deleting the downloaded models from the memory

on_server_start.py

@@ -1,17 +1,24 @@
 """
 A script that is run when the server starts.
 """
+import os
 
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+DOWNLOADED_MODELS_DIR = os.path.join(os.path.dirname(__file__), 'models')
+
 
 def download_model(model_name: str):
     """
-    Downloads the model with the given name.
+    Downloads a model from hugging face hub to the disk but not to the RAM.
     :param model_name: The name of the model to download.
     """
-    AutoModelForCausalLM.from_pretrained(model_name)
-    AutoTokenizer.from_pretrained(model_name)
+    print(f"Downloading model: {model_name}")
+    model = AutoModelForCausalLM.from_pretrained(model_name)
+    del model
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    del tokenizer
+    print(f"Downloaded model: {model_name}")
 
 
 def download_useful_models():
@@ -19,6 +26,8 @@ def download_useful_models():
     Downloads the models that are useful for this project.
     So that the user doesn't have to wait for the models to download when they first use the app.
     """
+    print("Downloading useful models...")
+    os.makedirs(DOWNLOADED_MODELS_DIR, exist_ok=True)
     useful_models = (
         "gpt2",
         "EleutherAI/gpt-j-6B",