app.py

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("TheBloke/Yarn-Mistral-7B-128k-GPTQ")
model = AutoModelForCausalLM.from_pretrained("TheBloke/Yarn-Mistral-7B-128k-GPTQ")