Update app.py

app.py

@@ -5,7 +5,7 @@ from llama_cpp import Llama          # GGUF CPU backend
 # ---------- model loading (one‑time) ----------
 MODEL_REPO = "MaziyarPanahi/gemma-2b-it-GGUF"
 MODEL_FILE = "gemma-2b-it.Q4_K_M.gguf"     # 1.6 GB 4‑bit
-CTX_SIZE   = 2048
+CTX_SIZE   = 4096
 
 model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
 llm = Llama(
@@ -63,8 +63,8 @@ def analyze_ads(file):
     # stream=True → токены сразу в логах, ответ ~25‑30 с
     stream = llm(
         prompt,
-        max_tokens=250,
-        temperature=0.6,
+        max_tokens=800,
+        temperature=0.4,
         top_p=0.8,
         stream=True
     )