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|>


def chat(message, history):
    history = history or []
    if history == []:
        context = '<|context|> <|user|> ' + message + ' <|endofcontext|> '
    else:
        context, _ = history[-1][0].split('<|endofcontext|>')
        context += ' <|system|> ' 
        context += history[-1][1].split('<|response|>')[1]
        context = context.replace('<|endofresponse|>', '')
        context += ' <|user|> ' + message + ' <|endofcontext|> '

    output = generate_response(context)
    _ , response = output.split('<|endofcontext|>')

    history.append((message, response))
    
    return history, history


import gradio as gr

chatbot = gr.Chatbot(color_map=("gray", "blue"))

iface = gr.Interface(chat,
                    ["text", "state"],
                    [chatbot, "state"],
                    allow_screenshot=False,
                    allow_flagging="never",
)



"""
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|>"]],
                     description="Passing in a task-oriented dialogue context generates a belief state, actions to take and a response based on those actions",
                     )
"""
iface.launch(debug=True)