app.py

import gradio as gr
import openai
from transformers import AutoTokenizer
import os


openai.api_key = os.environ.get("OPENAI_API_KEY")

def openai_summarize(prompt):
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "system", "content": "You are a helpful assistant that summarizes the emails between 2 parties in chronological order."}, {"role": "user", "content": f"summarize, maintaining the correct time sequence and make sure to include all the dates and at the end of the summary, add a Chinese sentence to sum up both parties' intents: {prompt}"}],
        temperature = 0.1,
        max_tokens=2000,
    )

    return response['choices'][0]['message']['content'].strip()

def generate_reply(prompt, context, system_role):
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "system", "content": f"You are {system_role} that generates English email replies based on given prompts and context."},
            {"role": "user", "content": context},
            {"role": "user", "content": prompt},
        ],
        temperature = 0.7,
        max_tokens=1900,
    )

    return response['choices'][0]['message']['content'].strip()


def count_tokens(tokenizer, text):
    tokens = tokenizer.encode(text)
    return len(tokens)

def summarize_chunks(text):
    tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-2.7B", model_max_length=4096)
    tokens = tokenizer.encode(text)
    
    chunk_size = 2000
    chunks = [tokens[i:i + chunk_size] for i in range(0, len(tokens), chunk_size)]

    summaries = []
    for chunk in chunks:
        text_chunk = tokenizer.decode(chunk)
        summary = openai_summarize(text_chunk)
        summaries.append(summary)

    # Reverse the list of summaries
    summaries.reverse()

    return "\n".join(summaries)

'''
def gradio_interface(text, prompt, system_role):
    summary = summarize_chunks(text)
    reply = generate_reply(prompt, summary, system_role)
    return summary, reply
'''

# 历史邮件内容超过2000才压缩，2000以内则直接作为语料
def gradio_interface(text, prompt, system_role):
    tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-2.7B", model_max_length=4096)
    token_count = count_tokens(tokenizer, text)

    if token_count > 2000:
        summary = summarize_chunks(text)
    else:
        summary = text

    reply = generate_reply(prompt, summary, system_role)
    return summary, reply    

inputs = [
    gr.inputs.Textbox(lines=5, label="所有过往邮件"),
    gr.inputs.Textbox(label="按以下提示生成回复内容"),
    gr.inputs.Radio(["helpful assistant", "customer service at maxfull hair company", "customer service at mhot hair company", "social marketing specialist at maxfull hair company","social marketing specialist at mhot hair company"], label="设置AI角色"),
]

outputs = [
    gr.outputs.Textbox(label="总结过往邮件(短邮件不总结)"),
    gr.outputs.Textbox(label="邮件回复"),
]

iface = gr.Interface(fn=gradio_interface, inputs=inputs, outputs=outputs, title="AI回复邮件")
iface.launch()