Full Rebrand: Transitioned intelligence engine from Llama to Qwen

download_qwen.py

@@ -0,0 +1,46 @@
+import os
+import requests
+from tqdm import tqdm
+
+def download_qwen():
+    # Using bartowski's high-quality Qwen 3.5 4B GGUF
+    url = "https://huggingface.co/bartowski/Qwen_Qwen3.5-4B-GGUF/resolve/main/Qwen3.5-4B-Instruct-Q4_K_M.gguf?download=true"
+    local_dir = "ml_models"
+    filename = "qwen-3.5-4b.gguf"
+    save_path = os.path.join(local_dir, filename)
+    
+    if not os.path.exists(local_dir):
+        os.makedirs(local_dir)
+        
+    if os.path.exists(save_path):
+        print(f"--- INFO: Qwen model already exists at {save_path} ---")
+        return
+
+    print(f"--- STARTING DIRECT DOWNLOAD: {filename} ---")
+    print("Size: ~2.5 GB. This may take 5-10 minutes.")
+    
+    try:
+        response = requests.get(url, stream=True, timeout=30)
+        response.raise_for_status()
+        total_size = int(response.headers.get('content-length', 0))
+        
+        with open(save_path, 'wb') as f, tqdm(
+            desc=filename,
+            total=total_size,
+            unit='B',
+            unit_scale=True,
+            unit_divisor=1024,
+        ) as bar:
+            for chunk in response.iter_content(chunk_size=8192):
+                if chunk:
+                    f.write(chunk)
+                    bar.update(len(chunk))
+                    
+        print(f"\n--- SUCCESS: Qwen model saved to {save_path} ---")
+    except Exception as e:
+        print(f"\n--- ERROR: Download failed: {e} ---")
+        if os.path.exists(save_path):
+            os.remove(save_path)
+
+if __name__ == "__main__":
+    download_qwen()

scraper.py

@@ -129,17 +129,17 @@ def search_app_store(query):
     except: return []
 
 # Neural Config
-LLAMA_PATH = "ml_models/qwen-3.5-4b.gguf"
-_Llama = None
+QWEN_PATH = "ml_models/qwen-3.5-4b.gguf"
+_Qwen = None
 
-def get_llama():
-    global _Llama
-    if _Llama: return _Llama
-    if not os.path.exists(LLAMA_PATH): return None
+def get_qwen():
+    global _Qwen
+    if _Qwen: return _Qwen
+    if not os.path.exists(QWEN_PATH): return None
     try:
         from llama_cpp import Llama
-        _Llama = Llama(model_path=LLAMA_PATH, n_gpu_layers=-1, verbose=False, n_ctx=2048)
-        return _Llama
+        _Qwen = Llama(model_path=QWEN_PATH, n_gpu_layers=-1, verbose=False, n_ctx=2048)
+        return _Qwen
     except Exception: return None
 
 def get_representative_reviews(reviews, count=25):
@@ -169,8 +169,8 @@ def get_representative_reviews(reviews, count=25):
         return [r[:200] for r in reviews[:count]]
 
 def ai_generate_overview(neg_reviews, pos_reviews):
-    llm = get_llama()
-    if not llm: return "Llama engine offline. Unable to generate review overview."
+    llm = get_qwen()
+    if not llm: return "Qwen engine offline. Unable to generate review overview."
     
     try:
         # Phase 1: Hierarchical Clustering & Mini-Summarization
@@ -192,7 +192,7 @@ Reviews to analyze:
                 return res['choices'][0]['text'].strip()
             except: return ""
 
-        # Batch process to ensure we don't hit Llama's context limit in one go
+        # Batch process to ensure we don't hit Qwen's context limit in one go
         neg_mini = get_mini_summary(neg_clusters, "negative")
         pos_mini = get_mini_summary(pos_clusters, "positive")
 
@@ -224,7 +224,7 @@ POSITIVE THEMES:
         return "Error generating hierarchical overview."
 
 def ai_generate_synthesis(overview_text):
-    llm = get_llama()
+    llm = get_qwen()
     if not llm: return "Synthesis engine offline."
     
     prompt = f"""<|start_header_id|>system<|end_header_id|>
@@ -246,7 +246,7 @@ def ai_generate_battle_analysis(data_a, data_b):
     """
     Performs a side-by-side neural clash analysis between two competitors.
     """
-    llm = get_llama()
+    llm = get_qwen()
     if not llm: return "Battle engine offline."
 
     name_a = data_a.get('app_name', 'App A')
@@ -282,7 +282,7 @@ INTEL ON B: {analysis_b}
         return "Neural Clash synthesis failed. Manual comparison required."
 
 def ai_extract_search_keywords(idea_text):
-    llm = get_llama()
+    llm = get_qwen()
     if not llm:
         # Fallback to basic keyword extraction if no LLM
         words = [w for w in re.split(r'\W+', idea_text.lower()) if len(w) > 3]
@@ -302,7 +302,7 @@ App Idea: {idea_text}
         return ",".join(idea_text.split()[:5])
 
 def ai_validate_match(title1, title2):
-    llm = get_llama()
+    llm = get_qwen()
     if not llm: return "YES" # Optimistic fallback
     
     prompt = f"""<|start_header_id|>system<|end_header_id|>
@@ -374,10 +374,10 @@ def run_full_scan(gp_url=None, as_url=None):
 
     app_description = gp_desc or as_desc
         
-    # Llama cross-check for different apps
+    # Qwen cross-check for different apps
     if gp_title and as_title:
         ans = ai_validate_match(gp_title, as_title)
-        print(f"[AI Validation] Comparing '{gp_title}' vs '{as_title}' -> Llama said: {ans}")
+        print(f"[AI Validation] Comparing '{gp_title}' vs '{as_title}' -> Qwen said: {ans}")
         if "NO" in ans and "YES" not in ans:
             return {"mismatch_error": True, "gp_title": gp_title, "as_title": as_title}
 
@@ -402,7 +402,7 @@ def run_full_scan(gp_url=None, as_url=None):
     # 2. Positive Highlights (Positive 4-5)
     pos_reviews = df[(df['rating'] >= 4) & (df['review'].str.len() > 10)]['review'].dropna().tolist()
     
-    # 3. Generating General Llama Overview
+    # 3. Generating General Qwen Overview
     general_overview = ai_generate_overview(neg_reviews, pos_reviews)
     
     # 3. Sentiment Timeline (Weekly)