Update app.py

app.py

@@ -1,149 +1,47 @@
-import gradio as gr
+from fastapi import FastAPI, Request
+from fastapi.responses import StreamingResponse
 from openai import OpenAI
 import os
 import chromadb
-from chromadb.utils import embedding_functions
-import pypdf
-import uuid
 
-# --- الإعدادات الفنية ---
-STORAGE_PATH = "/data/neural_memory" if os.path.exists("/data") else "./neural_memory"
-chroma_client = chromadb.PersistentClient(path=STORAGE_PATH)
-default_ef = embedding_functions.DefaultEmbeddingFunction()
-collection = chroma_client.get_or_create_collection(name="advanced_brain", embedding_function=default_ef)
-
-# --- 1. الابتكار في الحقن (Semantic Ingestion) ---
-def advanced_ingest(file_path):
-    """حقن متقدم مع Metadata و Overlap و Normalization."""
-    try:
-        text = ""
-        filename = os.path.basename(file_path)
-        
-        if file_path.endswith('.pdf'):
-            reader = pypdf.PdfReader(file_path)
-            pages_data = [(p.extract_text(), i+1) for i, p in enumerate(reader.pages)]
-        else:
-            with open(file_path, 'r', encoding='utf-8') as f:
-                pages_data = [(f.read(), 1)]
+app = FastAPI()
 
-        documents, metadatas, ids = [], [], []
-        
-        # إعدادات الـ Chunking الاحترافية
-        chunk_size = 1000
-        overlap = 200 # تداخل 20% لضمان اتصال المعنى
-        
-        for content, page_num in pages_data:
-            content = content.replace('\t', ' ').strip() # Normalization بسيط
-            
-            for i in range(0, len(content), chunk_size - overlap):
-                chunk = content[i : i + chunk_size]
-                documents.append(chunk)
-                metadatas.append({
-                    "source": filename,
-                    "page": page_num,
-                    "length": len(chunk)
-                })
-                ids.append(str(uuid.uuid4()))
-        
-        collection.add(documents=documents, metadatas=metadatas, ids=ids)
-        return f"✅ تم حقن {len(documents)} قطعة معرفية من '{filename}' مع حفظ الميتا-داتا."
-    except Exception as e:
-        return f"❌ فشل الحقن: {str(e)}"
-
-# --- 2. الاسترجاع الذكي (Filtered Query) ---
-def smart_query(user_query, threshold=0.6):
-    """استرجاع مع تصفية حسب درجة التشابه (Score Filtering)."""
-    # نطلب نتائج أكثر ثم نصفيها
-    results = collection.query(
-        query_texts=[user_query],
-        n_results=10,
-        include=['documents', 'metadatas', 'distances']
-    )
-    
-    context_parts = []
-    for doc, meta, dist in zip(results['documents'][0], results['metadatas'][0], results['distances'][0]):
-        # في ChromaDB الـ distance الأقل تعني تشابه أكبر (0 = متطابق)
-        # نحولها إلى Score افتراضي (1 - dist)
-        score = 1 - dist
-        if score >= threshold:
-            source_info = f"[المصدر: {meta['source']} | صفحة: {meta['page']}]"
-            context_parts.append(f"{source_info}\n{doc}")
-    
-    return "\n\n---\n\n".join(context_parts) if context_parts else "لم يتم العثور على معرفة وثيقة الصلة."
-
-# --- 3. المحرك العصبي (The Engine) ---
-def neural_engine(message, history, system_prompt, base_url, api_key, temp, score_threshold):
-    client = OpenAI(
-        base_url=base_url or "https://router.huggingface.co/hf-inference/v1",
-        api_key=api_key or os.getenv("HF_TOKEN")
-    )
-    
-    knowledge = smart_query(message, threshold=score_threshold)
+# إعدادات الذاكرة (نفس منطق الكود السابق)
+STORAGE_PATH = "./neural_memory"
+chroma_client = chromadb.PersistentClient(path=STORAGE_PATH)
+collection = chroma_client.get_or_create_collection(name="advanced_brain_v6")
+
+client = OpenAI(
+    base_url="https://router.huggingface.co/hf-inference/v1",
+    api_key=os.getenv("HF_TOKEN")
+)
+
+@app.post("/v1/chat/completions")
+async def chat_proxy(request: Request):
+    data = await request.json()
+    messages = data.get("messages", [])
+    user_query = messages[-1]["content"]
     
