Update app.py

app.py

@@ -24,7 +24,7 @@ def generate_conversation_id():
     return str(uuid.uuid4())[:8]
 
 
-def generate_response(user_message, max_tokens, temperature, top_p, history_state):
+'''def generate_response(user_message, max_tokens, temperature, top_p, history_state):
     if not user_message.strip():
         return history_state, history_state
 
@@ -72,9 +72,75 @@ def generate_response(user_message, max_tokens, temperature, top_p, history_stat
     except Exception:
         pass
 
+    yield new_history, new_history'''
+
+
+import tiktoken                               
+enc = tiktoken.encoding_for_model("gpt-3.5-turbo")  # any OpenAI encoding works
+
+def generate_response(user_message,
+                      max_tokens,
+                      temperature,
+                      top_p,
+                      history_state):
+    if not user_message.strip():
+        return history_state, history_state
+
+    system_message = "Your role as an assistant..."
+    messages = [{"role": "system", "content": system_message}]
+    for m in history_state:
+        messages.append({"role": m["role"], "content": m["content"]})
+    messages.append({"role": "user", "content": user_message})
+
+    try:
+        response = client.chat.completions.create(
+            model="tgi",
+            messages=messages,
+            max_tokens=int(max_tokens),        # server-side limit
+            temperature=temperature,
+            top_p=top_p,
+            stream=True
+        )
+    except Exception as e:
+        print(f"[ERROR] OpenAI API call failed: {e}")
+        yield history_state + [
+            {"role": "user", "content": user_message},
+            {"role": "assistant", "content": "⚠️ Generation failed."}
+        ], history_state
+        return
+
+    assistant_response = ""
+    new_history = history_state + [
+        {"role": "user", "content": user_message},
+        {"role": "assistant", "content": ""}
+    ]
+
+    token_budget = int(max_tokens)
+    tokens_seen = 0
+
+    try:
+        for chunk in response:
+            if (not chunk.choices
+                or not chunk.choices[0].delta
+                or not chunk.choices[0].delta.content):
+                continue
+
+            token_text = chunk.choices[0].delta.content
+            assistant_response += token_text
+            # count how many tokens that piece is worth
+            tokens_seen += len(enc.encode(token_text))
+
+            new_history[-1]["content"] = assistant_response.strip()
+            yield new_history, new_history
+
+            if tokens_seen >= token_budget:
+                break                            # stop the local loop
+    except Exception:
+        pass
+
     yield new_history, new_history
 
-    
+
 example_messages = {
     "JEE Main 2025 Combinatorics": "From all the English alphabets, five letters are chosen and are arranged in alphabetical order. The total number of ways, in which the middle letter is 'M', is?",
     "JEE Main 2025 Coordinate Geometry": "A circle \\(C\\) of radius 2 lies in the second quadrant and touches both the coordinate axes. Let \\(r\\) be the radius of a circle that has centre at the point \\((2, 5)\\) and intersects the circle \\(C\\) at exactly two points. If the set of all possible values of \\(r\\) is the interval \\((\\alpha, \\beta)\\), then \\(3\\beta - 2\\alpha\\) is?",
@@ -91,7 +157,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         <div style="background-color: black; padding: 6px; border-radius: 8px;">
             <img src="https://framerusercontent.com/images/j0KjQQyrUfkFw4NwSaxQOLAoBU.png" alt="Fractal AI Logo" style="height: 48px;">
         </div>
-        <h1 style="margin: 0;">Ramanujan Ganit R1 14B V1 Chatbot</h1>
+        <h1 style="margin: 0;">Ramanujan Ganit R1 14B Chatbot</h1>
     </div>
     """
 )
@@ -109,7 +175,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
             # INTRO TEXT MOVED HERE
             gr.Markdown(
                 """
-                Welcome to the Ramanujan Ganit R1 14B V1 Chatbot, developed by Fractal AI Research! 
+                Welcome to the Ramanujan Ganit R1 14B Chatbot, developed by Fractal AI Research! 
                 
                 Our model excels at reasoning tasks in mathematics and science.