[CI] Mark 1-GPU L4 test steps with device: l4 for EKS migration (#54549)

Signed-off-by: khluu <khluu000@gmail.com>

.buildkite/test_areas/distributed.yaml

@@ -20,6 +20,7 @@ steps:
 
 - label: ":nvidia: (L4) Sharded RDT Weight Transfer"
   key: sharded-rdt-weight-transfer
+  device: l4
   timeout_in_minutes: 20
   working_dir: "/vllm-workspace/tests"
   num_devices: 1

.buildkite/test_areas/engine.yaml

@@ -38,6 +38,8 @@ steps:
 
 - label: ":nvidia: (L4) V1 Engine"
   key: engine-1-gpu
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 45
   source_file_dependencies:
     - vllm/v1/engine/

.buildkite/test_areas/kernels.yaml

@@ -104,6 +104,8 @@ steps:
 
 - label: ":nvidia: (L4) Attention Kernels Shard %N"
   key: kernels-attention-test
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 65
   source_file_dependencies:
   - csrc/attention/
@@ -165,6 +167,8 @@ steps:
 
 - label: ":nvidia: (L4) Quantization Kernels Shard %N"
   key: kernels-quantization-test
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 60
   source_file_dependencies:
   - csrc/quantization/
@@ -194,6 +198,8 @@ steps:
 
 - label: ":nvidia: (L4) MoE Kernels Shard %N"
   key: kernels-moe-test
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 50
   source_file_dependencies:
   - csrc/quantization/cutlass_w8a8/moe/

.buildkite/test_areas/models_language.yaml

@@ -92,6 +92,8 @@ steps:
 # correctness test on L4 until its H200 output matches the Transformers reference.
 - label: ":nvidia: (L4) Granite Language Model Compatibility"
   key: language-models-tests-granite-l4-compatibility
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 65
   source_file_dependencies:
   - vllm/

.buildkite/test_areas/pytorch.yaml

@@ -90,6 +90,8 @@ steps:
 
 - label: ":nvidia: (L4) PyTorch Compilation Passes"
   key: pytorch-compilation-passes-unit-tests
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 45
   source_file_dependencies:
     - vllm/__init__.py
@@ -164,6 +166,8 @@ steps:
 # those SM90 cases skip while the architecture-compatible cases still run.
 - label: ":nvidia: (L4) PyTorch Fullgraph CUDAGraph Compatibility"
   key: pytorch-fullgraph-cudagraph-l4-compatibility
+  device: l4
+  num_devices: 1
   timeout_in_minutes: 60
   source_file_dependencies:
   - vllm/__init__.py