-    enhanced_system = f"{system_prompt}\n\n[سياق المعرفة الموثق]:\n{knowledge}"
+    # البحث في الذاكرة
+    results = collection.query(query_texts=[user_query], n_results=3)
+    knowledge = "\n".join(results['documents'][0]) if results['documents'] else ""
     
-    messages = [{"role": "system", "content": enhanced_system}]
-    for u, a in history:
-        messages.append({"role": "user", "content": u})
-        messages.append({"role": "assistant", "content": a})
-    messages.append({"role": "user", "content": message})
+    # حقن المعرفة في أول رسالة (System Prompt)
+    messages.insert(0, {"role": "system", "content": f"Context: {knowledge}"})
 
-    try:
+    def stream_response():
         response = client.chat.completions.create(
             model="huihui-ai/Qwen2.5-72B-Instruct-abliterated",
             messages=messages,
-            temperature=temp,
             stream=True
         )
-        full_resp = ""
         for chunk in response:
             if chunk.choices[0].delta.content:
-                full_resp += chunk.choices[0].delta.content
-                yield full_resp
-    except Exception as e:
-        yield f"⚠️ Neural Glitch: {str(e)}"
-
-# --- 4. واجهة المستخدم المتقدمة ---
-with gr.Blocks(theme=gr.themes.Soft(primary_hue="teal"), css=".gradio-container {background: #f9f9f9}") as demo:
-    gr.Markdown("# 🧬 Neural OS v4.0 (Semantic Edition)")
-    
-    with gr.Tabs():
-        with gr.Tab("💬 Interaction Console"):
-            chatbot = gr.Chatbot(height=600, show_label=False)
-            with gr.Row():
-                msg_input = gr.Textbox(placeholder="اسأل العقل الاصطناعي...", scale=8)
-                submit_btn = gr.Button("نفاذ", variant="primary")
-
-        with gr.Tab("📚 Knowledge Vault"):
-            with gr.Row():
-                with gr.Column():
-                    file_input = gr.File(label="وثائق التدريب (PDF/TXT)")
-                    upload_btn = gr.Button("بدء المعالجة الدلالية", variant="secondary")
-                with gr.Column():
-                    status_log = gr.TextArea(label="سجل العمليات", interactive=False)
-            
-        with gr.Tab("⚙️ Control Panel"):
-            with gr.Row():
-                with gr.Column():
-                    sys_p = gr.TextArea(label="System Persona", value="أنت محرك معرفي يستند إلى وثائق رسمية.")
-                    score_th = gr.Slider(0.0, 1.0, 0.4, label="Relevance Threshold", info="كلما زاد، كان الاسترجاع أدق وأقل كمية.")
-                with gr.Column():
-                    endpoint = gr.Textbox(label="API Endpoint")
-                    token = gr.Textbox(label="Access Token", type="password")
-                    temp = gr.Slider(0, 1.5, 0.7, label="Temperature")
-
-    # التفاعلات
-    upload_btn.click(lambda files: "\n".join([advanced_ingest(f.name) for f in files]), [file_input], [status_log])
-    
-    def chat_logic(m, h, sp, url, t, tmp, th):
-        gen = neural_engine(m, h, sp, url, t, tmp, th)
-        h.append([m, ""])
-        for res in gen:
-            h[-1][1] = res
-            yield "", h
+                yield f"data: {chunk.choices[0].delta.content}\n\n"
+        yield "data: [DONE]\n\n"
 
-    submit_btn.click(chat_logic, [msg_input, chatbot, sys_p, endpoint, token, temp, score_th], [msg_input, chatbot])
-    msg_input.submit(chat_logic, [msg_input, chatbot, sys_p, endpoint, token, temp, score_th], [msg_input, chatbot])
+    return StreamingResponse(stream_response(), media_type="text/event-stream")
 
 if __name__ == "__main__":
-    demo.queue().launch()
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=8000)