diff --git a/tests/v1/kv_connector/unit/test_nixl_connector_hma.py b/tests/v1/kv_connector/unit/test_nixl_connector_hma.py index 0dfe4df735..30913ff98e 100644 --- a/tests/v1/kv_connector/unit/test_nixl_connector_hma.py +++ b/tests/v1/kv_connector/unit/test_nixl_connector_hma.py @@ -195,7 +195,7 @@ def test_fewer_blocks_with_hma(monkeypatch, model_name, sw_size): llm_kwargs = { "model": model_name, "enforce_eager": True, - "gpu_memory_utilization": 0.5, + "gpu_memory_utilization": 0.47, "kv_transfer_config": kv_transfer_config, "max_model_len": 2048, # NOTE: Make sure HMA is enabled