Update app.py

app.py

@@ -2,8 +2,23 @@ import time
 import gradio as gr
 from openai import OpenAI
 
+def format_time(seconds_float):
+    total_seconds = int(round(seconds_float))
+    hours = total_seconds // 3600
+    remaining_seconds = total_seconds % 3600
+    minutes = remaining_seconds // 60
+    seconds = remaining_seconds % 60
+    
+    if hours > 0:
+        return f"{hours}h {minutes}m {seconds}s"
+    elif minutes > 0:
+        return f"{minutes}m {seconds}s"
+    else:
+        return f"{seconds}s"
+
 DESCRIPTION = '''
-# DeepSeek-R1 Distill Qwen-1.5 Demo
+# Duplicate the space for free private inference.
+## DeepSeek-R1 Distill Qwen-1.5 Demo
 A reasoning model trained using RL (Reinforcement Learning) that demonstrates structured reasoning capabilities.
 '''
 
@@ -46,13 +61,14 @@ def user(message, history):
     return "", history + [[message, None]]
 
 class ParserState:
-    __slots__ = ['answer', 'thought', 'in_think', 'start_time', 'last_pos']
+    __slots__ = ['answer', 'thought', 'in_think', 'start_time', 'last_pos', 'total_think_time']
     def __init__(self):
         self.answer = ""
         self.thought = ""
         self.in_think = False
         self.start_time = 0
         self.last_pos = 0
+        self.total_think_time = 0.0
 
 def parse_response(text, state):
     buffer = text[state.last_pos:]
@@ -73,6 +89,9 @@ def parse_response(text, state):
             think_end = buffer.find('</think>')
             if think_end != -1:
                 state.thought += buffer[:think_end]
+                # Calculate duration and accumulate
+                duration = time.perf_counter() - state.start_time
+                state.total_think_time += duration
                 state.in_think = False
                 buffer = buffer[think_end + 8:]
             else:
@@ -87,8 +106,15 @@ def format_response(state, elapsed):
     collapsible = []
 
     if state.thought or state.in_think:
-        status = (f"🌀 Thinking for {elapsed:.0f} seconds" 
-                  if state.in_think else f"✅ Thought for {elapsed:.0f} seconds")
+        if state.in_think:
+            # Ongoing think: total time = accumulated + current elapsed
+            total_elapsed = state.total_think_time + elapsed
+            formatted_time = format_time(total_elapsed)
+            status = f"🌀 Thinking for {formatted_time}"
+        else:
+            # Finished: show total accumulated time
+            formatted_time = format_time(state.total_think_time)
+            status = f"✅ Thought for {formatted_time}"
         collapsible.append(
             f"<details open><summary>{status}</summary>\n\n<div class='thinking-container'>\n{state.thought}\n</div>\n</details>"
         )
@@ -120,13 +146,13 @@ def generate_response(history, temperature, top_p, max_tokens, active_gen):
                 state, elapsed = parse_response(full_response, state)
                 
                 collapsible, answer_part = format_response(state, elapsed)
-                history[-1][1] = "\n\n".join(collapsible + [answer_part])  # Markdown-safe
+                history[-1][1] = "\n\n".join(collapsible + [answer_part])
                 yield history
         
-        # Final update
+        # Final update to ensure all content is parsed
         state, elapsed = parse_response(full_response, state)
         collapsible, answer_part = format_response(state, elapsed)
-        history[-1][1] = "\n\n".join(collapsible + [answer_part])  # Markdown-safe
+        history[-1][1] = "\n\n".join(collapsible + [answer_part])
         yield history
         
     except Exception as e: