diff --git a/python/ray/air/config.py b/python/ray/air/config.py
index 7515b731282d..fb2abe5d3355 100644
--- a/python/ray/air/config.py
+++ b/python/ray/air/config.py
@@ -206,7 +206,7 @@ def _resources_per_worker_not_none(self):
 
         if self.accelerator_type:
             accelerator = f"{RESOURCE_CONSTRAINT_PREFIX}{self.accelerator_type}"
-            resources_per_worker.setdefault(accelerator, 1)
+            resources_per_worker.setdefault(accelerator, 0.001)
         return resources_per_worker
 
     @property
diff --git a/python/ray/air/tests/test_api.py b/python/ray/air/tests/test_api.py
index 6ce8910e20ff..fcdd8f42dd9a 100644
--- a/python/ray/air/tests/test_api.py
+++ b/python/ray/air/tests/test_api.py
@@ -149,14 +149,14 @@ def test_scaling_config_accelerator_type():
     }
     assert scaling_config._resources_per_worker_not_none == {
         "GPU": 1,
-        "accelerator_type:A100": 1,
+        "accelerator_type:A100": 0.001,
     }
     assert scaling_config.additional_resources_per_worker == {
-        "accelerator_type:A100": 1
+        "accelerator_type:A100": 0.001
     }
     assert scaling_config.as_placement_group_factory().bundles == [
-        {"GPU": 1, "accelerator_type:A100": 1, "CPU": 1},
-        {"GPU": 1, "accelerator_type:A100": 1},
+        {"GPU": 1, "accelerator_type:A100": 0.001, "CPU": 1},
+        {"GPU": 1, "accelerator_type:A100": 0.001},
     ]
 
     # With resources_per_worker
@@ -172,15 +172,15 @@ def test_scaling_config_accelerator_type():
     assert scaling_config._resources_per_worker_not_none == {
         "GPU": 1,
         "custom_resource": 1,
-        "accelerator_type:A100": 1,
+        "accelerator_type:A100": 0.001,
     }
     assert scaling_config.additional_resources_per_worker == {
         "custom_resource": 1,
-        "accelerator_type:A100": 1,
+        "accelerator_type:A100": 0.001,
     }
     assert scaling_config.as_placement_group_factory().bundles == [
-        {"GPU": 1, "custom_resource": 1, "accelerator_type:A100": 1, "CPU": 1},
-        {"GPU": 1, "custom_resource": 1, "accelerator_type:A100": 1},
+        {"GPU": 1, "custom_resource": 1, "accelerator_type:A100": 0.001, "CPU": 1},
+        {"GPU": 1, "custom_resource": 1, "accelerator_type:A100": 0.001},
     ]
 
     # With trainer_resources
@@ -195,14 +195,14 @@ def test_scaling_config_accelerator_type():
     }
     assert scaling_config._resources_per_worker_not_none == {
         "GPU": 1,
-        "accelerator_type:A100": 1,
+        "accelerator_type:A100": 0.001,
     }
     assert scaling_config.additional_resources_per_worker == {
-        "accelerator_type:A100": 1
+        "accelerator_type:A100": 0.001
     }
     assert scaling_config.as_placement_group_factory().bundles == [
-        {"GPU": 1, "accelerator_type:A100": 1, "memory": 10 * 1024**3},
-        {"GPU": 1, "accelerator_type:A100": 1},
+        {"GPU": 1, "accelerator_type:A100": 0.001, "memory": 10 * 1024**3},
+        {"GPU": 1, "accelerator_type:A100": 0.001},
     ]
 
 
diff --git a/python/ray/train/tests/test_data_parallel_trainer.py b/python/ray/train/tests/test_data_parallel_trainer.py
index 19b67a44de12..c40b28537f89 100644
--- a/python/ray/train/tests/test_data_parallel_trainer.py
+++ b/python/ray/train/tests/test_data_parallel_trainer.py
@@ -51,9 +51,11 @@ def ray_start_heterogenous_cluster():
             cluster.add_node(
                 num_cpus=4,
                 num_gpus=4,
-                resources={f"{RESOURCE_CONSTRAINT_PREFIX}{accelerator_type}": 4}
-                if accelerator_type
-                else {},
+                resources=(
+                    {f"{RESOURCE_CONSTRAINT_PREFIX}{accelerator_type}": 1.0}
+                    if accelerator_type
+                    else {}
+                ),
             )
 
     ray.init(address=cluster.address)