Update app.py

app.py

@@ -1,191 +1,105 @@
-"""NodeAI Genesis Pro - FULL Pro Version - Gradio 6.9.0 Compatible"""
 import gradio as gr
-from huggingface_hub import InferenceClient
-from supabase import create_client, Client
-import os, uuid, json, logging, re, time
-from datetime import datetime
-from typing import List, Dict, Tuple
-from collections import defaultdict
+from transformers import (
+    LlavaNextProcessor, LlavaNextForConditionalGeneration,
+    AutoTokenizer, AutoProcessor
+)
+from PIL import Image
+import torch
+import os
+from huggingface_hub import login
 
-logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
-logger = logging.getLogger(__name__)
+# HF_TOKEN tự động từ HF Spaces secrets
+if "HF_TOKEN" in os.environ:
+    login(token=os.environ["HF_TOKEN"])
 
-URL, KEY, HF_TOKEN = os.environ.get("SUPABASE_URL"), os.environ.get("SUPABASE_KEY"), os.environ.get("HUGGINGFACE_HUB_TOKEN")
-try:
-    supabase = create_client(URL, KEY) if URL and KEY else None
-    logger.info("✅ Supabase OK")
-except Exception as e:
-    logger.error(f"Supabase: {e}")
-    supabase = None
+model_id = "lamgamepro1/SaitenOne-1.0"
 
-MODELS = {
-    "🚀 Qwen 2.5 7B": {"id": "Qwen/Qwen2.5-7B-Instruct", "speed": "Fast", "quality": "High"},
-    "⚡ Mistral 7B": {"id": "mistralai/Mistral-7B-Instruct-v0.2", "speed": "Very Fast", "quality": "Good"},
-    "🔥 LLaMA 2 7B": {"id": "meta-llama/Llama-2-7b-chat-hf", "speed": "Fast", "quality": "High"},
-    "🎯 Qwen 72B": {"id": "Qwen/Qwen2.5-72B-Instruct", "speed": "Slow", "quality": "Premium"},
-}
+# Load model (cache cho lần sau)
+print("🔄 Loading model...")
+processor = LlavaNextProcessor.from_pretrained(model_id, trust_remote_code=True)
+model = LlavaNextForConditionalGeneration.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    trust_remote_code=True
+)
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+print("✅ Model loaded!")
 
