diff --git a/litellm/llms/replicate.py b/litellm/llms/replicate.py
index 4e14fa50d6e..5329e3874d3 100644
--- a/litellm/llms/replicate.py
+++ b/litellm/llms/replicate.py
@@ -4,7 +4,14 @@ import requests
 import time
 from typing import Callable
 from litellm.utils import ModelResponse
-import tiktoken
+
+class ReplicateError(Exception):
+    def __init__(self, status_code, message):
+        self.status_code = status_code
+        self.message = message
+        super().__init__(
+            self.message
+        )  # Call the base class constructor with the parameters it needs
 
 # Function to start a prediction and get the prediction URL
 def start_prediction(version_id, input_data, api_token):
@@ -25,7 +32,7 @@ def start_prediction(version_id, input_data, api_token):
         response_data = response.json()
         return response_data.get("urls", {}).get("get")
     else:
-        raise ValueError(response.status_code, "Failed to start prediction.")
+        raise ReplicateError(response.status_code, "Failed to start prediction.")
 
 # Function to handle prediction response (non-streaming)
 def handle_prediction_response(prediction_url, api_token, print_verbose):
@@ -87,7 +94,7 @@ def completion(
     print_verbose: Callable,
     logging_obj,
     api_key,
-    encoding=tiktoken.get_encoding("cl100k_base"),
+    encoding,
     optional_params=None,
     litellm_params=None,
     logger_fn=None,