[ci] Separate vllm tests (#2930)

Author: xyang16

.github/workflows/integration.yml

@@ -157,6 +157,9 @@ jobs:
           - test: TestVllm1
             instance: g6
             failure-prefix: lmi
+          - test: TestVllm2
+            instance: g6
+            failure-prefix: lmi
           - test: TestVllmCustomHandlers
             instance: g6
             failure-prefix: lmi

tests/integration/tests.py

@@ -587,6 +587,11 @@ def test_llama_68m_speculative_medusa(self):
             r.launch()
             client.run("vllm llama-68m-speculative-medusa".split())
 
+
+@pytest.mark.vllm
+@pytest.mark.gpu_4
+class TestVllm2:
+
     def test_llama_68m_speculative_eagle(self):
         with Runner('lmi', 'llama-68m-speculative-eagle') as r:
             prepare.build_vllm_async_model("llama-68m-speculative-eagle")