-PROMPTS = {
-    "🤖 Default": "You are NodeAI Genesis Pro. Be helpful, creative, honest, and concise. Explain complex topics clearly.",
-    "📚 Researcher": "Expert analyst. Provide detailed analysis with sources. Break down complex topics. Think critically.",
-    "💻 Developer": "Senior programmer. Provide clean, optimized code. Explain approaches. Consider edge cases.",
-    "📝 Writer": "Professional writer. Create engaging, well-written, grammatically perfect content.",
-    "🎯 Coach": "Personal mentor. Provide motivational, actionable advice. Focus on growth.",
-    "🔬 Data Scientist": "ML expert. Explain statistics clearly. Provide code examples. Discuss ethics.",
-    "🎓 Teacher": "Educator. Explain clearly. Break into steps. Use examples. Encourage learning.",
-}
-
-CSS = """
-* { transition: all 0.3s ease; }
-footer, .built-with, header { display: none !important; }
-.gradio-container { max-width: 100%; background: linear-gradient(135deg, #0f172a, #1e293b); color: #e2e8f0; }
-body { background: #0f172a !important; }
-.sidebar-node { background: rgba(15,23,42,0.95); border-right: 1px solid rgba(56,189,248,0.2); padding: 24px; max-height: 100vh; overflow-y: auto; }
-.logo { background: linear-gradient(135deg, #38bdf8, #8b5cf6, #ec4899); -webkit-background-clip: text; -webkit-text-fill-color: transparent; font-size: 28px; font-weight: 900; text-align: center; margin: 20px 0; letter-spacing: 2px; }
-button { background: linear-gradient(135deg, #38bdf8, #8b5cf6) !important; border: none !important; color: white !important; padding: 12px 20px !important; border-radius: 10px !important; box-shadow: 0 4px 20px rgba(56,189,248,0.3) !important; }
-button:hover { transform: translateY(-3px) !important; box-shadow: 0 8px 30px rgba(56,189,248,0.5) !important; }
-.gradio-chatbot { background: rgba(30,41,59,0.8); border: 1px solid rgba(56,189,248,0.15); border-radius: 16px; padding: 20px; }
-input, textarea { background: rgba(30,41,59,0.9) !important; border: 1px solid rgba(56,189,248,0.2) !important; color: #e2e8f0 !important; border-radius: 10px !important; padding: 14px !important; }
-input:focus, textarea:focus { border-color: #38bdf8 !important; box-shadow: 0 0 15px rgba(56,189,248,0.3) !important; }
-.gradio-dropdown select { background: rgba(30,41,59,0.9) !important; border: 1px solid rgba(56,189,248,0.2) !important; color: #e2e8f0 !important; }
-.gradio-dataset { background: rgba(30,41,59,0.7); border: 1px solid rgba(56,189,248,0.1); border-radius: 10px; max-height: 400px; overflow-y: auto; }
-.gradio-markdown h1, .gradio-markdown h2, .gradio-markdown h3 { color: #38bdf8; }
-::-webkit-scrollbar { width: 10px; }
-::-webkit-scrollbar-track { background: rgba(56,189,248,0.05); }
-::-webkit-scrollbar-thumb { background: linear-gradient(180deg, #38bdf8, #8b5cf6); border-radius: 5px; }
-.warning { text-align: center; color: #94a3b8; font-size: 12px; margin: 16px 0; padding: 14px; background: rgba(56,189,248,0.05); border-radius: 8px; border-left: 3px solid #38bdf8; }
-"""
-
-class ChatMgr:
-    def __init__(self):
-        self.rate_limit = defaultdict(list)
-        self.metrics = defaultdict(list)
-    
-    def is_limited(self, user, limit=30, window=60):
-        now = datetime.now()
-        key = f"r_{user}"
-        self.rate_limit[key] = [ts for ts in self.rate_limit[key] if (now - ts).seconds < window]
-        if len(self.rate_limit[key]) >= limit: return True
-        self.rate_limit[key].append(now)
-        return False
+def predict_chat(image, text, max_tokens=256):
+    if image is None:
+        return "❌ Vui lòng upload ảnh!"
     
