Update app.py

initialized config sooner

app.py

@@ -13,15 +13,6 @@ import gradio as gr
 # local_llm = AutoModel.from_pretrained("TheBloke/zephyr-7B-beta-GGUF")
 
 
-local_llm = CTransformers(
-    model = "TheBloke/zephyr-7B-beta-GGUF",
-    model_file = "zephyr-7b-beta.Q4_0.gguf",
-    model_type="mistral",
-    lib="avx2", #for CPU use
-    **config
-)
-
-
 config = {
     "max_new_token": 1024,
     "repetition_penalty": 1.1,
@@ -32,6 +23,18 @@ config = {
     "threads": int(os.cpu_count() / 2),
 }
 
+
+local_llm = CTransformers(
+    model = "TheBloke/zephyr-7B-beta-GGUF",
+    model_file = "zephyr-7b-beta.Q4_0.gguf",
+    model_type="mistral",
+    lib="avx2", #for CPU use
+    **config
+)
+
+
+
+
 llm_init = CTransformers(model=local_llm, model_type="mistral", lib="avx2", **config)
 
 prompt_template = """Use the following piece of information to answers the question asked by the user.