Use bigger runners

Signed-off-by: Huy Do <huydhn@gmail.com>

Author: huydhn

.github/workflows/gpt-oss-benchmark.yml

@@ -26,13 +26,13 @@ jobs:
             model: openai/gpt-oss-120b
             docker-image: rocm/vllm-dev:open-mi300-08052025
           # gpt-oss-20b
-          - runner: linux.aws.h100
+          - runner: linux.aws.h100.4
             model: openai/gpt-oss-20b
             docker-image: 'public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:6d8d0a24c02bfd84d46b3016b865a44f048ae84b'
-          - runner: linux.dgx.b200
+          - runner: linux.dgx.b200.8
             model: openai/gpt-oss-20b
             docker-image: 'public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:6d8d0a24c02bfd84d46b3016b865a44f048ae84b'
-          - runner: linux.rocm.gpu.gfx942.2
+          - runner: linux.rocm.gpu.gfx942.8
             model: openai/gpt-oss-20b
             docker-image: rocm/vllm-dev:open-mi300-08052025
       fail-fast: false

vllm-benchmarks/benchmarks/rocm/serving-tests.json

@@ -421,7 +421,8 @@
             "disable_log_stats": "",
             "disable_log_requests": "",
             "load_format": "dummy",
-            "max_model_len": 8192
+            "max_model_len": 8192,
+            "compilation-config": "{'full_cuda_graph': true}"
         },
         "client_parameters": {
             "model": "",
@@ -437,11 +438,12 @@
         "server_parameters": {
             "model": "openai/gpt-oss-120b",
             "tensor_parallel_size": 2,
-            "swap_space": 128,
+            "swap_space": 16,
             "disable_log_stats": "",
             "disable_log_requests": "",
             "load_format": "dummy",
-            "max_model_len": 8192
+            "max_model_len": 8192,
+            "compilation-config": "{'full_cuda_graph': true}"
         },
         "client_parameters": {
             "model": "",
@@ -457,11 +459,12 @@
         "server_parameters": {
             "model": "openai/gpt-oss-120b",
             "tensor_parallel_size": 4,
-            "swap_space": 128,
+            "swap_space": 16,
             "disable_log_stats": "",
             "disable_log_requests": "",
             "load_format": "dummy",
-            "max_model_len": 8192
+            "max_model_len": 8192,
+            "compilation-config": "{'full_cuda_graph': true}"
         },
         "client_parameters": {
             "model": "",