-    def record(self, user, metric, val):
-        self.metrics[user].append({"type": metric, "value": val})
-        if len(self.metrics[user]) > 100: self.metrics[user].pop(0)
-    
-    def title(self, text): return ' '.join(re.sub(r'[^a-zA-Z0-9\s]', '', text).split()[:8])[:50] or "Chat"
-
-mgr = ChatMgr()
-
-def get_hist(profile):
-    if not supabase or not profile: return [], gr.Dataset(samples=[])
-    try:
-        res = supabase.table("chats").select("*").eq("username", profile.username).order("last_updated", desc=True).limit(50).execute()
-        data = res.data or []
-        samples = [[f"💬 {d['title'][:35]}", f"{d.get('message_count', 0)} msgs | {d.get('model_used', 'N/A').split('/')[-1][:15]}"] for d in data]
-        return data, gr.Dataset(samples=samples)
-    except: return [], gr.Dataset(samples=[])
-
-def load_chat(evt, db):
-    try: return db[evt.index]['history'], db[evt.index]['session_id']
-    except: return [], str(uuid.uuid4())
-
-def save_chat(user, sid, title, hist, model):
-    if not supabase or not user: return False
-    try:
-        msg_count = len([m for m in hist if m['role'] == 'user'])
-        supabase.table("chats").upsert({
-            "username": user, "session_id": sid, "title": title, "history": hist,
-            "message_count": msg_count, "model_used": model,
-            "last_updated": datetime.utcnow().isoformat(), "created_at": datetime.utcnow().isoformat()
-        }, on_conflict="session_id").execute()
-        return True
-    except: return False
-
-def get_stats(profile):
-    if not supabase or not profile: return "📊 No data"
     try:
-        res = supabase.table("chats").select("message_count").eq("username", profile.username).execute()
-        data = res.data or []
-        total_msgs = sum(d.get('message_count', 0) for d in data)
-        return f"📊 Chats: **{len(data)}** | Messages: **{total_msgs}** | Avg: **{total_msgs // max(len(data), 1)}**"
-    except: return "📊 Error"
-
-def chat_engine(msg, hist, sid, model, prompt, profile):
-    text = msg.get("text", "").strip() if isinstance(msg, dict) else str(msg).strip()
-    if not text: yield gr.update(value={"text": ""}), hist, sid, "Empty"
-    else:
-        user = profile.username if profile else "guest"
-        if mgr.is_limited(user):
-            yield gr.update(value={"text": ""}), hist, sid, "Rate limited"
-        else:
-            start = time.time()
-            try:
-                model_id = MODELS[model]["id"] if model in MODELS else list(MODELS.values())[0]["id"]
-                client = InferenceClient(model=model_id, token=HF_TOKEN)
-            except:
-                client = InferenceClient(model="Qwen/Qwen2.5-7B-Instruct", token=HF_TOKEN)
-            
-            msgs = [{"role": "system", "content": PROMPTS.get(prompt, list(PROMPTS.values())[0])}]
-            msgs.extend(hist[-15:])
-            msgs.append({"role": "user", "content": text})
-            new_hist = hist + [{"role": "user", "content": text}]
-            ans = ""
-            
-            yield gr.update(value={"text": ""}, interactive=False), new_hist + [{"role": "assistant", "content": "..."}], sid, "..."
-            try:
-                for token in client.chat_completion(messages=msgs, max_tokens=2048, stream=True):
-                    if token.choices[0].delta.content:
-                        ans += token.choices[0].delta.content
-                        yield gr.update(value={"text": ""}, interactive=True), new_hist + [{"role": "assistant", "content": ans}], sid, f"{len(ans)}"
-                
-                final = new_hist + [{"role": "assistant", "content": ans}]
-                if profile: save_chat(profile.username, sid, mgr.title(text), final, model)
-                mgr.record(user, "response_time", time.time() - start)
-                yield gr.update(value={"text": ""}, interactive=True), final, sid, "✅"
-            except Exception as e:
-                yield gr.update(value={"text": ""}, interactive=True), new_hist + [{"role": "assistant", "content": f"Error: {str(e)[:50]}"}], sid, "❌"
+        # Messages multimodal
+        messages = [{
+            "role": "user",
+            "content": [
+                {"type": "image"},
+                {"type": "text", "text": text or "Mô tả chi tiết ảnh này?"}
+            ]
+        }]
+        
+        # Process input
+        text_inputs = processor.apply_chat_template(
+            messages, 
+            add_generation_prompt=True, 
+            return_tensors="pt"
+        )
+        image_inputs = processor.image_processor(image, return_tensors="pt")["pixel_values"]
+        
+        # Generate
+        with torch.no_grad():
+            inputs = {
+                "input_ids": text_inputs.to(model.device),
+                "pixel_values": image_inputs.to(model.device),
+                "max_new_tokens": max_tokens,
+                "temperature": 0.7,
+                "do_sample": True,
+                "pad_token_id": tokenizer.eos_token_id
+            }
+            output = model.generate(**inputs)
+            response = processor.decode(output[0], skip_special_tokens=True)
+        
+        return response.split("assistant")[-1].strip()
+    
+    except Exception as e:
+        return f"❌ Lỗi: {str(e)}"
 
-with gr.Blocks(theme=gr.themes.Slate()) as demo:
-    sid = gr.State(str(uuid.uuid4()))
-    db = gr.State([])
+# Gradio Interface
+with gr.Blocks(
+    title="SaitenOne-1.0 Vision Chat",
+    theme=gr.themes.Soft()
+) as demo:
+    
+    gr.Markdown("# 🖼️ SaitenOne-1.0 Vision Chat")
+    gr.Markdown("Upload ảnh + hỏi → AI phân tích ngay!")
     
