Update app.py

app.py

@@ -7,130 +7,205 @@ import gc
 from datetime import datetime
 from bs4 import BeautifulSoup
 from contextlib import asynccontextmanager
+
 from fastapi import FastAPI, Request
 from fastapi.responses import JSONResponse
 from fastapi.middleware.cors import CORSMiddleware
-from llama_cpp import Llama  
+
+from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
 
 # ==========================================
 # 1. CONFIGURATION & CORE SETUP
 # ==========================================
+
 SECRET_KEY = os.environ.get("REAPERAI_SECRET", "jan30")
+
 MODEL_REPO = "bartowski/Qwen2.5-7B-Instruct-GGUF"
-MODEL_FILE = "Qwen2.5-7B-Instruct-Q4_K_M.gguf" # Specific filename required for download
+MODEL_FILENAME = "Qwen2.5-7B-Instruct-Q4_K_M.gguf"
 
 chat_memory = {}
 MAX_GLOBAL_USERS = 50
 memory_lock = threading.Lock()
+
+# Only 1 inference at a time (CPU safety)
 model_semaphore = threading.Semaphore(1)
-llm = None  # Initialize as None to avoid NameError
+
+llm = None  # will be set at startup
 
 # ==========================================
-# 2. MODEL LOADING (FIXED FOR COMPATIBILITY)
+# 2. MODEL LOADING (HF Spaces Compatible)
 # ==========================================
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
-    print(f"--- [SYSTEM] Downloading & Loading GGUF: {MODEL_FILE} ---")
     global llm
+    print(f"--- [SYSTEM] Downloading model from HF: {MODEL_REPO} ---")
+
     try:
-        # Step 1: Securely download the model file to local cache
-        path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
-        
-        # Step 2: Initialize using the local path
+        model_path = hf_hub_download(
+            repo_id=MODEL_REPO,
+            filename=MODEL_FILENAME
+        )
+
+        print(f"--- [SYSTEM] Loading GGUF model from: {model_path} ---")
+
         llm = Llama(
-            model_path=path,
-            n_ctx=2048,           
-            n_threads=2,          
+            model_path=model_path,
+            n_ctx=2048,
+            n_threads=4,
+            n_batch=512,
             verbose=False
         )
-        print("--- [SYSTEM] ReaperAI Core Online ---")
+
+        print("--- [SYSTEM] ReaperAI Core Online (GGUF) ---")
+
     except Exception as e:
-        print(f"--- [CRITICAL ERROR] Loading Failed: {e} ---")
+        print(f"--- [CRITICAL ERROR] Model load failed: {e} ---")
+        llm = None
+
     yield
 
+# ==========================================
+# 3. FASTAPI SETUP
+# ==========================================
+
 app = FastAPI(title="ReaperAI Secure Core", lifespan=lifespan)
-app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["POST"], allow_headers=["*"])
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_methods=["POST"],
+    allow_headers=["*"]
+)
 
 # ==========================================
-# 3. AUTONOMOUS TOOLS (REFINED)
+# 4. AUTONOMOUS TOOLS
 # ==========================================
+
 def ddg_search(query):
-    print(f"--- [TOOL] Searching Web: {query} ---")
     try:
         headers = {"User-Agent": "Mozilla/5.0"}
-        res = requests.get("https://html.duckduckgo.com/html/", params={"q": query}, headers=headers, timeout=5)
+        res = requests.get(
+            "https://html.duckduckgo.com/html/",
+            params={"q": query},
+            headers=headers,
+            timeout=6
+        )
         soup = BeautifulSoup(res.text, "html.parser")
-        results = [f"- {r.select_one('.result__a').get_text(strip=True)}: {r.select_one('.result__snippet').get_text(strip=True)}" 
-                   for r in soup.select(".result")[:3]]
-        return f"\n[REAL-TIME KNOWLEDGE]:\n" + "\n".join(results) + "\n" if results else ""
-    except: return ""
+
+        results = []
+        for r in soup.select(".result")[:3]:
+            title = r.select_one(".result__a")
+            snippet = r.select_one(".result__snippet")
+            if title and snippet:
+                results.append(f"- {title.get_text(strip=True)}: {snippet.get_text(strip=True)}")
+
+        return "\n[REAL-TIME KNOWLEDGE]\n" + "\n".join(results) + "\n" if results else ""
+    except:
+        return ""
 
