[1/n][CI] Load models in CI from S3 instead of HF (#13205)

Signed-off-by: <>
Co-authored-by: EC2 Default User <ec2-user@ip-172-31-20-117.us-west-2.compute.internal>

tests/entrypoints/llm/test_prompt_validation.py

@@ -3,6 +3,7 @@
 import pytest
 
 from vllm import LLM
+from vllm.config import LoadFormat
 
 
 @pytest.fixture(autouse=True)
@@ -14,13 +15,17 @@ def v1(run_with_both_engines):
 
 
 def test_empty_prompt():
-    llm = LLM(model="gpt2", enforce_eager=True)
+    llm = LLM(model="s3://vllm-ci-model-weights/gpt2",
+              load_format=LoadFormat.RUNAI_STREAMER,
+              enforce_eager=True)
     with pytest.raises(ValueError, match='Prompt cannot be empty'):
         llm.generate([""])
 
 
 @pytest.mark.skip_v1
 def test_out_of_vocab_token():
-    llm = LLM(model="gpt2", enforce_eager=True)
+    llm = LLM(model="s3://vllm-ci-model-weights/gpt2",
+              load_format=LoadFormat.RUNAI_STREAMER,
+              enforce_eager=True)
     with pytest.raises(ValueError, match='out of vocabulary'):
         llm.generate({"prompt_token_ids": [999999]})