diff --git a/instill/helpers/ray_config.py b/instill/helpers/ray_config.py
index dcba865..d4a6f69 100644
--- a/instill/helpers/ray_config.py
+++ b/instill/helpers/ray_config.py
@@ -1,3 +1,4 @@
+import os
 from typing import Callable, Optional
 
 import ray
@@ -11,6 +12,7 @@
     DEFAULT_RAY_ACTOR_OPRTIONS,
     DEFAULT_RUNTIME_ENV,
 )
+from instill.helpers.utils import get_dir_size
 
 
 class InstillDeployable:
@@ -21,6 +23,7 @@ def __init__(
         use_gpu: bool,
     ) -> None:
         self._deployment: Deployment = deployable
+        self.use_gpu = use_gpu
         # params
         self.model_weight_or_folder_name: str = model_weight_or_folder_name
         if use_gpu:
@@ -37,6 +40,17 @@ def _update_num_gpus(self, num_gpus: float):
         if self._deployment.ray_actor_options is not None:
             self._deployment.ray_actor_options.update({"num_gpus": num_gpus})
 
+    def _determine_vram_usage(self, model_path: str, vram: str):
+        if vram == "":
+            return 0.25
+        if os.path.isfile(model_path):
+            return (
+                1.1 * os.path.getsize(model_path) / (1024 * 1024 * 1024) / float(vram)
+            )
+        if os.path.isdir(model_path):
+            return 1.1 * get_dir_size(model_path) / (1024 * 1024 * 1024) / float(vram)
+        return 0.25
+
     def update_min_replicas(self, num_replicas: int):
         new_autoscaling_config = DEFAULT_AUTOSCALING_CONFIG
         new_autoscaling_config["min_replicas"] = num_replicas
@@ -51,15 +65,20 @@ def update_max_replicas(self, num_replicas: int):
             autoscaling_config=new_autoscaling_config
         )
 
-    def deploy(self, model_folder_path: str, ray_addr: str):
+    def deploy(self, model_folder_path: str, ray_addr: str, vram: str):
         if not ray.is_initialized():
             ray_addr = "ray://" + ray_addr.replace("9000", "10001")
             ray.init(address=ray_addr, runtime_env=DEFAULT_RUNTIME_ENV)
+
         model_path = "/".join([model_folder_path, self.model_weight_or_folder_name])
         model_path_string_parts = model_path.split("/")
         application_name = model_path_string_parts[5]
         model_name = "_".join(model_path_string_parts[3].split("#")[:2])
         route_prefix = f'/{model_name}/{model_path_string_parts[3].split("#")[3]}'
+
+        if self.use_gpu:
+            self._update_num_gpus(self._determine_vram_usage(model_path, vram))
+
         serve.run(
             self._deployment.options(name=application_name).bind(model_path),
             name=model_name,
diff --git a/instill/helpers/utils.py b/instill/helpers/utils.py
new file mode 100644
index 0000000..a81877d
--- /dev/null
+++ b/instill/helpers/utils.py
@@ -0,0 +1,12 @@
+import os
+
+
+def get_dir_size(path):
+    total = 0
+    with os.scandir(path) as it:
+        for entry in it:
+            if entry.is_file():
+                total += entry.stat().st_size
+            elif entry.is_dir():
+                total += get_dir_size(entry.path)
+    return total