Update app.py

app.py

@@ -1,167 +1,86 @@
 import gradio as gr
-import torch
-import pandas as pd
-import io
-import os 
-from transformers import AutoModelForCausalLM, AutoTokenizer
-from peft import PeftModel
+from llama_cpp import Llama
+from huggingface_hub import hf_hub_download
 
 # ----------------------------------------------------------------------
-# 1. MODEL SETUP (Load only once)
+# 1. إعداد النموذج (سحب من مستودع عام لتجاوز حدود التخزين)
 # ----------------------------------------------------------------------
 
-# العودة إلى نموذج Mistral-7B لضمان التوافق مع محوّل LoRA
-BASE_MODEL_ID = "mistralai/Mistral-7B-Instruct-v0.2" 
-LORA_ADAPTER_ID = "RootSystem2101/ZeroCyber-SLM-LoRA-Adapter"
+# نستخدم نموذج Qwen2.5-3B القوي والصغير (متوفر مسبقاً ولا يحتاج رفع)
+REPO_ID = "Qwen/Qwen2.5-3B-Instruct-GGUF"
+FILENAME = "qwen2.5-3b-instruct-q4_k_m.gguf"
 
-def load_zerocyber_model():
-    print("Loading Tokenizer...")
-    # تحميل التوكنايزر الخاص بـ Mistral
-    tokenizer = AutoTokenizer.from_pretrained(LORA_ADAPTER_ID)
-    
-    print("Loading Base Model in 4-bit...")
-    # device_map="sequential" يضمن تفريغ الذاكرة بشكل تسلسلي وفعال لحل مشكلة FATAL ERROR
-    model = AutoModelForCausalLM.from_pretrained(
-        BASE_MODEL_ID,
-        load_in_4bit=True,
-        torch_dtype=torch.float16,
-        device_map="sequential", 
-        # تم حذف offload_folder
-    )
-    
-    print("Merging LoRA Adapter...")
-    model = PeftModel.from_pretrained(model, LORA_ADAPTER_ID)
-    model = model.merge_and_unload()
-    model.eval()
-    
-    return tokenizer, model
-
-try:
-    ZEROCYBER_TOKENIZER, ZEROCYBER_MODEL = load_zerocyber_model()
-except Exception as e:
-    print(f"FATAL ERROR during model loading: {e}")
-    ZEROCYBER_TOKENIZER = None 
-    ZEROCYBER_MODEL = None
+def load_model():
+    print(f"Downloading model {FILENAME} from {REPO_ID}...")
+    try:
+        # تحميل النموذج إلى الذاكرة المؤقتة للسيرفر
+        model_path = hf_hub_download(
+            repo_id=REPO_ID, 
+            filename=FILENAME
+        )
+        print(f"Model downloaded to: {model_path}")
+        
+        # تشغيل النموذج (CPU)
+        llm = Llama(
+            model_path=model_path,
+            n_ctx=2048,
+            n_threads=2, 
+            verbose=True
+        )
+        return llm
+    except Exception as e:
+        print(f"Error loading model: {e}")
+        return None
 
+# تحميل النموذج عند الإقلاع
+ZEROCYBER_MODEL = load_model()
 
 # ----------------------------------------------------------------------
-# 2. CORE INFERENCE FUNCTIONS (FASTEST GENERATION MODE)
+# 2. منطق التحليل
 # ----------------------------------------------------------------------
 
-def generate_response(prompt_text: str):
-    """وظيفة توليد الاستجابة المُسرَّعة القصوى (Greedy Search)."""
+def generate_response(prompt_text, file_obj):
     if ZEROCYBER_MODEL is None:
-        return "❌ Model loading failed. Please check the command line for errors."
-        
-    # العودة إلى تنسيق المطالبة الخاص بـ Mistral
-    formatted_prompt = f"<s>[INST] {prompt_text} [/INST]"
-    inputs = ZEROCYBER_TOKENIZER(formatted_prompt, return_tensors="pt").to(ZEROCYBER_MODEL.device)
+        return "❌ Error: Model failed to load."
 
-    try:
-        with torch.no_grad():
-            outputs = ZEROCYBER_MODEL.generate(
-                **inputs, 
-                max_new_tokens=128,  # تثبيت السرعة
-                do_sample=False,      
-                pad_token_id=ZEROCYBER_TOKENIZER.eos_token_id 
-            )
-            
-        response = ZEROCYBER_TOKENIZER.decode(outputs[0], skip_special_tokens=True)
-        return response.split("[/INST]")[1].strip()
-        
-    except Exception as e:
-        return f"❌ Internal Error during Inference: {e}"
-
-
-def analyze_log_file(file_path: str):
-    """وظيفة تحليل ملف Log/CSV بأمان ضد مشاكل الترميز."""
-    
-    # 1. Safely read file content using common encodings
-    try:
-        with open(file_path, 'r', encoding='utf-8', errors='strict') as f:
-            log_content = f.read()
-    except UnicodeDecodeError:
+    # قراءة الملف إذا وجد
+    context = ""
+    if file_obj:
         try:
-            with open(file_path, 'r', encoding='latin-1', errors='strict') as f:
-                log_content = f.read()
+            with open(file_obj.name, 'r', encoding='utf-8', errors='ignore') as f:
+                content = f.read()[:2000]
+            context = f"\n\nFile Content to Analyze:\n{content}\n"
         except Exception as e:
-            return f"❌ File Reading Error: {e}\nCould not read the file using common text encodings."
+            return f"Error reading file: {e}"
 
