from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "deepseek-ai/deepseek-coder-1.3b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="cpu")

inputs = tokenizer("def add(x, y):", return_tensors="pt")
outputs = model.generate(**inputs, max_length=30)
print(tokenizer.decode(outputs[0]))