from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

def generate_response(input):
    '''
    simple test for the model
    '''
    # tokenzize the input
    tokenized_input = tokenizer.encode_plus(input, return_tensors='pt').to(device)
    
    # generate the response
    outputs = model.generate(
        input_ids=tokenized_input['input_ids'], 
        attention_mask=tokenized_input['attention_mask'],
        pad_token_id=tokenizer.pad_token_id,
        do_sample=False,
        repetition_penalty=1.3,
        max_length=500
    )
    
    # decode the response
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

if __name__ == '__main__':
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    model = AutoModelForCausalLM.from_pretrained("benchang1110/Taiwan-tinyllama-v1.0-base",device_map=device,torch_dtype=torch.bfloat16)
    tokenizer = AutoTokenizer.from_pretrained("benchang1110/Taiwan-tinyllama-v1.0-base")
    while(True):
        text = input("input a simple prompt:")
        print('System:', generate_response(text))

Using bfloat16, the VRAM required is around 3GB!!!