add passsage templates

app.py

@@ -5,7 +5,21 @@ import json
 import requests
 import random
 
-API_URL = "http://bore.testsprep.online:8082/v1/chat/completions"
+API_URL = ""
+
+
+LA_SERVICE_URL_v1 = "http://bore.testsprep.online:8082/v1/chat/completions"
+LA_SERVICE_URL_v2 = "http://bore.testsprep.online:8083/v1/chat/completions"
+LA_SERVICE_URL_v3 = "http://bore.testsprep.online:8084/v1/chat/completions"
+LA_SERVICE_URL_v4 = "http://bore.testsprep.online:8085/v1/chat/completions"
+
+MODEL2SERVICE = {
+    'LA-llama-3.1-7b-16k-sft-awq': LA_SERVICE_URL_v1,
+    'LA-storm-llama-3.1-7b-16k-sft-awq': LA_SERVICE_URL_v2,
+    'LA-cohere-aya-expanse-8b-16k-sft-awq': LA_SERVICE_URL_v3,
+    'LA-qwen2.5-7b-16k-sft-awq': LA_SERVICE_URL_v4,
+}
+
 DISABLED = os.getenv("DISABLED") == 'True'
 BEARER_TOKEN = "Prep@123"
 NUM_THREADS = 16
@@ -22,7 +36,7 @@ sys.excepthook = exception_handler
 sys.tracebacklimit = 0
 
 
-def predict(system_prompt, inputs, top_p, temperature, max_tokens, chat_counter, chatbot, history, request: gr.Request):
+def predict(model_selector, system_prompt, inputs, top_p, temperature, max_tokens, chat_counter, chatbot, history, request: gr.Request):
     messages = [{"role": "system", "content": system_prompt}] if system_prompt else []
 
     headers = {
@@ -87,7 +101,7 @@ def predict(system_prompt, inputs, top_p, temperature, max_tokens, chat_counter,
         if payload:
             print(f"\n>>> Payload: {payload}")
             # Gọi API với stream=True
-            response = requests.post(API_URL, headers=headers, json=payload, stream=True)
+            response = requests.post(MODEL2SERVICE[model_selector], headers=headers, json=payload, stream=True)
 
             for chunk in response.iter_lines():
                 if counter == 0:
@@ -125,6 +139,9 @@ with gr.Blocks(
     gr.HTML(title)
 
     with gr.Column(elem_id="col_container", visible=True) as main_block:
+
+        model_selector = gr.Dropdown(choices=list(MODEL2SERVICE.keys()), label="Select Model", value=list(MODEL2SERVICE.keys())[0])
+
         chatbot = gr.Chatbot(elem_id='chatbot')
         inputs = gr.Textbox(placeholder="Hi there!", label="Type an input and press Enter")
         state = gr.State([])
@@ -145,10 +162,10 @@ with gr.Blocks(
             chat_counter = gr.Number(value=0, visible=False, precision=0)
 
     inputs.submit(reset_textbox, [], [inputs, b1], queue=False)
-    inputs.submit(predict, [system_prompt, inputs, top_p, temperature, max_tokens, chat_counter, chatbot, state],
+    inputs.submit(predict, [model_selector, system_prompt, inputs, top_p, temperature, max_tokens, chat_counter, chatbot, state],
                   [chatbot, state, chat_counter, server_status_code, inputs, b1])
     b1.click(reset_textbox, [], [inputs, b1], queue=False)
-    b1.click(predict, [system_prompt, inputs, top_p, temperature, max_tokens, chat_counter, chatbot, state],
+    b1.click(predict, [model_selector, system_prompt, inputs, top_p, temperature, max_tokens, chat_counter, chatbot, state],
              [chatbot, state, chat_counter, server_status_code, inputs, b1])
 
     demo.queue(max_size=10, default_concurrency_limit=NUM_THREADS, api_open=False).launch(share=False)