app.py

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cognitivecomputations/dolphin-2.6-mixtral-8x7b")
model = AutoModelForCausalLM.from_pretrained("cognitivecomputations/dolphin-2.6-mixtral-8x7b")