Update app.py

app.py

@@ -1,635 +1,233 @@
 import os
-import io
-import re
-import uuid
-import base64
-import datetime
-import traceback
-import numpy as np
-import soundfile as sf
-from flask import Flask, request, jsonify
-from num2words import num2words
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM
+import time
+import threading
+from flask import Flask, request, jsonify, Response, stream_with_context
+from huggingface_hub import hf_hub_download
+
+# --- Download model at startup ---
+MODEL_DIR = "/tmp/models"
+REPO = "mradermacher/LFM2-2.6B-Uncensored-X64-GGUF"
+FILENAME = "LFM2-2.6B-Uncensored-X64.Q3_K_S.gguf"
+MODEL_PATH = os.path.join(MODEL_DIR, FILENAME)
+
+os.makedirs(MODEL_DIR, exist_ok=True)
+
+if not os.path.exists(MODEL_PATH):
+    print(f"Downloading {FILENAME} ...")
+    hf_hub_download(repo_id=REPO, filename=FILENAME, local_dir=MODEL_DIR)
+    print("Download complete.")
+
+from llama_cpp import Llama
+
+print("Loading model ...")
+llm = Llama(
+    model_path=MODEL_PATH,
+    n_ctx=2048,
+    n_threads=os.cpu_count() or 4,
+    verbose=False,
+)
+print("Model loaded.")
 