-    if not log_content.strip():
-        return "⚠️ Uploaded file is empty or does not contain readable text content."
-        
-    # 2. Prompt Engineering for Cybersecurity Report (Arabic language enforced)
-    
-    truncated_content = log_content[:5000]
-    
-    prompt = f"""
-    You are a specialized cybersecurity analyst. Analyze the following log file content.
-    
-    Your task is to:
-    1. Identify the most critical security events or errors.
-    2. Pinpoint suspicious patterns or explicit attack attempts.
-    3. **Generate a structured report in ARABIC (اللغة العربية)** including a clear summary and recommendations.
-    4. Provide immediate, actionable steps for defenders (Defenders) in a bulleted list.
-    
-    Log Content (Truncated):
-    ---
-    {truncated_content}
-    ---
-    """
-    
-    print(f"Analyzing log content from file: {os.path.basename(file_path)}")
-    return generate_response(prompt)
-
-
-# ----------------------------------------------------------------------
-# 3. UNIFIED GRADIO INTERFACE LOGIC (NO PROGRESS INDICATOR)
-# ----------------------------------------------------------------------
-
-# تم حذف المؤشر لحل مشكلة الـ 404
-def unified_interface(question: str, log_file): 
-    """Handles either text input or file upload."""
-    
-    if log_file is not None:
-        return analyze_log_file(log_file.name)
-    
-    elif question.strip():
-        print(f"Received question: {question}")
-        
-        # Language steering
-        if any(c in question for c in 'ءآأبتثجحخدذرزسشصضطظعغفقكلمنهويى'):
-             prompt_with_lang = f"أجب باللغة العربية. السؤال هو: {question}"
-        else:
-             prompt_with_lang = f"Answer in English. The question is: {question}"
+    # تجهيز البرومبت
+    # نماذج Qwen/Mistral تفضل هذا التنسيق
+    full_prompt = f"<|im_start|>system\nYou are a Cybersecurity Analyst.<|im_end|>\n<|im_start|>user\n{prompt_text}\n{context}<|im_end|>\n<|im_start|>assistant\n"
 
-        return generate_response(prompt_with_lang)
+    output = ZEROCYBER_MODEL(
+        full_prompt,
+        max_tokens=512,
+        stop=["<|im_end|>"],
+        echo=False
+    )
     
-    else:
-        return "Please submit a question or upload a file for analysis."
-
+    return output['choices'][0]['text']
 
 # ----------------------------------------------------------------------
-# 4. GRADIO INTERFACE BUILD (Professional English Titles)
+# 3. واجهة المستخدم
 # ----------------------------------------------------------------------
 
 if __name__ == "__main__":
-    
-    input_components = [
-        gr.Textbox(label="1. Ask your Cybersecurity Inquiry:", placeholder="Example: What are the steps to secure a web server?"),
-        gr.File(label="2. Or Upload any Log/Text File for Analysis:", file_types=None) 
-    ]
-    
-    output_component = gr.Markdown(label="ZeroCyber-SLM Report / Response")
-
     interface = gr.Interface(
-        fn=unified_interface,
-        inputs=input_components,
-        outputs=output_component,
-        title="ZeroCyber-SLM: Security analysis and response platform",
-        description="A specialized application for responding to security inquiries and analyzing Log/CSV files to identify incidents and provide actionable recommendations for defenders.",
+        fn=generate_response,
+        inputs=[
+            gr.Textbox(label="1. استفسارك الأمني:", placeholder="كيف أقوم بتأمين قاعدة البيانات؟"),
+            gr.File(label="2. تحليل ملف (Log/Code)")
+        ],
+        outputs=gr.Textbox(label="التقرير الأمني"),
+        title="🛡️ ZeroCyber Cloud Platform",
+        description="تطبيق سحابي للتحليل الأمني يعمل بنموذج Qwen2.5-3B (GGUF).",
         allow_flagging="never"
     )
-
-    if ZEROCYBER_MODEL is not None:
-         interface.launch(share=True) 
-    else:
-        print("\n❌ Interface failed to start due to model loading failure.")
+    
+    interface.launch()