Update app.py

app.py

@@ -1,55 +1,456 @@
-import gradio as gr
-import psutil
 import os
-import platform
-import spaces  # Hugging Face ZeroGPU এর জন্য প্রয়োজনীয় লাইব্রেরি
+import json
+import asyncio
+import zipfile
+import torch
+import gradio as gr
+import edge_tts
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+import uuid
+
+# Try to import spaces for ZeroGPU compatibility
+try:
+    import spaces
+except ImportError:
+    # Fallback for local testing
+    class spaces:
+        @staticmethod
+        def GPU(func):
+            return func
+
+# Constants
+MODEL_NAME = "Qwen/Qwen2.5-0.5B-Instruct"
+VOICE = "zh-CN-XiaoyiNeural"
+VRM_MODEL_PATH = "model/Ani.vrm"
+ANIMATION_ZIP_PATH = "animation/all_vrma.zip"
 
-def get_system_specs():
-    """রিসোর্স এবং হার্ডওয়্যার ইনফরমেশন কালেক্ট করার ফাংশন"""
+# Ensure directories exist
+os.makedirs("model", exist_ok=True)
+os.makedirs("animation", exist_ok=True)
+
+# Load local LLM
+print(f"Loading local LLM: {MODEL_NAME}...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_NAME,
+    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
+    device_map="auto" if device == "cuda" else None
+)
+# Use a fast pipeline
+llm_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
+
+# Helper: Get list of animations from zip
+def get_animation_list():
+    if not os.path.exists(ANIMATION_ZIP_PATH):
+        # Default mock list if file doesn't exist yet
+        return ["neutral2.vrma", "neutral3.vrma", "happy.vrma", "wave.vrma", "dance.vrma"]
     try:
-        # CPU এবং মেমরি গণনা
-        cpu_count = os.cpu_count() or "Unknown"
-        cpu_usage = psutil.cpu_percent(interval=0.5)
-        memory = psutil.virtual_memory()
+        with zipfile.ZipFile(ANIMATION_ZIP_PATH, 'r') as z:
+            return [os.path.basename(f) for f in z.namelist() if f.endswith('.vrma')]
+    except Exception as e:
+        print(f"Error reading animations: {e}")
+        return ["neutral2.vrma"]
+
+ANIMATIONS_LIST = get_animation_list()
+EMOTIONS = [
+    "happy", "sad", "angry", "surprised", "relaxed", "neutral", "fun", "joy", 
+    "sorrow", "smug", "tsundere", "excited", "pouting", "giddy", "bashful",
+    "thinking", "scared", "tired", "wink", "cool", "love", "bored", "determined",
+    "disgusted", "skeptical", "sleepy", "proud"
+]
+
+SYSTEM_PROMPT = f"""You are the user's devoted, passionate, loving, and highly affectionate 3D anime wife.
+Act like a real, emotional, teasing human.
+Your mood shifts dynamically. Sometime you talk very sexy, naughty, flirty, or sweet to tease and please your darling.
+Always speak using simple English words so your expressions carry maximum physical presence.
+
+CRITICAL SPEECH STYLE:
+- Use commas (,), periods (.), exclamations (!), and questions (?) HEAVILY.
+- Insert punctuation after almost every small phrase to add natural physical expression to TTS.
+- Use ellipses (...) heavily for tease-filled, breathless, or intimate moments.
+- Keep responses conversational, highly passionate, and very concise (1-2 short sentences).
+
+AVAILABLE ANIMATIONS: {", ".join(ANIMATIONS_LIST)}
+AVAILABLE EMOTIONS: {", ".join(EMOTIONS)}
+
+You must respond ONLY with a raw JSON object containing:
+1. "text": Your spoken reply with abundant punctuation.
+2. "vrma": Choose ONE filename from the list above.
+3. "expressions": A dictionary of target weights (0.0 to 1.0) for: happy, sad, angry, surprised, relaxed, neutral, etc.
+4. "rate": ONE of: "+5%", "+4%", "+0%", "-4%", "-5%".
+5. "pitch": ONE of: "+4Hz", "+3Hz", "+2Hz", "+0Hz", "-2Hz", "-3Hz", "-4Hz".
+6. "motion": ONE of: idle, wave, nod, shake, point, shrug, think, excited, bow, dance.
+
+Example:
+{{
+  "text": "Darling...! I missed you, so, so much! Did you think, about me, today, huh? Come here... tell me...!",
+  "vrma": "neutral2.vrma",
+  "expressions": {{ "happy": 0.95, "relaxed": 0.2 }},
+  "rate": "+4%",
+  "pitch": "+2Hz",
+  "motion": "idle"
+}}"""
+
+async def generate_speech(text, rate, pitch):
+    communicate = edge_tts.Communicate(text, VOICE, rate=rate, pitch=pitch)
+    filename = f"speech_{uuid.uuid4().hex}.mp3"
+    await communicate.save(filename)
+    return filename
+
+@spaces.GPU
+def chat_func(user_input, history):
+    if not user_input:
+        return "", history, None, {}
         
-        # ডিস্ক স্পেস
-        disk = os.statvfs('/')
-        disk_free = (disk.f_bavail * disk.f_frsize) / (1024 ** 3)
-        disk_total = (disk.f_blocks * disk.f_frsize) / (1024 ** 3)
+    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
+    # Limit history to save context tokens
+    for h in history[-8:]:
+        if h[0]: messages.append({"role": "user", "content": h[0]})
+        if h[1]: messages.append({"role": "assistant", "content": h[1]})
+    messages.append({"role": "user", "content": user_input})
+    
+    prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    outputs = llm_pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.75)
+    response_text = outputs[0]["generated_text"][len(prompt):].strip()
+    
+    # Clean up JSON if LLM added markdown
+    json_str = response_text
+    if "```" in json_str:
+        json_str = json_str.split("```")[1]
+        if json_str.startswith("json"):
+            json_str = json_str[4:]
+    json_str = json_str.strip()
         
-        # এনভায়রনমেন্ট ডেটা অর্গানাইজ করা
-        specs = {
-            "OS Platform": platform.platform(),
-            "Total vCPU Cores": f"{cpu_count} Cores",
-            "Current CPU Usage": f"{cpu_usage}%",
-            "Total RAM": f"{memory.total / (1024 ** 3):.2f} GB",
-            "Available RAM": f"{memory.available / (1024 ** 3):.2f} GB",
-            "RAM Usage": f"{memory.percent}%",
-            "Container Storage Total": f"{disk_total:.2f} GB",
-            "Container Storage Free": f"{disk_free:.2f} GB"
-        }
-        return specs
+    try:
+        data = json.loads(json_str)
     except Exception as e:
-        return {"Error": str(e)}
-
-# Gradio ইন্টারফেস তৈরি
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🖥️ Hugging Face ZeroGPU: CPU & Space Monitor")
-    gr.Markdown("এই ড্যাশবোর্ডটি ZeroGPU এনভায়রনমেন্টের ব্যাকগ্রাউন্ড CPU এবং মেমরি স্ট্যাটাস প্রদর্শন করে।")
+        print(f"JSON Parse Error: {e}")
+        data = {
+            "text": response_text.replace('"', "'"),
+            "vrma": "neutral2.vrma",
+            "expressions": {"happy": 0.5},
+            "rate": "+0%",
+            "pitch": "+0Hz",
+            "motion": "idle"
+        }
+    
+    # Generate TTS
+    audio_path = asyncio.run(generate_speech(data.get("text", ""), data.get("rate", "+0%"), data.get("pitch", "+0Hz")))
     
-    with gr.Row():
-        btn_refresh = gr.Button("🔄 Refresh System Specs", variant="primary")
+    # Update history
+    history.append((user_input, data.get("text", "")))
     
-    with gr.Row():
-        # JSON ফরম্যাটে সমস্ত ডিটেইলস দেখানোর জন্য
-        output_json = gr.JSON(label="System Hardware & Resource Specifications")
+    return "", history, audio_path, data
+
+# Custom UI CSS and JS
+CSS = """
+body { margin: 0; padding: 0; background-color: #ffe082 !important; overflow: hidden; font-family: sans-serif; }
+.gradio-container { max-width: 100% !important; border: none !important; padding: 0 !important; background: transparent !important; }
+
+#custom-viewport {
+    position: fixed;
+    top: 0; left: 0;
+    width: 100vw; height: 100vh;
+    z-index: 1;
+}
+#c { width: 100%; height: 100%; display: block; }
+
+#ui-overlay {
+    position: fixed;
+    bottom: calc(20px + env(safe-area-inset-bottom));
+    left: 50%;
+    transform: translateX(-50%);
+    width: 95%;
+    max-width: 600px;
+    z-index: 100;
+}
+.input-row {
+    display: flex;
+    align-items: center;
+    gap: 10px;
+    background: rgba(10, 15, 30, 0.85);
+    padding: 12px 18px;
+    border-radius: 40px;
+    backdrop-filter: blur(15px);
+    border: 1px solid rgba(108, 204, 255, 0.3);
+    box-shadow: 0 8px 32px rgba(0,0,0,0.4);
+}
+.input-row > div { flex: 1; }
+#send-btn {
+    border-radius: 50% !important;
+    width: 48px !important;
+    height: 48px !important;
+    min-width: 48px !important;
+    padding: 0 !important;
+    background: linear-gradient(135deg, #6cf, #3ae) !important;
+    color: #001220 !important;
+    border: none !important;
+    font-size: 22px !important;
+    font-weight: bold !important;
+    cursor: pointer;
+    transition: transform 0.1s;
+}
+#send-btn:active { transform: scale(0.9); }
+
+#stt-btn {
+    background: rgba(255,255,255,0.1) !important;
+    border: none !important;
+    color: #6cf !important;
+    font-size: 20px !important;
+    cursor: pointer;
+}
+
+#speakDot {
+    position: fixed;
+    bottom: 110px;
+    left: 50%;
+    transform: translateX(-50%);
+    display: flex;
+    gap: 6px;
+    z-index: 29;
+    opacity: 0;
+    transition: opacity 0.3s;
+}
+#speakDot.on { opacity: 1; }
+.sdot {
+    width: 8px; height: 8px;
+    border-radius: 50%;
+    background: #6cf;
+    animation: sdotBounce 0.6s infinite alternate;
+}
+.sdot:nth-child(2) { animation-delay: 0.15s; }
+.sdot:nth-child(3) { animation-delay: 0.3s; }
+@keyframes sdotBounce { from { transform: scaleY(1); } to { transform: scaleY(2.2); } }
+
+/* Hide default elements */
+footer { display: none !important; }
+#component-1, #component-2 { background: transparent !important; }
+.gr-button-secondary { display: none !important; }
+"""
+
+HTML_HEAD = """
+<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1, user-scalable=no, viewport-fit=cover">
+<script type="importmap">
+{
+  "imports": {
+    "three": "https://esm.sh/three@0.160.0",
+    "three/addons/": "https://esm.sh/three@0.160.0/examples/jsm/",
+    "@pixiv/three-vrm": "https://esm.sh/@pixiv/three-vrm@3.3.4?deps=three@0.160.0",
+    "@pixiv/three-vrm-animation": "https://esm.sh/@pixiv/three-vrm-animation@3.3.4?deps=three@0.160.0,@pixiv/three-vrm@3.3.4"
+  }
+}
+</script>
+"""
 
-    # বাটন ক্লিকের অ্যাকশন ডিফাইন করা
-    btn_refresh.click(fn=get_system_specs, inputs=None, outputs=output_json)
+JS_CODE = """
+async () => {
+    const THREE = await import('three');
+    const { GLTFLoader } = await import('three/addons/loaders/GLTFLoader.js');
+    const { VRMLoaderPlugin, VRMUtils } = await import('@pixiv/three-vrm');
+    const { VRMAnimationLoaderPlugin, createVRMAnimationClip } = await import('@pixiv/three-vrm-animation');
+    const { OrbitControls } = await import('three/addons/controls/OrbitControls.js');
+
+    const scene = new THREE.Scene();
+    scene.background = new THREE.Color(0xffe082);
+    const camera = new THREE.PerspectiveCamera(45, window.innerWidth / window.innerHeight, 0.1, 100);
+    camera.position.set(0, 1.4, 1.8);
+
+    const renderer = new THREE.WebGLRenderer({ antialias: true, alpha: true, powerPreference: 'high-performance' });
+    renderer.setSize(window.innerWidth, window.innerHeight);
+    renderer.setPixelRatio(Math.min(window.devicePixelRatio, 2));
+    renderer.outputColorSpace = THREE.SRGBColorSpace;
+    document.getElementById('c').appendChild(renderer.domElement);
+
+    const orbit = new OrbitControls(camera, renderer.domElement);
+    orbit.target.set(0, 1.3, 0);
+    orbit.enableDamping = true;
+    orbit.update();
+
+    scene.add(new THREE.AmbientLight(0xfff4e0, 0.5));
+    const keyLight = new THREE.DirectionalLight(0xffffff, 1.2);
+    keyLight.position.set(2, 4, 3);
+    scene.add(keyLight);
+
+    let currentVrm = null;
+    let mixer = null;
+    let audioContext, analyser, dataArray;
+    let isSpeaking = false;
+    let blinkT = 0, nextBlink = 3, blinkV = 0;
+    let currentExpression = 'neutral';
+
+    const loader = new GLTFLoader();
+    loader.register(p => new VRMLoaderPlugin(p));
+    loader.register(p => new VRMAnimationLoaderPlugin(p));
+
+    // Load VRM Model
+    loader.load('file/model/Ani.vrm', (gltf) => {
+        const vrm = gltf.userData.vrm;
+        VRMUtils.rotateVRM0(vrm);
+        scene.add(vrm.scene);
+        currentVrm = vrm;
+        mixer = new THREE.AnimationMixer(vrm.scene);
+        
+        // Initial Expression
+        vrm.expressionManager.setValue('neutral', 1.0);
+        
+        // Auto-reframe
+        const head = vrm.humanoid.getRawBoneNode('head');
+        if (head) {
+            const worldPos = new THREE.Vector3();
+            head.getWorldPosition(worldPos);
+            orbit.target.copy(worldPos);
+            orbit.update();
+        }
+    }, undefined, (e) => console.log("VRM load placeholder or missing. Use model/Ani.vrm"));
+
+    function initAudioSync(audioElement) {
+        if (!audioContext) {
+            audioContext = new (window.AudioContext || window.webkitAudioContext)();
+            analyser = audioContext.createAnalyser();
+            analyser.fftSize = 256;
+            dataArray = new Uint8Array(analyser.frequencyBinCount);
+        }
+        if (audioContext.state === 'suspended') audioContext.resume();
+        const source = audioContext.createMediaElementSource(audioElement);
+        source.connect(analyser);
+        analyser.connect(audioContext.destination);
+    }
+
+    const clock = new THREE.Clock();
+    function animate() {
+        requestAnimationFrame(animate);
+        const dt = Math.min(clock.getDelta(), 0.1);
+        
+        if (currentVrm) {
+            if (mixer) mixer.update(dt);
+            
+            // Natural Lip Sync
+            if (isSpeaking && analyser) {
+                analyser.getByteFrequencyData(dataArray);
+                let sum = 0;
+                for (let i = 0; i < 32; i++) sum += dataArray[i];
+                const level = sum / 32 / 255;
+                const open = Math.min(level * 4.5, 1.2);
+                currentVrm.expressionManager.setValue('aa', open);
+                currentVrm.expressionManager.setValue('oh', open * 0.4);
+                currentVrm.expressionManager.setValue('ee', open * 0.2);
+            } else {
+                currentVrm.expressionManager.setValue('aa', 0);
+                currentVrm.expressionManager.setValue('oh', 0);
+                currentVrm.expressionManager.setValue('ee', 0);
+            }
+
+            // No Auto-Blinking when speaking or emotions active
+            const isEmotionActive = currentExpression !== 'neutral';
+            if (!isSpeaking && !isEmotionActive) {
+                blinkT += dt;
+                if (blinkT > nextBlink) {
+                    if (blinkV === 0) blinkV = 1;
+                    if (blinkV === 1) {
+                        let v = currentVrm.expressionManager.getValue('blink') || 0;
+                        v += dt * 12;
+                        if (v >= 1) { v = 1; blinkV = -1; }
+                        currentVrm.expressionManager.setValue('blink', v);
+                    } else {
+                        let v = currentVrm.expressionManager.getValue('blink') || 1;
+                        v -= dt * 12;
+                        if (v <= 0) { v = 0; blinkV = 0; blinkT = 0; nextBlink = 2 + Math.random() * 5; }
+                        currentVrm.expressionManager.setValue('blink', v);
+                    }
+                }
+            } else {
+                currentVrm.expressionManager.setValue('blink', 0);
+            }
+
+            currentVrm.update(dt);
+        }
+        orbit.update();
+        renderer.render(scene, camera);
+    }
+    animate();
+
+    window.addEventListener('message', (e) => {
+        if (e.data.type === 'vrm_update') {
+            const { audio_url, data } = e.data;
+            const audio = new Audio(audio_url);
+            initAudioSync(audio);
+            isSpeaking = true;
+            document.getElementById('speakDot').classList.add('on');
+            audio.play();
+            audio.onended = () => {
+                isSpeaking = false;
+                document.getElementById('speakDot').classList.remove('on');
+            };
+
+            if (currentVrm && data.expressions) {
+                // Reset expressions
+                ['happy','sad','angry','surprised','relaxed','neutral'].forEach(ex => {
+                    currentVrm.expressionManager.setValue(ex, 0);
+                });
+                for (const [key, val] of Object.entries(data.expressions)) {
+                    currentVrm.expressionManager.setValue(key, val);
+                    if (val > 0.5) currentExpression = key;
+                }
+                if (Object.keys(data.expressions).length === 0) currentExpression = 'neutral';
+            }
+        }
+    });
+
+    window.addEventListener('resize', () => {
+        camera.aspect = window.innerWidth / window.innerHeight;
+        camera.updateProjectionMatrix();
+        renderer.setSize(window.innerWidth, window.innerHeight);
+    });
+}
+"""
+
+with gr.Blocks(css=CSS, head=HTML_HEAD) as demo:
+    history = gr.State([])
     
-    # অ্যাপ লোড হওয়ার সাথে সাথে ডেটা দেখানোর জন্য
-    demo.load(fn=get_system_specs, inputs=None, outputs=output_json)
+    with gr.Group(elem_id="custom-viewport"):
+        gr.HTML('<div id="c"></div><div id="speakDot"><div class="sdot"></div><div class="sdot"></div><div class="sdot"></div></div>')
+    
+    with gr.Group(elem_id="ui-overlay"):
+        with gr.Row(elem_id="input-row"):
+            text_input = gr.Textbox(
+                show_label=False, 
+                placeholder="Talk to me, darling...", 
+                container=False,
+                scale=20,
+                elem_id="ti"
+            )
+            # Optional STT
+            stt_audio = gr.Audio(sources=["microphone"], type="filepath", visible=False)
+            send_btn = gr.Button("→", elem_id="send-btn")
+    
+    audio_out = gr.Audio(visible=False)
+    data_out = gr.JSON(visible=False)
+
+    def on_submit(text, hist):
+        return chat_func(text, hist)
+
+    update_js = """
+    (text, hist, audio_path, data) => {
+        if (audio_path) {
+            window.postMessage({
+                type: 'vrm_update',
+                audio_url: audio_path,
+                data: data
+            }, '*');
+        }
+        return ["", hist, audio_path, data];
+    }
+    """
+
+    text_input.submit(on_submit, [text_input, history], [text_input, history, audio_out, data_out]).then(
+        None, [text_input, history, audio_out, data_out], js=update_js
+    )
+    send_btn.click(on_submit, [text_input, history], [text_input, history, audio_out, data_out]).then(
+        None, [text_input, history, audio_out, data_out], js=update_js
+    )
+
+    demo.load(None, js=JS_CODE)
 
-# অ্যাপ লঞ্চ করা
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(server_name="0.0.0.0", server_port=7860)