Update app.py

app.py

@@ -10,10 +10,13 @@ from tokenization_xgen import XgenTokenizer
 title = "Welcome to 🙋🏻‍♂️Tonic's😈Xgen-8K Chat!"
 description = "Interestingly there simply wasnt a public demo for Xgen, So I made one. You can use [Salesforce/xgen-7b-8k-inst](https://huggingface.co/Salesforce/xgen-7b-8k-inst) via API using Gradio by scrolling down and clicking Use 'Via API' or privately by [cloning this space on huggingface](https://huggingface.co/spaces/Tonic1/Xgen?duplicate=true) . [Join my active builders' server on discord](https://discord.gg/VqTxc76K3u). Let's build together!."
 
+os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:50'
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 model_name = "Salesforce/xgen-7b-8k-base"
 tokenizer = XgenTokenizer.from_pretrained("./")
-model = AutoModelForCausalLM.from_pretrained(model_name, dtype=torch.bfloat16, device_map="auto")
+model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
+model = model.to(dtype=torch.bfloat16)
+model = model.to(device)
 
 class XgenChatBot:
     def __init__(self, model, tokenizer, system_message="You are Xgen, an AI language model created by Tonic-AI. You are a cautious assistant. You carefully follow instructions. You are helpful and harmless and you follow ethical guidelines and promote positive behavior."):