max_steps=1, gpus=gpus, strategy="ddp")
+    trainer = Trainer(default_root_dir=tmpdir, max_steps=1, accelerator="gpu", devices=gpus, strategy="ddp")
     trainer.fit(model)
     barrier_mock.assert_any_call(device_ids=[gpus[trainer.local_rank]])
 

tests/strategies/test_ddp_strategy_with_comm_hook.py

@@ -37,7 +37,8 @@ def test_ddp_fp16_compress_comm_hook(tmpdir):
     strategy = DDPStrategy(ddp_comm_hook=default.fp16_compress_hook)
     trainer = Trainer(
         max_epochs=1,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy=strategy,
         default_root_dir=tmpdir,
         sync_batchnorm=True,
@@ -60,7 +61,8 @@ def test_ddp_sgd_comm_hook(tmpdir):
     )
     trainer = Trainer(
         max_epochs=1,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy=strategy,
         default_root_dir=tmpdir,
         sync_batchnorm=True,
@@ -84,7 +86,8 @@ def test_ddp_fp16_compress_wrap_sgd_comm_hook(tmpdir):
     )
     trainer = Trainer(
         max_epochs=1,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy=strategy,
         default_root_dir=tmpdir,
         sync_batchnorm=True,
@@ -104,7 +107,8 @@ def test_ddp_spawn_fp16_compress_comm_hook(tmpdir):
     strategy = DDPSpawnStrategy(ddp_comm_hook=default.fp16_compress_hook)
     trainer = Trainer(
         max_epochs=1,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy=strategy,
         default_root_dir=tmpdir,
         sync_batchnorm=True,
@@ -130,7 +134,8 @@ def test_ddp_post_local_sgd_comm_hook(tmpdir):
     )
     trainer = Trainer(
         fast_dev_run=True,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy=strategy,
         default_root_dir=tmpdir,
         sync_batchnorm=True,
@@ -151,7 +156,8 @@ def test_post_local_sgd_model_averaging(average_parameters_mock, tmpdir):
     # test regular ddp does not call model averaging
     trainer = Trainer(
         fast_dev_run=True,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy="ddp",
         default_root_dir=tmpdir,
         sync_batchnorm=True,
@@ -207,7 +213,8 @@ def configure_optimizers(self):
 
     trainer = Trainer(
         fast_dev_run=True,
-        gpus=2,
+        accelerator="gpu",
+        devices=2,
         strategy=strategy,
         default_root_dir=tmpdir,
         sync_batchnorm=True,