 def jina_read(url):
-    print(f"--- [TOOL] Reading Source: {url} ---")
     try:
-        res = requests.get(f"https://r.jina.ai/{url}", headers={"x-respond-with": "text"}, timeout=8)
-        return f"\n[SOURCE CONTENT]:\n{res.text[:1200]}\n" if res.status_code == 200 else ""
-    except: return ""
+        res = requests.get(
+            f"https://r.jina.ai/{url}",
+            headers={"x-respond-with": "text"},
+            timeout=8
+        )
+        if res.status_code == 200:
+            return f"\n[SOURCE CONTENT]\n{res.text[:1200]}\n"
+    except:
+        pass
+    return ""
 
 def fast_intent_detection(message):
     url_match = re.search(r"(https?://\S+)", message)
-    if url_match: return "URL", url_match.group(1)
+    if url_match:
+        return "URL", url_match.group(1)
+
     search_keywords = ["who is", "price", "latest", "current", "news", "today", "weather", "score", "stock"]
     if any(k in message.lower() for k in search_keywords):
         return "SEARCH", message
+
     return "CHAT", None
 
 # ==========================================
-# 4. SECURE ENDPOINT
+# 5. SECURITY MIDDLEWARE
 # ==========================================
+
 @app.middleware("http")
 async def security_guard(request: Request, call_next):
-    if request.url.path == "/": return await call_next(request)
+    if request.url.path == "/":
+        return await call_next(request)
+
     if request.headers.get("x-reaperai-key") != SECRET_KEY:
         return JSONResponse(status_code=403, content={"error": "ACCESS_DENIED"})
+
     return await call_next(request)
 
+# ==========================================
+# 6. CHAT ENDPOINT
+# ==========================================
+
 @app.post("/chat")
-def chat_endpoint(request: Request, payload: dict):
+async def chat_endpoint(request: Request):
     if llm is None:
-        return JSONResponse(status_code=503, content={"error": "Model not initialized"})
+        return JSONResponse(status_code=503, content={"error": "MODEL_NOT_READY"})
+
+    payload = await request.json()
 
     user_id = payload.get("userId", "default")
     message = payload.get("message", "").strip()
+
     current_date = datetime.now().strftime("%A, %B %d, %Y")
-    
+
     with memory_lock:
         if len(chat_memory) > MAX_GLOBAL_USERS:
             chat_memory.pop(next(iter(chat_memory)))
         history = chat_memory.get(user_id, [])[-5:]
 
     intent, data = fast_intent_detection(message)
-    context = jina_read(data) if intent == "URL" else (ddg_search(message) if intent == "SEARCH" else "")
 
-    prompt_messages = [{"role": "system", "content": f"You are ReaperAI. Precise and concise. Today: {current_date}"}]
+    context = ""
+    if intent == "URL":
+        context = jina_read(data)
+    elif intent == "SEARCH":
+        context = ddg_search(message)
+
+    prompt_messages = [
+        {"role": "system", "content": f"You are ReaperAI. Precise and concise. Today is {current_date}."}
+    ]
+
     for h in history:
-        prompt_messages.append({"role": "user", "content": h['u']})
-        prompt_messages.append({"role": "assistant", "content": h['a']})
-    
-    final_input = f"Context: {context}\nQuery: {message}" if context else message
+        prompt_messages.append({"role": "user", "content": h["u"]})
+        prompt_messages.append({"role": "assistant", "content": h["a"]})
+
+    final_input = f"Context:\n{context}\n\nUser query:\n{message}" if context else message
     prompt_messages.append({"role": "user", "content": final_input})
 
     with model_semaphore:
         try:
-            output = llm.create_chat_completion(messages=prompt_messages, max_tokens=400, temperature=0.7)
+            output = llm.create_chat_completion(
+                messages=prompt_messages,
+                max_tokens=400,
+                temperature=0.7
+            )
             response = output["choices"][0]["message"]["content"].strip()
         except Exception as e:
             response = f"Core Error: {str(e)}"
 
     with memory_lock:
-        if user_id not in chat_memory: chat_memory[user_id] = []
-        chat_memory[user_id].append({"u": message, "a": response})
+        chat_memory.setdefault(user_id, []).append({"u": message, "a": response})
         chat_memory[user_id] = chat_memory[user_id][-8:]
 
     gc.collect()
+
     return {"intent": intent, "response": response}
 
+# ==========================================
+# 7. LOCAL DEV ENTRY
+# ==========================================
+
 if __name__ == "__main__":
     import uvicorn
-    uvicorn.run(app, host="0.0.0.0", port=7860, timeout_keep_alive=60)
+    uvicorn.run(app, host="0.0.0.0", port=7860)