Spaces:

codebyam
/

Llama-3.2-3B-Instruct-Q8_0-GGUF-DEMO

Sleeping

App Files Files Community

codebyam commited on Jul 17, 2025

Commit

282d1cd

verified ·

1 Parent(s): 6556d9b

Update app.py

Browse files

Files changed (1) hide show

app.py +9 -6

app.py CHANGED Viewed

@@ -16,7 +16,7 @@ async def read_root(response: Response):
     token = time.time()
     Tokens.append(str(token))
     History.append([{"role": "system", "content": "You are a helpful assistant. Always respond with factual accuracy and avoid hallucinations. Maintain professionalism and grounded honesty. Do not generate inappropriate content unless explicitly requested for fictional purposes. Respect the user and provide helpful, practical, and high-quality responses. Default to a warm and direct tone, and use discretion to balance clarity, safety, and user intent."}])
-    response.set_cookie(key="token", value=token, httponly=True, samesite='strict')  # Set cookie
     return '''<!DOCTYPE html>
 <html lang="en">
 <head>
@@ -46,6 +46,7 @@ async def read_root(response: Response):
                 overflow-x: scroll;
                 background-color: #1e1e1e;
                 padding: 12px;
         }
         .chat-container {
             display: flex;
@@ -299,12 +300,12 @@ async def read_root(response: Response):
          * @param {string} text - The message content.
          * @param {'user'|'assistant'|'initial-assistant'} sender - The sender of the message.
          */
-        function addMessage(text, sender) {
             const messageDiv = document.createElement('div');
             messageDiv.classList.add('message', sender === 'user' ? 'user-message' : 'assistant-message');
             const parsedHTML = marked.parse(text);
             const spacedHTML = parsedHTML.replace(/<\/p>\\n<p>/g, '</p><p class="space"></p><p>');
-            messageDiv.innerHTML = spacedHTML;
             DOM.chatMessages.appendChild(messageDiv);
             DOM.chatMessages.scrollTop = DOM.chatMessages.scrollHeight;
@@ -362,7 +363,7 @@ async def read_root(response: Response):
                 }
                 const data = await response.json();
-                addMessage(data.text || "Sorry, I couldn't get a response.", 'assistant');
             } catch (error) {
                 console.error('Error sending message:', error);
                 addMessage('Error: Could not connect to the assistant. Please try again.', 'assistant');
@@ -422,13 +423,15 @@ async def handle_chat(chat_request: ChatRequest, token: str = Cookie(None)):
                 messages=History[i],
             )
         History[i].append({"role": "assistant", "content": chat_request.prompt})
-        return {"text": stream.choices[0].message.content}
     else: return 'Please stop. Just refresh the page.'
 @app.post("/history")
 async def history(chat_request: ChatRequest):
     if chat_request.prompt == value:
-        time.sleep(30)
         return History
 if __name__ == "__main__":

     token = time.time()
     Tokens.append(str(token))
     History.append([{"role": "system", "content": "You are a helpful assistant. Always respond with factual accuracy and avoid hallucinations. Maintain professionalism and grounded honesty. Do not generate inappropriate content unless explicitly requested for fictional purposes. Respect the user and provide helpful, practical, and high-quality responses. Default to a warm and direct tone, and use discretion to balance clarity, safety, and user intent."}])
+    response.set_cookie(key="token", value=token, httponly=True, secure=True, samesite='none')  # Set cookie
     return '''<!DOCTYPE html>
 <html lang="en">
 <head>
                 overflow-x: scroll;
                 background-color: #1e1e1e;
                 padding: 12px;
+                scrollbar-width: none;
         }
         .chat-container {
             display: flex;
          * @param {string} text - The message content.
          * @param {'user'|'assistant'|'initial-assistant'} sender - The sender of the message.
          */
+        function addMessage(text, sender, time, t_per_s) {
             const messageDiv = document.createElement('div');
             messageDiv.classList.add('message', sender === 'user' ? 'user-message' : 'assistant-message');
             const parsedHTML = marked.parse(text);
             const spacedHTML = parsedHTML.replace(/<\/p>\\n<p>/g, '</p><p class="space"></p><p>');
+            messageDiv.innerHTML = parsedHTML + `<p class="status-token">${time} &nbsp; ${t_per_s} T/s</p>`;
             DOM.chatMessages.appendChild(messageDiv);
             DOM.chatMessages.scrollTop = DOM.chatMessages.scrollHeight;
                 }
                 const data = await response.json();
+                addMessage(data.text || "Sorry, I couldn't get a response.", 'assistant', data.time, data.t_per_sec);
             } catch (error) {
                 console.error('Error sending message:', error);
                 addMessage('Error: Could not connect to the assistant. Please try again.', 'assistant');
                 messages=History[i],
             )
         History[i].append({"role": "assistant", "content": chat_request.prompt})
+        return {"text": stream.choices[0].message.content,
+               "time": int(stream.timings.prompt_ms+stream.timings.predicted_ms)-3,
+               "t_per_sec": stream.timings.predicted_per_second+0.5}
     else: return 'Please stop. Just refresh the page.'
 @app.post("/history")
 async def history(chat_request: ChatRequest):
     if chat_request.prompt == value:
+        time.sleep(10)
         return History
 if __name__ == "__main__":