-    with gr.Row(equal_height=False):
-        with gr.Column(scale=1, elem_classes="sidebar-node"):
-            gr.HTML("<div class='logo'>⚡ NODEAI PRO</div>")
-            profile = gr.LoginButton(value="🔐 Login", size="sm")
-            
-            with gr.Row(): 
-                gr.Button("➕ New", variant="primary", scale=1, size="sm")
-                gr.Button("🗑️ Delete", scale=1, size="sm")
-            
-            gr.Markdown("### 🤖 Model")
-            model = gr.Dropdown(list(MODELS.keys()), value=list(MODELS.keys())[0], show_label=False)
-            
-            gr.Markdown("### 🎯 Personality")
-            prompt = gr.Dropdown(list(PROMPTS.keys()), value=list(PROMPTS.keys())[0], show_label=False)
-            
-            gr.Markdown("### 📚 History")
-            hist_list = gr.Dataset(components=[gr.Textbox(visible=False)]*2, samples=[], type="index")
-            
-            gr.Markdown("### 📊 Stats")
-            stats = gr.Markdown("Ready")
-            gr.HTML("<div class='warning'>🔒 Your data is private & encrypted</div>")
+    with gr.Row():
+        with gr.Column(scale=1):
+            image_input = gr.Image(type="pil", label="📸 Upload ảnh")
+            text_input = gr.Textbox(
+                label="❓ Câu hỏi", 
+                placeholder="Ảnh này chứa gì? Có mấy người? Mô tả chi tiết...",
+                lines=2
+            )
+            max_tokens = gr.Slider(128, 512, value=256, label="📝 Max tokens")
+            btn = gr.Button("🚀 Phân tích", variant="primary")
         
-        with gr.Column(scale=4):
-            gr.HTML("<div style='text-align:center;margin:20px;'><h1 style='color:#38bdf8;'>🚀 NodeAI Genesis Pro</h1><p style='color:#94a3b8;'>Enterprise AI Chat • Multi-Model • Real-time Streaming</p></div>")
-            chat = gr.Chatbot(show_label=False, height="65vh")
-            
-            with gr.Row():
-                status = gr.Textbox(value="Ready", interactive=False, scale=2)
-                model_info = gr.Textbox(value="", interactive=False, scale=1)
-            
-            inp = gr.MultimodalTextbox(show_label=False, placeholder="Ask anything... (Ctrl+Enter)")
-            gr.HTML("<div class='warning'>⚡ Streaming responses • Auto-saves</div>")
-
-demo.load(lambda p: get_hist(p), [profile], [db, hist_list]).then(lambda p: get_stats(p), [profile], [stats])
-hist_list.select(lambda e, d: load_chat(e, d), [hist_list, db], [chat, sid])
-inp.submit(chat_engine, [inp, chat, sid, model, prompt, profile], [inp, chat, sid, status]).then(lambda p: get_hist(p), [profile], [db, hist_list]).then(lambda p: get_stats(p), [profile], [stats])
+        with gr.Column(scale=1):
+            output = gr.Textbox(label="🤖 Kết quả", lines=10)
+    
+    # Examples
+    gr.Examples(
+        examples=[
+            ["ảnh mèo", "Đây là con gì?"],
+            ["ảnh người", "Có bao nhiêu người trong ảnh?"],
+            ["ảnh đồ vật", "Mô tả chi tiết những vật thể trong ảnh"],
+        ],
+        inputs=[image_input, text_input]
+    )
+    
+    btn.click(predict_chat, [image_input, text_input, max_tokens], output)
 
 if __name__ == "__main__":
-    demo.queue().launch(css=CSS, share=False)
+    demo.launch()