[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>

vllm/model_executor/model_loader/loader.py

@@ -1327,6 +1327,7 @@ class RunaiModelStreamerLoader(BaseModelLoader):
         """Prepare weights for the model.
 
         If the model is not local, it will be downloaded."""
+
         is_s3_path = is_s3(model_name_or_path)
         is_local = os.path.isdir(model_name_or_path)
         safetensors_pattern = "*.safetensors"
@@ -1340,7 +1341,6 @@ class RunaiModelStreamerLoader(BaseModelLoader):
                          revision,
                          ignore_patterns=self.load_config.ignore_patterns,
                      ))
-
         if is_s3_path:
             hf_weights_files = s3_glob(path=hf_folder,
                                        allow_pattern=[safetensors_pattern])

vllm/model_executor/model_loader/weight_utils.py

@@ -27,6 +27,8 @@ from vllm.model_executor.layers.quantization import (QuantizationConfig,
 from vllm.platforms import current_platform
 from vllm.utils import PlaceholderModule
 
+logger = init_logger(__name__)
+
 try:
     from runai_model_streamer import SafetensorsStreamer
 except (ImportError, OSError):
@@ -37,8 +39,6 @@ except (ImportError, OSError):
     SafetensorsStreamer = runai_model_streamer.placeholder_attr(
         "SafetensorsStreamer")
 
-logger = init_logger(__name__)
-
 # use system-level temp directory for file locks, so that multiple users
 # can share the same lock without error.
 # lock files in the temp directory will be automatically deleted when the