[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/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

vllm/transformers_utils/config.py

@@ -144,7 +144,6 @@ def file_exists(
     revision: Optional[str] = None,
     token: Union[str, bool, None] = None,
 ) -> bool:
-
     file_list = list_repo_files(repo_id,
                                 repo_type=repo_type,
                                 revision=revision,
@@ -498,7 +497,7 @@ def get_sentence_transformer_tokenizer_config(model: str,
             if encoder_dict:
                 break
 
-    if not encoder_dict:
+    if not encoder_dict and not model.startswith("/"):
         try:
             # If model is on HuggingfaceHub, get the repo files
             repo_files = list_repo_files(model,

vllm/transformers_utils/s3_utils.py

@@ -46,6 +46,8 @@ def glob(s3=None,
     """
     if s3 is None:
         s3 = boto3.client("s3")
+    if not path.endswith("/"):
+        path = path + "/"
     bucket_name, _, paths = list_files(s3,
                                        path=path,
                                        allow_pattern=allow_pattern)
@@ -109,6 +111,7 @@ class S3Model:
         for sig in (signal.SIGINT, signal.SIGTERM):
             existing_handler = signal.getsignal(sig)
             signal.signal(sig, self._close_by_signal(existing_handler))
+
         self.dir = tempfile.mkdtemp()
 
     def __del__(self):
@@ -140,6 +143,9 @@ class S3Model:
             ignore_pattern: A list of patterns of which files not to pull.
 
         """
+        if not s3_model_path.endswith("/"):
+            s3_model_path = s3_model_path + "/"
+
         bucket_name, base_dir, files = list_files(self.s3, s3_model_path,
                                                   allow_pattern,
                                                   ignore_pattern)
@@ -147,8 +153,9 @@ class S3Model:
             return
 
         for file in files:
-            destination_file = os.path.join(self.dir,
-                                            file.removeprefix(base_dir))
+            destination_file = os.path.join(
+                self.dir,
+                file.removeprefix(base_dir).lstrip("/"))
             local_dir = Path(destination_file).parent
             os.makedirs(local_dir, exist_ok=True)
             self.s3.download_file(bucket_name, file, destination_file)