diff --git a/app.py b/app.py
index b8c20c3..bf4968f 100644
--- a/app.py
+++ b/app.py
@@ -44,12 +44,13 @@ def startup_event():
         direct_tokenize = True
 
     model_dir = "./models/model"
-    def get_model_directory() -> str:
+    def get_model_directory() -> (str, bool):
         if os.path.exists(f"{model_dir}/model_name"):
             with open(f"{model_dir}/model_name", "r") as f:
                 model_name = f.read()
-                return f"{model_dir}/{model_name}"
-        return model_dir
+                return f"{model_dir}/{model_name}", True
+        # Default model directory is ./models/model
+        return model_dir, False
 
     def get_onnx_runtime() -> bool:
         if os.path.exists(f"{model_dir}/onnx_runtime"):
@@ -66,13 +67,14 @@ def log_info_about_onnx(onnx_runtime: bool):
                     onnx_quantization_info = f.read()
             logger.info(f"Running ONNX vectorizer with quantized model for {onnx_quantization_info}")
 
+    model_dir, use_sentence_transformer_vectorizer = get_model_directory()
     onnx_runtime = get_onnx_runtime()
     log_info_about_onnx(onnx_runtime)
 
-    meta_config = Meta(get_model_directory())
-    vec = Vectorizer(get_model_directory(), cuda_support, cuda_core, cuda_per_process_memory_fraction,
+    meta_config = Meta(model_dir)
+    vec = Vectorizer(model_dir, cuda_support, cuda_core, cuda_per_process_memory_fraction,
                      meta_config.get_model_type(), meta_config.get_architecture(), 
-                     direct_tokenize, onnx_runtime)
+                     direct_tokenize, onnx_runtime, use_sentence_transformer_vectorizer)
 
 
 @app.get("/.well-known/live", response_class=Response)
diff --git a/download.py b/download.py
index 0b54bf5..e6267de 100755
--- a/download.py
+++ b/download.py
@@ -34,6 +34,10 @@
 if not onnx_cpu_arch:
     onnx_cpu_arch = "arm64"
 
+use_sentence_transformers_vectorizer = os.getenv('USE_SENTENCE_TRANSFORMERS_VECTORIZER')
+if not use_sentence_transformers_vectorizer:
+    use_sentence_transformers_vectorizer = "false"
+
 print(f"Downloading MODEL_NAME={model_name} with FORCE_AUTOMODEL={force_automodel} ONNX_RUNTIME={onnx_runtime} ONNX_CPU={onnx_cpu_arch}")
 
 def download_onnx_model(model_name: str, model_dir: str):
@@ -82,7 +86,7 @@ def download_model(model_name: str, model_dir: str):
     config = AutoConfig.from_pretrained(model_name)
     model_type = config.to_dict()['model_type']
 
-    if model_type is not None and model_type == "t5":
+    if (model_type is not None and model_type == "t5") or use_sentence_transformers_vectorizer.lower() == "true":
         SentenceTransformer(model_name, cache_folder=model_dir)
         with open(f"{model_dir}/model_name", "w") as f:
             f.write(model_name.replace("/", "_"))
diff --git a/vectorizer.py b/vectorizer.py
index 8d13f34..6e403db 100644
--- a/vectorizer.py
+++ b/vectorizer.py
@@ -37,12 +37,13 @@ class VectorInput(BaseModel):
 class Vectorizer:
     executor: ThreadPoolExecutor
 
-    def __init__(self, model_path: str, cuda_support: bool, cuda_core: str, cuda_per_process_memory_fraction: float, model_type: str, architecture: str, direct_tokenize: bool, onnx_runtime: bool):
+    def __init__(self, model_path: str, cuda_support: bool, cuda_core: str, cuda_per_process_memory_fraction: float, 
+                 model_type: str, architecture: str, direct_tokenize: bool, onnx_runtime: bool, use_sentence_transformer_vectorizer: bool):
         self.executor = ThreadPoolExecutor()
         if onnx_runtime:
             self.vectorizer = ONNXVectorizer(model_path)
         else:
-            if model_type == 't5':
+            if model_type == 't5' or use_sentence_transformer_vectorizer:
                 self.vectorizer = SentenceTransformerVectorizer(model_path, cuda_core)
             else:
                 self.vectorizer = HuggingFaceVectorizer(model_path, cuda_support, cuda_core, cuda_per_process_memory_fraction, model_type, architecture, direct_tokenize)