from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = 'armandnlp/gpt2-TOD_finetuned_SGD'
tokenizer_TOD = AutoTokenizer.from_pretrained(model_name)
model_TOD = AutoModelForCausalLM.from_pretrained(model_name)


def generate_response(prompt):
    input_ids = tokenizer_TOD(prompt, return_tensors="pt").input_ids
    outputs = model_TOD.generate(input_ids, 
                                 do_sample=False, 
                                 max_length=1024, 
                                 eos_token_id=50262)
    return tokenizer_TOD.batch_decode(outputs)[0]

#<|context|> <|user|> I want to go to the restaurant.<|endofcontext|>




iface = gr.Interface(fn=generate_response,
                     inputs="text",
                     outputs="text",
                     title="gpt2-TOD",
                     examples=[["<|context|> <|user|> I'm super hungry ! I want to go to the restaurant.<|endofcontext|>"],
                                "<|context|> <|user|> I want to go to the restaurant.\
                                <|system|> What food would you like to eat ? <|user|> Italian sounds good. <|endofcontext|>"],
                     description="Passing in a task-oriented dialogue context generates a belief state, actions to take and a response based on those actions",
                     )

iface.launch()