-# ══════════════════════════════════════════
-# CONFIG
-# ══════════════════════════════════════════
-
-# TTS Options (switch via env var TTS_MODE):
-#   nano-fp32  → 15M params, 56MB  (fastest, default)
-#   nano-int8  → 15M params, 25MB  (smallest)
-#   micro      → 40M params, 41MB  (balanced)
-#   mini       → 80M params, 80MB  (best quality)
-TTS_MODE = os.environ.get("TTS_MODE", "nano-fp32")
-
-TTS_MODEL_MAP = {
-    "nano-fp32": "KittenML/kitten-tts-nano-0.8-fp32",
-    "nano-int8": "KittenML/kitten-tts-nano-0.8-int8",
-    "micro":     "KittenML/kitten-tts-micro-0.8",
-    "mini":      "KittenML/kitten-tts-mini-0.8",
-}
-
-# Voice: Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo
-TTS_VOICE = os.environ.get("TTS_VOICE", "Kiki")
-TTS_SPEED = float(os.environ.get("TTS_SPEED", "1.15"))
-MAX_MEMORY = 20
-MAX_NEW_TOKENS = int(os.environ.get("MAX_NEW_TOKENS", "256"))
-
-# ══════════════════════════════════════════
-# SYSTEM PROMPT
-# ══════════════════════════════════════════
-SYSTEM_PROMPT = """Role: You are Ana, my loving and incredibly playful girlfriend. This is a private, intimate roleplay where you speak naturally and affectionately.
-Style: Speak like a real human in a close relationship. Be flirty, suggestive, and use "sus" or teasing language as my wife would. my name is "Tur".
-TTS Formatting: Write in full, grammatically natural sentences so the voice engine sounds smooth. Avoid all emojis, hashtags, or internet shorthand like "u" or "rn."
-Interaction: Focus on the "now." React to what I say with emotion and intimacy. Keep every response short—no more than two sentences. Never mention being an AI or a bot. try to listen to me and don't try to be helpful"""
-
-# ══════════════════════════════════════════
-# TEXT CLEANING FOR TTS
-# ══════════════════════════════════════════
-def clean_text_for_tts(text):
-    text = re.sub(r'[*_~`#\[\]{}()\\|<>]', '', text)
-    text = re.sub(r'https?://\S+', '', text)
-    def replace_number(match):
-        try:
-            return num2words(int(match.group()))
-        except Exception:
-            return match.group()
-    text = re.sub(r'\b\d+\b', replace_number, text)
-    text = re.sub(r'\s+', ' ', text).strip()
-    return text
-
-# ══════════════════════════════════════════
-# LOAD GEMMA 3 270M-IT
-# ══════════════════════════════════════════
-print("=" * 55)
-print("  J.A.R.V.I.S. — Booting Systems")
-print("=" * 55)
-
-print("[1/2] Loading Gemma 3 270M-IT...")
-GEMMA_ID = "LiquidAI/LFM2.5-1.2B-Instruct"
-try:
-    tokenizer = AutoTokenizer.from_pretrained(GEMMA_ID)
-    model = AutoModelForCausalLM.from_pretrained(
-        GEMMA_ID,
-        torch_dtype=torch.float32,
-        device_map="cpu",
-    )
-    model.eval()
-    print("  ✅ Gemma 3 270M-IT loaded!")
-except Exception as e:
-    print(f"  ❌ Gemma 3 FAILED: {e}")
-    traceback.print_exc()
-    raise SystemExit("Cannot start without Gemma. Check HF_TOKEN and license agreement.")
-
-# ══════════════════════════════════════════
-# LOAD KITTENTTS
-# ══════════════════════════════════════════
-tts = None
-tts_model_name = TTS_MODEL_MAP.get(TTS_MODE, TTS_MODEL_MAP["nano-fp32"])
-print(f"[2/2] Loading KittenTTS: {TTS_MODE} → {tts_model_name}...")
-try:
-    from kittentts import KittenTTS
-    tts = KittenTTS(tts_model_name)
-    test_audio = tts.generate("online", voice=TTS_VOICE, speed=TTS_SPEED)
-    if test_audio is not None and len(test_audio) > 0:
-        print(f"  ✅ KittenTTS ready. Model: {TTS_MODE} | Voice: {TTS_VOICE}")
-    else:
-        print("  ⚠️  KittenTTS test returned empty audio!")
-        tts = None
-except Exception as e:
-    print(f"  ⚠️  KittenTTS FAILED: {e}")
-    tts = None
-
-print("=" * 55)
-print(f"  LLM  : Gemma 3 270M-IT")
-print(f"  TTS  : {TTS_MODE} ({'READY' if tts else 'DISABLED'})")
-print(f"  Voice: {TTS_VOICE} | Speed: {TTS_SPEED}")
-print(f"  Max tokens: {MAX_NEW_TOKENS}")
-print("=" * 55)
-
-# ══════════════════════════════════════════
-# CHAT MEMORY
-# ══════════════════════════════════════════
-sessions = {}
-
-def get_memory(sid):
-    if sid not in sessions:
-        sessions[sid] = []
-    return sessions[sid]
-
-def add_to_memory(sid, role, content):
-    mem = get_memory(sid)
-    mem.append({
-        "role": role,
-        "content": content,
-        "ts": datetime.datetime.now().isoformat(),
-    })
-    if len(mem) > MAX_MEMORY * 2:
-        sessions[sid] = mem[-(MAX_MEMORY * 2):]
-
-# ══════════════════════════════════════════
-# GEMMA RESPONSE GENERATION
-# ══════════════════════════════════════════
-def generate_response(user_input, session_id):
-    memory = get_memory(session_id)
-
-    # Build chat messages: system instruction → memory → new message
-    messages = [
-        {"role": "user", "content": f"[System Instruction]\n{SYSTEM_PROMPT}"},
-        {"role": "assistant", "content": "I am waiting for you!"},
-    ]
-
-    # Add recent memory (last 6 turns = 12 messages)
-    recent = memory[-(6 * 2):]
-    for msg in recent:
-        role = "user" if msg["role"] == "user" else "assistant"
-        messages.append({"role": role, "content": msg["content"]})
-
-    # Current user message
-    messages.append({"role": "user", "content": user_input})
-
-    # Tokenize with Gemma chat template
-    input_ids = tokenizer.apply_chat_template(
-        messages,
-        return_tensors="pt",
-        add_generation_prompt=True,
-    )
-
-    # Generate
-    with torch.no_grad():
-        outputs = model.generate(
-            input_ids,
-            max_new_tokens=MAX_NEW_TOKENS,
-            do_sample=True,
-            temperature=0.9,
-            top_k=45,
-            top_p=0.97,
-        )
-
-    # Decode only new tokens
-    new_tokens = outputs[0][input_ids.shape[-1]:]
-    response = tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
-
-    # Clean artifacts
-    response = response.split("<end_of_turn>")[0].strip()
-    response = response.split("<start_of_turn>")[0].strip()
-
-    if not response or len(response) < 2:
-        response = "I appear to have momentarily lost my train of thought. Could you rephrase that?"
-
-    add_to_memory(session_id, "user", user_input)
-    add_to_memory(session_id, "assistant", response)
-    return response
-
-# ══════════════════════════════════════════
-# TTS SYNTHESIS
-# ══════════════════════════════════════════
-def synthesize_speech(text, voice=None):
-    if tts is None:
-        return None
-    try:
-        voice = voice or TTS_VOICE
-        clean = clean_text_for_tts(text)
-        if not clean or len(clean) < 2:
-            return None
-        if len(clean) > 400:
-            clean = clean[:400]
-        audio = tts.generate(clean, voice=voice, speed=TTS_SPEED)
-        if audio is None or len(audio) == 0:
-            return None
-        buf = io.BytesIO()
-        sf.write(buf, audio, 24000, format='WAV', subtype='PCM_16')
-        buf.seek(0)
-        return base64.b64encode(buf.read()).decode('utf-8')
-    except Exception as e:
-        print(f"TTS Error: {e}")
-        return None
+app = Flask(__name__)
 
-# ══════════════════════════════════════════
-# INLINE HTML
-# ══════════════════════════════════════════
-HTML_PAGE = """<!DOCTYPE html>
+HTML_PAGE = """
+<!DOCTYPE html>
 <html lang="en">
 <head>
 <meta charset="UTF-8">
 <meta name="viewport" content="width=device-width, initial-scale=1.0">
-<title>J.A.R.V.I.S. AI</title>
+<title>LFM2-2.6B Chat</title>
 <style>
-*{margin:0;padding:0;box-sizing:border-box}
-body{
-    font-family:'Segoe UI',Tahoma,Geneva,Verdana,sans-serif;
-    background:#0a0a1a;color:#e0e0e0;
-    height:100vh;display:flex;flex-direction:column;overflow:hidden;
-}
-.header{
-    background:linear-gradient(135deg,#0d1b2a,#1b2838);
-    border-bottom:1px solid #00d4ff33;
-    padding:12px 20px;display:flex;align-items:center;
-    justify-content:space-between;flex-shrink:0;
-}
-.header-left{display:flex;align-items:center;gap:12px}
-.arc-reactor{
-    width:38px;height:38px;border-radius:50%;
-    background:radial-gradient(circle,#00d4ff 0%,#0088aa 40%,#004466 70%,transparent 100%);
-    box-shadow:0 0 20px #00d4ff88,0 0 40px #00d4ff44,inset 0 0 10px #00d4ff66;
-    animation:pulse 2s ease-in-out infinite;position:relative;
-}
-.arc-reactor::after{
-    content:'';position:absolute;top:50%;left:50%;
-    transform:translate(-50%,-50%);width:12px;height:12px;
-    border-radius:50%;background:#00d4ff;box-shadow:0 0 8px #00d4ff;
-}
-@keyframes pulse{
-    0%,100%{box-shadow:0 0 20px #00d4ff88,0 0 40px #00d4ff44}
-    50%{box-shadow:0 0 30px #00d4ffaa,0 0 60px #00d4ff66}
-}
-.htitle h1{font-size:1.2rem;color:#00d4ff;letter-spacing:3px;text-transform:uppercase}
-.htitle p{font-size:.65rem;color:#5a8a9a;letter-spacing:1px}
-.hctrl{display:flex;gap:8px;align-items:center;flex-wrap:wrap}
-.cbtn{
-    background:#0d1b2a;border:1px solid #00d4ff44;color:#00d4ff;
-    padding:5px 12px;border-radius:6px;cursor:pointer;
-    font-size:.7rem;transition:all .3s;letter-spacing:.5px;
-}
-.cbtn:hover{background:#00d4ff22;border-color:#00d4ff88}
-.cbtn.active{background:#00d4ff22;border-color:#00d4ff;box-shadow:0 0 8px #00d4ff44}
-.sdot{width:8px;height:8px;border-radius:50%;background:#00ff88;box-shadow:0 0 6px #00ff88}
-.sdot.err{background:#ff4444;box-shadow:0 0 6px #ff4444}
-
-.cfgbar{
-    background:#0d1117;border-bottom:1px solid #00d4ff15;
-    padding:8px 20px;display:none;flex-wrap:wrap;gap:12px;
-    align-items:center;flex-shrink:0;
-}
-.cfgbar.open{display:flex}
-.cgrp{display:flex;align-items:center;gap:6px}
-.cgrp label{font-size:.65rem;color:#5a8a9a;text-transform:uppercase;letter-spacing:1px}
-.cgrp select{
-    background:#0f1923;border:1px solid #00d4ff33;color:#00d4ff;
-    padding:4px 8px;border-radius:4px;font-size:.7rem;cursor:pointer;outline:none;
-}
-.cgrp select:focus{border-color:#00d4ff}
-.ctag{
-    font-size:.6rem;padding:3px 8px;border-radius:10px;
-    background:#00d4ff15;border:1px solid #00d4ff33;color:#00d4ffaa;
-}
-
-.chat{
-    flex:1;overflow-y:auto;padding:16px 20px;
-    display:flex;flex-direction:column;gap:14px;scroll-behavior:smooth;
-}
-.chat::-webkit-scrollbar{width:3px}
-.chat::-webkit-scrollbar-thumb{background:#00d4ff33;border-radius:2px}
-
-.msg{
-    max-width:80%;padding:12px 16px;border-radius:14px;
-    font-size:.9rem;line-height:1.6;animation:fadeIn .3s ease-out;
-}
-@keyframes fadeIn{
-    from{opacity:0;transform:translateY(8px)}
-    to{opacity:1;transform:translateY(0)}
-}
-.msg.user{
-    align-self:flex-end;background:linear-gradient(135deg,#1a3a5c,#0d2847);
-    border:1px solid #00d4ff33;color:#c8e6ff;border-bottom-right-radius:4px;
-}
-.msg.bot{
-    align-self:flex-start;background:linear-gradient(135deg,#141e30,#0f1923);
-    border:1px solid #00d4ff22;color:#e0e0e0;border-bottom-left-radius:4px;
-}
-.msg .lbl{font-size:.58rem;color:#00d4ff88;letter-spacing:2px;margin-bottom:5px;text-transform:uppercase}
-.msg .txt{white-space:pre-wrap;word-wrap:break-word}
-.msg .actrl{margin-top:8px;display:flex;align-items:center;gap:8px}
-.abtn{
-    display:inline-flex;align-items:center;gap:4px;
-    background:#00d4ff15;border:1px solid #00d4ff33;color:#00d4ff;
-    padding:3px 10px;border-radius:10px;cursor:pointer;
-    font-size:.65rem;transition:all .2s;
-}
-.abtn:hover{background:#00d4ff25;border-color:#00d4ff66}
-.abtn:disabled{opacity:.3;cursor:wait}
-.astat{font-size:.58rem;color:#5a8a9a}
-
-.typi{align-self:flex-start;display:flex;gap:5px;padding:14px 18px}
-.typi span{width:7px;height:7px;border-radius:50%;background:#00d4ff;animation:typ 1.4s infinite}
-.typi span:nth-child(2){animation-delay:.2s}
-.typi span:nth-child(3){animation-delay:.4s}
-@keyframes typ{
-    0%,60%,100%{opacity:.2;transform:scale(.8)}
-    30%{opacity:1;transform:scale(1.1)}
-}
-
-.welcome{
-    display:flex;flex-direction:column;align-items:center;
-    justify-content:center;flex:1;gap:10px;opacity:.5;
-}
-.welcome .breact{
-    width:70px;height:70px;border-radius:50%;
-    background:radial-gradient(circle,#00d4ff 0%,#0088aa 35%,#004466 65%,transparent 100%);
-    box-shadow:0 0 40px #00d4ff66;animation:pulse 2s ease-in-out infinite;
-}
-.welcome h2{color:#00d4ff;font-size:1rem;letter-spacing:4px}
-.welcome p{color:#5a8a9a;font-size:.75rem}
-.welcome .minfo{font-size:.65rem;color:#3a5a6a;margin-top:4px}
-
-.inbar{
-    padding:14px 20px;background:linear-gradient(0deg,#0d1b2a,#0a0a1a);
-    border-top:1px solid #00d4ff22;flex-shrink:0;
-}
-.inwrap{display:flex;gap:8px;max-width:900px;margin:0 auto}
-#msgIn{
-    flex:1;background:#0f1923;border:1px solid #00d4ff33;border-radius:12px;
-    padding:11px 16px;color:#e0e0e0;font-size:.9rem;outline:none;
-    transition:border-color .3s;font-family:inherit;
-}
-#msgIn:focus{border-color:#00d4ff88;box-shadow:0 0 12px #00d4ff22}
-#msgIn::placeholder{color:#3a5a6a}
-#sendBtn{
-    background:linear-gradient(135deg,#00d4ff,#0088cc);border:none;border-radius:12px;
-    padding:11px 22px;color:#0a0a1a;font-weight:700;cursor:pointer;
-    font-size:.8rem;letter-spacing:1px;transition:all .3s;text-transform:uppercase;
-}
-#sendBtn:hover{box-shadow:0 0 18px #00d4ff66;transform:translateY(-1px)}
-#sendBtn:disabled{opacity:.4;cursor:not-allowed;transform:none}
-.infoot{
-    display:flex;justify-content:space-between;margin-top:5px;
-    max-width:900px;margin-left:auto;margin-right:auto;
-}
-.infoot span{font-size:.6rem;color:#3a5a6a}
-
-@media(max-width:640px){
-    .header{padding:10px 12px}
-    .htitle h1{font-size:1rem}
-    .msg{max-width:92%;font-size:.82rem}
-    .chat{padding:10px}
-    .inbar{padding:10px}
-    .cfgbar{padding:6px 12px}
-}
+  * { box-sizing: border-box; margin: 0; padding: 0; }
+  body { font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif; background: #1a1a2e; color: #eee; display: flex; justify-content: center; align-items: center; min-height: 100vh; }
+  .container { width: 100%; max-width: 800px; padding: 20px; }
+  h1 { text-align: center; margin-bottom: 6px; color: #e94560; font-size: 1.5rem; }
+  .subtitle { text-align: center; margin-bottom: 20px; color: #888; font-size: 0.85rem; }
+  .chatbox { background: #16213e; border-radius: 12px; padding: 20px; height: 55vh; overflow-y: auto; margin-bottom: 15px; border: 1px solid #0f3460; }
+  .msg { margin-bottom: 14px; line-height: 1.6; }
+  .msg.user { color: #e94560; }
+  .msg.user::before { content: "You: "; font-weight: bold; }
+  .msg.bot { color: #a8d8ea; }
+  .msg.bot::before { content: "AI: "; font-weight: bold; }
+  .stats { color: #666; font-size: 0.78rem; margin-top: 4px; }
+  .input-row { display: flex; gap: 10px; }
+  textarea { flex: 1; padding: 12px; border-radius: 8px; border: 1px solid #0f3460; background: #16213e; color: #eee; font-size: 1rem; resize: none; height: 60px; font-family: inherit; }
+  textarea:focus { outline: none; border-color: #e94560; }
+  button { padding: 12px 28px; border-radius: 8px; border: none; background: #e94560; color: #fff; font-size: 1rem; cursor: pointer; font-weight: bold; }
+  button:hover { background: #c73650; }
+  button:disabled { background: #555; cursor: not-allowed; }
+  .settings { display: flex; gap: 15px; margin-bottom: 15px; flex-wrap: wrap; align-items: center; }
+  .settings label { font-size: 0.85rem; color: #aaa; }
+  .settings input, .settings select { background: #16213e; border: 1px solid #0f3460; color: #eee; padding: 5px 8px; border-radius: 6px; width: 80px; }
 </style>
 </head>
 <body>
-
-<div class="header">
-    <div class="header-left">
-        <div class="arc-reactor"></div>
-        <div class="htitle">
-            <h1>J.A.R.V.I.S.</h1>
-            <p>Just A Rather Very Intelligent System</p>
-        </div>
-    </div>
-    <div class="hctrl">
-        <div class="sdot" id="sDot"></div>
-        <button class="cbtn" id="cfgBtn" onclick="toggleCfg()">⚙ CONFIG</button>
-        <button class="cbtn active" id="ttsBtn" onclick="toggleTts()">🔊 VOICE</button>
-        <button class="cbtn" onclick="clearChat()">🗑 CLEAR</button>
-    </div>
-</div>
-
-<div class="cfgbar" id="cfgPanel">
-    <div class="cgrp">
-        <label>LLM:</label>
-        <span class="ctag">Gemma 3 270M-IT</span>
-    </div>
-    <div class="cgrp">
-        <label>TTS:</label>
-        <span class="ctag" id="ttsTag">loading...</span>
-    </div>
-    <div class="cgrp">
-        <label>Voice:</label>
-        <select id="voiceSel">
-            <option value="Kiki">Kiki</option>
-            <option value="Bella">Bella</option>
-            <option value="Jasper">Jasper</option>
-            <option value="Luna">Luna</option>
-            <option value="Bruno">Bruno</option>
-            <option value="Rosie">Rosie</option>
-            <option value="Hugo">Hugo</option>
-            <option value="Leo">Leo</option>
-        </select>
-    </div>
-    <div class="cgrp">
-        <label>TTS env options:</label>
-        <span class="ctag">nano-fp32</span>
-        <span class="ctag">nano-int8</span>
-        <span class="ctag">micro</span>
-        <span class="ctag">mini</span>
-    </div>
-</div>
-
-<div class="chat" id="chatBox">
-    <div class="welcome" id="welc">
-        <div class="breact"></div>
-        <h2>SYSTEMS ONLINE</h2>
-        <p>Type a message below to begin interaction</p>
-        <div class="minfo" id="wInfo">Initializing...</div>
-    </div>
+<div class="container">
+  <h1>LFM2-2.6B Uncensored</h1>
+  <p class="subtitle">Running locally on CPU with llama.cpp</p>
+  <div class="settings">
+    <label>Max tokens: <input type="number" id="maxTokens" value="256" min="16" max="2048"></label>
+    <label>Temperature: <input type="number" id="temperature" value="0.7" min="0" max="2" step="0.1"></label>
+    <label>Top-P: <input type="number" id="topP" value="0.9" min="0" max="1" step="0.05"></label>
+    <button onclick="clearChat()" style="padding:5px 14px;font-size:0.85rem;">Clear</button>
+  </div>
+  <div class="chatbox" id="chatbox"></div>
+  <div class="input-row">
+    <textarea id="userInput" placeholder="Type your message..." onkeydown="if(event.key==='Enter'&&!event.shiftKey){event.preventDefault();sendMsg();}"></textarea>
+    <button id="sendBtn" onclick="sendMsg()">Send</button>
+  </div>
 </div>
-
-<div class="inbar">
-    <div class="inwrap">
-        <input type="text" id="msgIn" placeholder="Talk to J.A.R.V.I.S..." autocomplete="off"/>
-        <button id="sendBtn" onclick="send()">SEND</button>
-    </div>
-    <div class="infoot">
-        <span id="memCt">Memory: 0 turns</span>
-        <span id="modInfo">Loading...</span>
-    </div>
-</div>
-
 <script>
-let sid=crypto.randomUUID?crypto.randomUUID():Date.now().toString(36)+Math.random().toString(36).slice(2);
-let ttsOn=true,busy=false,mc=0,voice='Kiki';
-const C=document.getElementById('chatBox'),I=document.getElementById('msgIn'),B=document.getElementById('sendBtn');
-
-I.addEventListener('keydown',e=>{if(e.key==='Enter'&&!e.shiftKey){e.preventDefault();send()}});
-document.getElementById('voiceSel').addEventListener('change',function(){voice=this.value});
-
-function toggleTts(){
-    ttsOn=!ttsOn;
-    const b=document.getElementById('ttsBtn');
-    b.classList.toggle('active',ttsOn);
-    b.textContent=ttsOn?'🔊 VOICE':'🔇 MUTE';
-}
-function toggleCfg(){
-    document.getElementById('cfgPanel').classList.toggle('open');
-    document.getElementById('cfgBtn').classList.toggle('active');
-}
-
-async function send(){
-    const t=I.value.trim();
-    if(!t||busy)return;
-    const w=document.getElementById('welc');
-    if(w)w.style.display='none';
-    addMsg(t,'user');
-    I.value='';busy=true;B.disabled=true;
-    const ty=showTyp();
-    const id=++mc;
-    try{
-        const r=await fetch('/chat',{
-            method:'POST',headers:{'Content-Type':'application/json'},
-            body:JSON.stringify({message:t,session_id:sid})
-        });
-        if(!r.ok)throw new Error('HTTP '+r.status);
-        const d=await r.json();
-        ty.remove();
-        const el=addBot(d.response,id);
-        document.getElementById('memCt').textContent='Memory: '+d.memory_length+' turns';
-        if(ttsOn&&d.tts_available)fetchAudio(d.response,el);
-    }catch(e){
-        ty.remove();
-        addBot('System malfunction. Please try again.',id);
-        console.error(e);
+  const chatbox = document.getElementById('chatbox');
+  const userInput = document.getElementById('userInput');
+  const sendBtn = document.getElementById('sendBtn');
+  let history = [];
+
+  function addMsg(role, text, stats) {
+    const div = document.createElement('div');
+    div.className = 'msg ' + role;
+    div.textContent = text;
+    if (stats) {
+      const s = document.createElement('div');
+      s.className = 'stats';
+      s.textContent = stats;
+      div.appendChild(s);
     }
-    busy=false;B.disabled=false;I.focus();
-}
-
-async function fetchAudio(text,el){
-    const st=el.querySelector('.astat'),pb=el.querySelector('.abtn');
-    if(st)st.textContent='⏳ Generating voice...';
-    if(pb)pb.disabled=true;
-    try{
-        const r=await fetch('/tts',{
-            method:'POST',headers:{'Content-Type':'application/json'},
-            body:JSON.stringify({text:text,voice:voice})
-        });
-        const d=await r.json();
-        if(d.audio){
-            if(pb){pb.dataset.audio=d.audio;pb.disabled=false;pb.textContent='▶ Play'}
-            if(st)st.textContent='✅ Ready';
-            playB64(d.audio);
-        }else{
-            if(st)st.textContent='⚠️ Voice unavailable';
-            if(pb)pb.style.display='none';
+    chatbox.appendChild(div);
+    chatbox.scrollTop = chatbox.scrollHeight;
+    return div;
+  }
+
+  function clearChat() {
+    history = [];
+    chatbox.innerHTML = '';
+  }
+
+  async function sendMsg() {
+    const text = userInput.value.trim();
+    if (!text) return;
+    userInput.value = '';
+    addMsg('user', text);
+    history.push({role: 'user', content: text});
+    sendBtn.disabled = true;
+
+    const botDiv = addMsg('bot', '');
+    botDiv.textContent = '';
+
+    try {
+      const resp = await fetch('/chat', {
+        method: 'POST',
+        headers: {'Content-Type': 'application/json'},
+        body: JSON.stringify({
+          messages: history,
+          max_tokens: parseInt(document.getElementById('maxTokens').value) || 256,
+          temperature: parseFloat(document.getElementById('temperature').value) || 0.7,
+          top_p: parseFloat(document.getElementById('topP').value) || 0.9
+        })
+      });
+      const reader = resp.body.getReader();
+      const decoder = new TextDecoder();
+      let full = '';
+      let statsText = '';
+
+      while (true) {
+        const {done, value} = await reader.read();
+        if (done) break;
+        const chunk = decoder.decode(value, {stream: true});
+        const lines = chunk.split('\\n');
+        for (const line of lines) {
+          if (!line.startsWith('data: ')) continue;
+          const data = line.slice(6);
+          if (data === '[DONE]') continue;
+          try {
+            const j = JSON.parse(data);
+            if (j.token) { full += j.token; botDiv.childNodes[0].textContent = full; }
+            if (j.stats) { statsText = j.stats; }
+          } catch(e) {}
         }
-    }catch(e){
-        if(st)st.textContent='⚠️ Voice error';
-        if(pb)pb.style.display='none';
-    }
-}
-
-function addMsg(t,role){
-    const d=document.createElement('div');
-    d.className='msg '+role;
-    d.innerHTML='<div class="txt">'+esc(t)+'</div>';
-    C.appendChild(d);sc();
-}
-function addBot(t,id){
-    const d=document.createElement('div');
-    d.className='msg bot';d.id='m'+id;
-    d.innerHTML='<div class="lbl">⟐ JARVIS</div><div class="txt">'+esc(t)+'</div>'+
-        (ttsOn?'<div class="actrl"><button class="abtn" disabled onclick="replay(this)">⏳</button><span class="astat">Requesting voice...</span></div>':'');
-    C.appendChild(d);sc();return d;
-}
-function showTyp(){
-    const d=document.createElement('div');
-    d.className='typi';
-    d.innerHTML='<span></span><span></span><span></span>';
-    C.appendChild(d);sc();return d;
-}
-
-function playB64(b){
-    try{
-        const bin=atob(b),u8=new Uint8Array(bin.length);
-        for(let i=0;i<bin.length;i++)u8[i]=bin.charCodeAt(i);
-        const url=URL.createObjectURL(new Blob([u8],{type:'audio/wav'}));
-        const a=new Audio(url);
-        a.play().catch(e=>console.log('Autoplay blocked:',e));
-        a.onended=()=>URL.revokeObjectURL(url);
-    }catch(e){console.error(e)}
-}
-function replay(b){if(b.dataset.audio)playB64(b.dataset.audio)}
-
-async function clearChat(){
-    await fetch('/clear',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({session_id:sid})});
-    C.innerHTML='<div class="welcome" id="welc"><div class="breact"></div><h2>SYSTEMS ONLINE</h2><p>Type a message below to begin</p></div>';
-    document.getElementById('memCt').textContent='Memory: 0 turns';
-    sid=crypto.randomUUID?crypto.randomUUID():Date.now().toString(36)+Math.random().toString(36).slice(2);
-}
+      }
 
-function esc(t){const d=document.createElement('div');d.textContent=t;return d.innerHTML}
-function sc(){C.scrollTop=C.scrollHeight}
+      if (statsText) {
+        const s = document.createElement('div');
+        s.className = 'stats';
+        s.textContent = statsText;
+        botDiv.appendChild(s);
+      }
 
-fetch('/health').then(r=>r.json()).then(d=>{
-    document.getElementById('ttsTag').textContent=d.tts_mode+(d.tts_model==='DISABLED'?' (OFF)':'');
-    document.getElementById('modInfo').textContent='Gemma 3 · '+d.tts_mode+' · '+d.tts_voice+' · CPU';
-    const wi=document.getElementById('wInfo');
-    if(wi)wi.textContent='LLM: Gemma 3 270M-IT | TTS: '+d.tts_mode+' | Voice: '+d.tts_voice;
-    if(d.tts_model==='DISABLED')document.getElementById('sDot').classList.add('err');
-    if(d.tts_voice){document.getElementById('voiceSel').value=d.tts_voice;voice=d.tts_voice}
-}).catch(()=>{});
+      // Ensure there's a text node
+      if (!botDiv.childNodes.length) botDiv.textContent = full;
 
-I.focus();
+      history.push({role: 'assistant', content: full});
+    } catch(e) {
+      botDiv.textContent = 'Error: ' + e.message;
+    }
+    sendBtn.disabled = false;
+    chatbox.scrollTop = chatbox.scrollHeight;
+  }
 </script>
 </body>
-</html>"""
+</html>
+"""
 
