Update app.py

app.py

@@ -1,21 +1,12 @@
 from fastapi import FastAPI
 import uvicorn
-import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
 model_name = "Tap-M/Luna-AI-Llama2-Uncensored"
 
-bnb_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_quant_type="nf4",       # Must be 'nf4' for CPU compatibility
-    bnb_4bit_compute_dtype="float32", # Use float32 for CPU computation
-    bnb_4bit_use_double_quant=True
-)
-
 model = AutoModelForCausalLM.from_pretrained(
     model_name,  # Example model
     device_map="auto",            # Auto-distribute across GPU/CPU
-    quantization_config=bnb_config,
     offload_folder="./offload",  # Temporary directory
     trust_remote_code=True        # Required for some models
 )