[ci] Use env var to control whether to use S3 bucket in CI (#13634)

tests/mq_llm_engine/test_abort.py

@@ -10,8 +10,8 @@ import pytest
 from tests.mq_llm_engine.utils import RemoteMQLLMEngine, generate
 from vllm.engine.arg_utils import AsyncEngineArgs
 
-MODEL = "s3://vllm-ci-model-weights/gemma-1.1-2b-it"
-ENGINE_ARGS = AsyncEngineArgs(model=MODEL, load_format="runai_streamer")
+MODEL = "google/gemma-1.1-2b-it"
+ENGINE_ARGS = AsyncEngineArgs(model=MODEL)
 RAISED_ERROR = KeyError
 RAISED_VALUE = "foo"
 EXPECTED_TOKENS = 250

tests/mq_llm_engine/test_error_handling.py

@@ -21,10 +21,8 @@ from vllm.lora.request import LoRARequest
 from vllm.usage.usage_lib import UsageContext
 from vllm.utils import FlexibleArgumentParser
 
-MODEL = "s3://vllm-ci-model-weights/gemma-1.1-2b-it"
-ENGINE_ARGS = AsyncEngineArgs(model=MODEL,
-                              load_format="runai_streamer",
-                              enforce_eager=True)
+MODEL = "google/gemma-1.1-2b-it"
+ENGINE_ARGS = AsyncEngineArgs(model=MODEL, enforce_eager=True)
 RAISED_ERROR = KeyError
 RAISED_VALUE = "foo"
 

tests/mq_llm_engine/test_load.py

@@ -10,14 +10,12 @@ import pytest
 from tests.mq_llm_engine.utils import RemoteMQLLMEngine, generate
 from vllm.engine.arg_utils import AsyncEngineArgs
 
-MODEL = "s3://vllm-ci-model-weights/gemma-1.1-2b-it"
+MODEL = "google/gemma-1.1-2b-it"
 NUM_EXPECTED_TOKENS = 10
 NUM_REQUESTS = 10000
 
 # Scenarios to test for num generated token.
-ENGINE_ARGS = AsyncEngineArgs(model=MODEL,
-                              load_format="runai_streamer",
-                              disable_log_requests=True)
+ENGINE_ARGS = AsyncEngineArgs(model=MODEL, disable_log_requests=True)
 
 
 @pytest.fixture(scope="function")