app.py

#!/usr/bin/env python
# coding: utf-8

# In[ ]:


import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("microsoft/GODEL-v1_1-large-seq2seq")
model = AutoModelForSeq2SeqLM.from_pretrained("microsoft/GODEL-v1_1-large-seq2seq")

def predict(input,knowledge, history=[]):
#     instruction="Instruction: given a dialog context and related knowledge, you need to answer the question based on the knowledge."
    instruction="Instruction: given a dialog context, you need to response empathically"
    knowledge = '[KNOWLEDGE]' + knowledge
    s = list(sum(history, ()))
    s.append(input)
    dialog = ' EOS ' .join(s)
    query = f"{instruction} [CONTEXT] {dialog} {knowledge}"
    top_p = 0.9
    min_length = 8
    max_length = 64
    new_user_input_ids = tokenizer.encode(f"{query}", return_tensors='pt')
    print(input,s)
    output = model.generate(new_user_input_ids, min_length=int(
        min_length), max_length=int(max_length), top_p=top_p, do_sample=True).tolist()
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    history.append((input, response))
    return history, history

gr.Interface(fn=predict,
             inputs=["text","text",'state'],
            
             outputs=["chatbot",'state']).launch()