[Misc][ModelScope] Change to use runtime VLLM_USE_MODELSCOPE

Signed-off-by: Mengqing Cao <cmq0113@163.com>

Author: MengqingCao

vllm/model_executor/model_loader/default_loader.py

@@ -11,8 +11,8 @@
 from torch import nn
 from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
 
+from vllm import envs
 from vllm.config import LoadConfig, LoadFormat, ModelConfig, VllmConfig
-from vllm.envs import VLLM_USE_MODELSCOPE
 from vllm.logger import init_logger
 from vllm.model_executor.model_loader.base_loader import BaseModelLoader
 from vllm.model_executor.model_loader.utils import (
@@ -64,7 +64,7 @@ def _maybe_download_from_modelscope(
 
         Returns the path to the downloaded model, or None if the model is not
         downloaded from ModelScope."""
-        if VLLM_USE_MODELSCOPE:
+        if envs.VLLM_USE_MODELSCOPE:
             # download model from ModelScope hub,
             # lazy import so that modelscope is not required for normal use.
             # pylint: disable=C.

vllm/transformers_utils/__init__.py

@@ -1,8 +1,8 @@
 # SPDX-License-Identifier: Apache-2.0
 
-from vllm.envs import VLLM_USE_MODELSCOPE
+from vllm import envs
 
-if VLLM_USE_MODELSCOPE:
+if envs.VLLM_USE_MODELSCOPE:
     try:
         # Patch here, before each import happens
         import modelscope

vllm/transformers_utils/config.py

@@ -24,7 +24,7 @@
     MODEL_FOR_CAUSAL_LM_MAPPING_NAMES)
 from transformers.utils import CONFIG_NAME as HF_CONFIG_NAME
 
-from vllm.envs import VLLM_USE_MODELSCOPE
+from vllm import envs
 from vllm.logger import init_logger
 # yapf conflicts with isort for this block
 # yapf: disable
@@ -45,7 +45,7 @@
 from vllm.transformers_utils.utils import check_gguf_file
 from vllm.utils import resolve_obj_by_qualname
 
-if VLLM_USE_MODELSCOPE:
+if envs.VLLM_USE_MODELSCOPE:
     from modelscope import AutoConfig
 else:
     from transformers import AutoConfig
@@ -130,7 +130,7 @@ def lookup_files() -> list[str]:
             ]
         # if model is remote, use hf_hub api to list files
         try:
-            if VLLM_USE_MODELSCOPE:
+            if envs.VLLM_USE_MODELSCOPE:
                 from vllm.transformers_utils.utils import (
                     modelscope_list_repo_files)
                 return modelscope_list_repo_files(repo_id,
@@ -768,7 +768,7 @@ def get_hf_image_processor_config(
     **kwargs,
 ) -> dict[str, Any]:
     # ModelScope does not provide an interface for image_processor
-    if VLLM_USE_MODELSCOPE:
+    if envs.VLLM_USE_MODELSCOPE:
         return dict()
     # Separate model folder from file path for GGUF models
     if check_gguf_file(model):

vllm/transformers_utils/tokenizer.py

@@ -13,7 +13,7 @@
 from transformers import (AutoTokenizer, PreTrainedTokenizer,
                           PreTrainedTokenizerFast)
 
-from vllm.envs import VLLM_USE_MODELSCOPE
+from vllm import envs
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.transformers_utils.tokenizer_base import (TokenizerBase,
@@ -168,7 +168,7 @@ def get_tokenizer(
 ) -> AnyTokenizer:
     """Gets a tokenizer for the given model name via HuggingFace or ModelScope.
     """
-    if VLLM_USE_MODELSCOPE:
+    if envs.VLLM_USE_MODELSCOPE:
         # download model from ModelScope hub,
         # lazy import so that modelscope is not required for normal use.
         # pylint: disable=C.