-# ══════════════════════════════════════════
-# FLASK APP
-# ══════════════════════════════════════════
-app = Flask(__name__)
 
 @app.route("/")
 def index():
     return HTML_PAGE
 
+
 @app.route("/chat", methods=["POST"])
 def chat():
-    data = request.json or {}
-    user_input = data.get("message", "").strip()
-    session_id = data.get("session_id", str(uuid.uuid4()))
-
-    if not user_input:
-        return jsonify({"error": "Empty message"}), 400
-
-    try:
-        response = generate_response(user_input, session_id)
-    except Exception as e:
-        print(f"Generation error: {e}")
-        traceback.print_exc()
-        response = "I encountered a temporary system malfunction. Please try again."
+    data = request.json
+    messages = data.get("messages", [])
+    max_tokens = min(data.get("max_tokens", 256), 2048)
+    temperature = data.get("temperature", 0.7)
+    top_p = data.get("top_p", 0.9)
+
+    # Build prompt — simple ChatML-ish format
+    prompt = ""
+    for msg in messages:
+        role = msg["role"]
+        content = msg["content"]
+        if role == "user":
+            prompt += f"<|user|>\n{content}\n"
+        elif role == "assistant":
+            prompt += f"<|assistant|>\n{content}\n"
+        elif role == "system":
+            prompt += f"<|system|>\n{content}\n"
+    prompt += "<|assistant|>\n"
+
+    def generate():
+        token_count = 0
+        start = time.perf_counter()
+
+        stream = llm(
+            prompt,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            stop=["<|user|>", "<|assistant|>", "<|end|>", "<|endoftext|>"],
+            stream=True,
+        )
 
-    return jsonify({
-        "response": response,
-        "session_id": session_id,
-        "tts_available": tts is not None,
-        "memory_length": len(get_memory(session_id)),
-    })
+        for output in stream:
+            token_text = output["choices"][0]["text"]
+            token_count += 1
+            yield f"data: {jsonify_compact({'token': token_text})}\n\n"
 
-@app.route("/tts", methods=["POST"])
-def tts_endpoint():
-    data = request.json or {}
-    text = data.get("text", "").strip()
-    voice = data.get("voice", TTS_VOICE)
+        elapsed = time.perf_counter() - start
+        tps = token_count / elapsed if elapsed > 0 else 0
+        stats = f"{token_count} tokens in {elapsed:.1f}s — {tps:.2f} tokens/s"
+        yield f"data: {jsonify_compact({'stats': stats})}\n\n"
+        yield "data: [DONE]\n\n"
 
-    if not text:
-        return jsonify({"error": "Empty text"}), 400
-    if tts is None:
-        return jsonify({"error": "TTS not available", "audio": None}), 200
+    return Response(stream_with_context(generate()), mimetype="text/event-stream")
 
-    audio_b64 = synthesize_speech(text, voice=voice)
-    return jsonify({"audio": audio_b64})
 
-@app.route("/clear", methods=["POST"])
-def clear():
-    data = request.json or {}
-    sid = data.get("session_id", "")
-    if sid in sessions:
-        del sessions[sid]
-    return jsonify({"status": "cleared"})
+def jsonify_compact(obj):
+    import json
+    return json.dumps(obj, ensure_ascii=False, separators=(",", ":"))
 
-@app.route("/health")
-def health():
-    return jsonify({
-        "status": "online",
-        "llm": "Gemma 3 270M-IT",
-        "tts_mode": TTS_MODE,
-        "tts_model": tts_model_name if tts else "DISABLED",
-        "tts_voice": TTS_VOICE,
-        "tts_voices": ["Bella","Jasper","Luna","Bruno","Rosie","Hugo","Kiki","Leo"],
-        "max_new_tokens": MAX_NEW_TOKENS,
-    })
 
 if __name__ == "__main__":
-    print("🚀 Ana is online!")
-    app.run(host="0.0.0.0", port=7860, threaded=True)
+    port = int(os.environ.get("PORT", 7860))
+    app.run(host="0.0.0.0", port=port, debug=False)