Download Support for Custom HuggingFace Models

llm/download.py

@@ -15,7 +15,6 @@
 import uuid
 from typing import List
 import huggingface_hub as hfh
-from huggingface_hub.utils import HfHubHTTPError
 from utils.marsgen import get_mar_name, generate_mars
 from utils.system_utils import (
     check_if_path_exists,
@@ -176,65 +175,68 @@ def read_config_for_download(gen_model: GenerateDataModel) -> GenerateDataModel:
                         "repo_version"
                     ]
 
-                # Make sure there is HF hub token for LLAMA(2)
-                if (
-                    gen_model.repo_info.repo_id.startswith("meta-llama")
-                    and gen_model.repo_info.hf_token is None
-                ):
-                    print(
-                        "## Error: HuggingFace Hub token is required for llama download."
-                        " Please specify it using --hf_token=<your token>. "
-                        "Refer https://huggingface.co/docs/hub/security-tokens"
-                    )
-                    sys.exit(1)
-
-                # Validate downloaded files
-                hf_api = hfh.HfApi()
-                hf_api.list_repo_commits(
-                    repo_id=gen_model.repo_info.repo_id,
-                    revision=gen_model.repo_info.repo_version,
-                    token=gen_model.repo_info.hf_token,
-                )
-
                 # Read handler file name
                 if not gen_model.mar_utils.handler_path:
                     gen_model.mar_utils.handler_path = os.path.join(
                         os.path.dirname(__file__),
                         models[gen_model.model_name]["handler"],
                     )
 
-            except (KeyError, HfHubHTTPError):
+                # Validate hf_token
+                gen_model.validate_hf_token()
+
+                # Validate repository info
+                gen_model.validate_commit_info()
+
+            except (KeyError, ValueError):
                 print(
-                    "## Error: Please check either repo_id, repo_version"
-                    " or HuggingFace ID is not correct\n"
+                    "## There seems to be an error in the model_config.json file. "
+                    "Please check the same."
                 )
                 sys.exit(1)
+
         else:  # Custom model case
-            if not gen_model.skip_download:
-                print(
-                    "## Please check your model name,"
-                    " it should be one of the following : "
-                )
-                print(list(models.keys()))
-                print(
-                    "\n## If you want to use custom model files,"
-                    " use the '--no_download' argument"
-                )
-                sys.exit(1)
+            gen_model.is_custom_model = True
+            if gen_model.skip_download:
+                if check_if_folder_empty(gen_model.mar_utils.model_path):
+                    print("## Error: The given model path folder is empty\n")
+                    sys.exit(1)
 
-            if check_if_folder_empty(gen_model.mar_utils.model_path):
-                print("## Error: The given model path folder is empty\n")
-                sys.exit(1)
+                if not gen_model.repo_info.repo_version:
+                    gen_model.repo_info.repo_version = "1.0"
+
+            else:
+                if not gen_model.repo_info.repo_id:
+                    print(
+                        "## If you want to create a model archive file with the supported models, "
+                        "make sure you're model name is present in the below : "
+                    )
+                    print(list(models.keys()))
+                    print(
+                        "\nIf you want to create a model archive file for"
+                        " a custom model,there are two methods:\n"
+                        "1. If you have already downloaded the custom model"
+                        " files, please include"
+                        " the --no_download flag and provide the model_path "
+                        "directory which contains the model files.\n"
+                        "2. If you need to download the model files, provide "
+                        "the HuggingFace repository ID using 'repo_id'"
+                        " along with an empty model_path driectory where the "
+                        "model files will be downloaded.\n"
+                    )
+                    sys.exit(1)
+
+                # Validate hf_token
+                gen_model.validate_hf_token()
+
+                # Validate repository info
+                gen_model.validate_commit_info()
 
             if not gen_model.mar_utils.handler_path:
                 gen_model.mar_utils.handler_path = os.path.join(
                     os.path.dirname(__file__), "handler.py"
                 )
 
-            if not gen_model.repo_info.repo_version:
-                gen_model.repo_info.repo_version = "1.0"
-
-            gen_model.is_custom_model = True
             print(
                 f"\n## Generating MAR file for "
                 f"custom model files: {gen_model.model_name}"
@@ -260,7 +262,9 @@ def run_download(gen_model: GenerateDataModel) -> GenerateDataModel:
         GenerateDataModel: An instance of the GenerateDataModel class.
     """
     if not check_if_folder_empty(gen_model.mar_utils.model_path):
-        print("## Make sure the path provided to download model files is empty\n")
+        print(
+            "## Make sure the model_path provided to download model files through is empty\n"
+        )
         sys.exit(1)
 
     print(
@@ -290,7 +294,9 @@ def create_mar(gen_model: GenerateDataModel) -> None:
     Args:
         gen_model (GenerateDataModel): An instance of the GenerateDataModel dataclass
     """
-    if not gen_model.is_custom_model and not check_if_model_files_exist(gen_model):
+    if not (
+        gen_model.is_custom_model and gen_model.skip_download
+    ) and not check_if_model_files_exist(gen_model):
         print("## Model files do not match HuggingFace repository files")
         sys.exit(1)
 
@@ -347,13 +353,20 @@ def run_script(params: argparse.Namespace) -> bool:
         type=str,
         default="",
         required=True,
-        metavar="mn",
+        metavar="n",
         help="Name of model",
     )
+    parser.add_argument(
+        "--repo_id",
+        type=str,
+        default=None,
+        metavar="ri",
+        help="HuggingFace repository ID (In case of custom model download)",
+    )
     parser.add_argument(
         "--repo_version",
         type=str,
-        default="",
+        default=None,
         metavar="rv",
         help="Commit ID of models repo from HuggingFace repository",
     )
@@ -367,15 +380,15 @@ def run_script(params: argparse.Namespace) -> bool:
         type=str,
         default="",
         required=True,
-        metavar="mp",
+        metavar="p",
         help="Absolute path of model files (should be empty if downloading)",
     )
     parser.add_argument(
         "--mar_output",
         type=str,
         default="",
         required=True,
-        metavar="mx",
+        metavar="a",
         help="Absolute path of exported MAR file (.mar)",
     )
     parser.add_argument(
@@ -389,7 +402,7 @@ def run_script(params: argparse.Namespace) -> bool:
         "--hf_token",
         type=str,
         default=None,
-        metavar="hft",
+        metavar="ht",
         help="HuggingFace Hub token to download LLAMA(2) models",
     )
     parser.add_argument("--debug", action="store_true", help="flag to debug")

llm/tests/test_download.py

@@ -60,7 +60,7 @@ def cleanup_folders() -> None:
 
 def set_generate_args(
     model_name: str = MODEL_NAME,
-    repo_version: str = "",
+    repo_version: str = None,
     model_path: str = MODEL_PATH,
     mar_output: str = MODEL_STORE,
     handler_path: str = "",
@@ -82,6 +82,7 @@ def set_generate_args(
     args.model_path = model_path
     args.mar_output = mar_output
     args.no_download = False
+    args.repo_id = None
     args.repo_version = repo_version
     args.handler_path = handler_path
     args.debug = False
@@ -250,16 +251,20 @@ def test_skip_download_success() -> None:
         assert result is True
 
 
-def custom_model_setup() -> None:
+def custom_model_setup(download_model: bool = True) -> None:
     """
     This function is used to setup custom model case.
     It runs download.py to download model files and
     deletes the contents of 'model_config.json' after
     making a backup.
+
+    Args:
+        download (bool): Set to download model files (defaults to True)
     """
     download_setup()
-    args = set_generate_args()
-    download.run_script(args)
+    if download_model:
+        args = set_generate_args()
+        download.run_script(args)
 
     # creating a backup of original model_config.json
     copy_file(MODEL_CONFIG_PATH, MODEL_TEMP_CONFIG_PATH)
@@ -277,9 +282,9 @@ def custom_model_restore() -> None:
     cleanup_folders()
 
 
-def test_custom_model_success() -> None:
+def test_custom_model_skip_download_success() -> None:
     """
-    This function tests the custom model case.
+    This function tests the no download custom model case.
     This is done by clearing the 'model_config.json' and
     generating the 'GPT2' MAR file.
     Expected result: Success.
@@ -296,6 +301,62 @@ def test_custom_model_success() -> None:
     custom_model_restore()
 
 
+def test_custom_model_download_success() -> None:
+    """
+    This function tests the download custom model case.
+    This is done by clearing the 'model_config.json' and
+    generating the 'GPT2' MAR file.
+    Expected result: Success.
+    """
+    custom_model_setup(download_model=False)
+    args = set_generate_args()
+    args.repo_id = "gpt2"
+    try:
+        result = download.run_script(args)
+    except SystemExit:
+        assert False
+    else:
+        assert result is True
+    custom_model_restore()
+
+
+def test_custom_model_download_wrong_repo_id_throw_error() -> None:
+    """
+    This function tests the download custom model case and
+    passes a wrong repo_id.
+    Expected result: Failure.
+    """
+    custom_model_setup(download_model=False)
+    args = set_generate_args()
+    args.repo_id = "wrong_repo_id"
+    try:
+        download.run_script(args)
+    except SystemExit as e:
+        assert e.code == 1
+    else:
+        assert False
+    custom_model_restore()
+
+
+def test_custom_model_download_wrong_repo_version_throw_error() -> None:
+    """
+    This function tests the download custom model case and
+    passes a correct repo_id but wrong repo_version.
+    Expected result: Failure.
+    """
+    custom_model_setup(download_model=False)
+    args = set_generate_args()
+    args.repo_id = "gpt2"
+    args.repo_version = "wrong_repo_version"
+    try:
+        download.run_script(args)
+    except SystemExit as e:
+        assert e.code == 1
+    else:
+        assert False
+    custom_model_restore()
+
+
 # Run the tests
 if __name__ == "__main__":
     pytest.main(["-v", __file__])

llm/tests/test_torchserve_run.py

@@ -201,9 +201,9 @@ def test_inference_json_file_success() -> None:
         assert False
 
 
-def test_custom_model_success() -> None:
+def test_custom_model_skip_download_success() -> None:
     """
-    This function tests custom model with input folder.
+    This function tests custom model skipping download with input folder.
     Expected result: Success.
     """
     custom_model_setup()
@@ -221,6 +221,26 @@ def test_custom_model_success() -> None:
     process = subprocess.run(["python3", "cleanup.py"], check=False)
 
 
+def test_custom_model_download_success() -> None:
+    """
+    This function tests download custom model input folder.
+    Expected result: Success.
+    """
+    custom_model_setup(download_model=False)
+    args = set_generate_args()
+    args.repo_id = "gpt2"
+    try:
+        download.run_script(args)
+    except SystemExit:
+        assert False
+
+    process = subprocess.run(get_run_cmd(input_path=INPUT_PATH), check=False)
+    assert process.returncode == 0
+
+    custom_model_restore()
+    process = subprocess.run(["python3", "cleanup.py"], check=False)
+
+
 # Run the tests
 if __name__ == "__main__":
     pytest.main(["-v", __file__])