[Data] Update data ingest release tests (ray-project#49406)

Signed-off-by: Balaji Veeramani <[email protected]>

Author: bveeramani

release/nightly_tests/dataset/autoscaling_gpu_compute.yaml

@@ -8,12 +8,16 @@ advanced_configurations_json:
 
 head_node_type:
     name: head-node
-    instance_type: m5.2xlarge
+    # Use a GPU head node so we can test `iter_torch_batches` with `device='cuda'`.
+    # We use the same number of vCPUs and memory as the default m5.2xlarge head node.
+    instance_type: g4dn.2xlarge
     resources:
       cpu: 0
 
 worker_node_types:
     - name: worker-node
+      # Anyscale workspaces use m5.2xlarge worker nodes by default. For consistency, we 
+      # use GPU nodes with the same number of vCPUs and memory.
       instance_type: g4dn.2xlarge
       min_workers: 0
       max_workers: 10

release/nightly_tests/dataset/data_ingest_benchmark.py

@@ -0,0 +1,20 @@
+cloud_id: {{env["ANYSCALE_CLOUD_ID"]}}
+region: us-west-2
+
+advanced_configurations_json:
+    IamInstanceProfile: {"Name": "ray-autoscaler-v1"}
+
+head_node_type:
+    name: head-node
+    # Default head node type on Anyscale workspaces.
+    instance_type: m5.2xlarge
+    resources:
+      cpu: 0
+
+worker_node_types:
+    - name: worker-node
+      # Default worker node type on Anyscale workspaces.
+      instance_type: m5.2xlarge
+      min_workers: 10
+      max_workers: 10
+      use_spot: false

release/nightly_tests/dataset/data_ingest_benchmark_compute.yaml

@@ -79,8 +79,10 @@ def consume_fn(ds):
 
     elif args.iter_torch_batches:
 
+        # In addition to consuming the data, we also want to test the performance of
+        # moving data to GPU.
         def consume_fn(ds):
-            for _ in ds.iter_torch_batches():
+            for _ in ds.iter_torch_batches(device="cuda"):
                 pass
 
     elif args.to_tf: