Update app.py

app.py

@@ -7,7 +7,7 @@ import time
 api_key = os.environ.get("RUNPOD_API_KEY")  # Make sure your API key is correctly configured
 client = OpenAI(
     api_key=api_key,
-    base_url="https://api.runpod.ai/v2/vllm-kwgeym0i9drfcy/openai/v1",
+    base_url="https://api.runpod.ai/v2/vllm-7os1t1701rdva2/openai/v1",
 )
 
 def get_response(user_message, history, verbosity):
@@ -25,7 +25,7 @@ def get_response(user_message, history, verbosity):
 
     # Make the API call with a custom stop sequence
     response = client.chat.completions.create(
-        model='ambrosfitz/llama-3-history',
+        model='ambrosfitz/autotrain-nvidia-llama3',
         messages=history_openai_format,
         temperature=temperature,
         max_tokens=512,