Update app.py

app.py

@@ -66,8 +66,8 @@ from langchain.chains import create_retrieval_chain
 #from langchain_community.llms import Ollama
 #READER_MODEL=Ollama(model="meta-llama/Meta-Llama-Guard-2-8B")
 
-from transformers import AutoModel
-lll_model = AutoModel.from_pretrained("HuggingFaceH4/zephyr-7b-beta", load_in_4bit=True)
+from transformers import AutoModelForCausalLM
+lll_model = AutoModelForCausalLM.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
 
 #READER_MODEL = "HuggingFaceH4/zephyr-7b-beta"