Download model to /tmp to avoid permission error

app.py

@@ -1,21 +1,26 @@
-import os, gradio as gr
+import os, gradio as gr, requests, tempfile
 from llama_cpp import Llama
 
 MODEL_URL = (
     "https://huggingface.co/fdtn-ai/Foundation-Sec-8B-Q4_K_M-GGUF/"
     "resolve/main/foundation-sec-8b-q4_k_m.gguf"
 )
-MODEL_PATH = "foundation-sec-8b-q4_k_m.gguf"
 
-# download once (Gradio blocks until the file exists)
+# writable directory
+CACHE_DIR = "/tmp"
+MODEL_PATH = os.path.join(CACHE_DIR, "foundation-sec-8b-q4_k_m.gguf")
+
+# download only once
 if not os.path.exists(MODEL_PATH):
-    import requests, time
+    print("Downloading model … (~4.9 GB)")
     with requests.get(MODEL_URL, stream=True) as r:
         r.raise_for_status()
         with open(MODEL_PATH, "wb") as f:
             for chunk in r.iter_content(chunk_size=8192):
                 f.write(chunk)
+    print("Download finished.")
 
+# load model
 llm = Llama(model_path=MODEL_PATH, n_ctx=4096, verbose=False)
 
 def chat_fn(message, history):