Update app.py

app.py

@@ -21,27 +21,29 @@ def http_bot(messages, argsbox):
     print(messages)
     print(argsbox)
 
-    baseurl="https://hkust.azure-api.net"
-    # client = OpenAI(api_key=args["api_key"], base_url = args["base_url"])
-    client = AzureOpenAI(
-        api_version="2023-07-01-preview",
-        azure_endpoint=baseurl,
-        api_key=args["api_key"],
-    )
-    OpenAI(timeout=60, max_retries=3, base_url=baseurl,api_key=["api_key"])
-    # n = 0
-    # while True:
-        # try:
-    chat_completion = client.chat.completions.create(
-        messages=messages["messages"],
-        model=args["model"],
-        temperature=float(args["temperature"]),
-        max_tokens=int(args["max_tokens"])
-    )
-        #     break
-        # except Exception as e:
-        #     continue
-    print(chat_completion)
+    if args["model"]=="gpt-35-turbo":
+        baseurl="https://hkust.azure-api.net"
+        client = AzureOpenAI(
+            api_version="2023-07-01-preview",
+            azure_endpoint=baseurl,
+            api_key=args["api_key"],
+        )
+        chat_completion = client.chat.completions.create(
+            messages=messages["messages"],
+            model=args["model"],
+            temperature=float(args["temperature"]),
+            max_tokens=int(args["max_tokens"])
+        )
+        print(chat_completion)
+    else:
+        client = OpenAI(api_key=args["api_key"], base_url = args["base_url"])
+        chat_completion = client.chat.completions.create(
+            messages=messages,
+            model=args["model"],
+            temperature=float(args["temperature"]),
+            max_tokens=int(args["max_tokens"])
+        )
+        print(chat_completion)
     return chat_completion.choices[0].message.content
 
 
@@ -49,8 +51,8 @@ def http_bot(messages, argsbox):
 with gr.Blocks() as demo:
     gr.Markdown("# vLLM text completion demo\n")
     inputbox = gr.Textbox(label="Input",
-                          placeholder="Enter text and press ENTER")
-    argsbox = gr.Textbox(label="Args", placeholder="a dict of {api_key, base_url, model, temperature, max_tokens}")
+                          placeholder="Dict of infos")
+    argsbox = gr.Textbox(label="Args", placeholder="a dict of {api_key, base_url, model, temperature, max_tokens,....}")
     outputbox = gr.Textbox(label="Output",
                            placeholder="Generated result from the model")
     submit = gr.Button("Submit")