Update app.py

app.py

@@ -2,6 +2,8 @@ import gradio as gr
 import random
 import time
 from ctransformers import AutoModelForCausalLM
+from dl_hf_model import dl_hf_model
+
 
 params = {
         "max_new_tokens":512,
@@ -12,7 +14,9 @@ params = {
         "batch_size": 8}
 
 
-llm = AutoModelForCausalLM.from_pretrained("Aspik101/trurl-2-7b-GGML", model_type="llama")
+url = "https://huggingface.co/Aspik101/trurl-2-7b-GGML/blob/main/trurl-2-7b.ggmlv3.q8_0.bin"
+model_loc, file_size = dl_hf_model(url)
+llm = AutoModelForCausalLM.from_pretrained(model_loc, model_type="llama")
 
 with gr.Blocks() as demo:
     chatbot = gr.Chatbot()