boto.py

import bottle
from bottle import route, run, template, static_file, request, response
import json
import logging
import llm

LOG_FORMAT = "%(asctime)s - %(levelname)s - %(message)s"
DATE_FORMAT = "%m/%d/%Y %H:%M:%S %p"
logging.basicConfig(filename='qa.log', level=logging.WARNING, format=LOG_FORMAT, datefmt=DATE_FORMAT)
app = bottle.app()

def get_reply(querys):
    prompts = []
    if len(querys) >= 5:
        querys = querys[-5:]
    for query in querys:
        if query["role"] == "bot":
            query["role"] = "system"
        prompts.append(llm.build_prompt(query["role"], query["content"]))
    text = llm.tokenizer.apply_chat_template(
        prompts,
        tokenize=False,
        add_generation_prompt=True
    )

    inputs = llm.tokenizer(text, return_tensors="pt", padding=True).to('cuda')

    from transformers import TextIteratorStreamer
    from threading import Thread

    streamer = TextIteratorStreamer(llm.tokenizer, skip_prompt=True)
    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=4000, temperature=0.6)
    thread = Thread(target=llm.model.generate, kwargs=generation_kwargs)
    thread.start()
    generated_text = ""
    for new_text in streamer:
        generated_text += new_text
    logging.warning(generated_text)
    return "**以下为思考过程**\n" + generated_text.replace("</think>", "**以下为回答内容**\n").replace("<｜end▁of▁sentence｜>","")

def analyze_document(content, question):
    # 构建分析提示
    system_prompt = {
        "role": "system",
        "content": f"""你是一个专业的文档分析助手。请基于以下文档内容回答用户的问题。
文档内容开始：
{content}
文档内容结束。

请仔细分析文档内容，并以专业、准确的方式回答用户问题。如果问题超出文档范围，请明确指出。
分析要求：
1. 仔细阅读文档内容，理解用户问题
2. 在文档中定位相关信息
3. 如果需要引用文档内容，请使用引号标注
4. 如果文档中没有相关信息，请明确说明"""
    }
    
    user_prompt = {
        "role": "user",
        "content": question
    }
    
    prompts = [system_prompt, user_prompt]
    
    text = llm.tokenizer.apply_chat_template(
        prompts,
        tokenize=False,
        add_generation_prompt=True
    )

    inputs = llm.tokenizer(text, return_tensors="pt", padding=True).to('cuda')

    from transformers import TextIteratorStreamer
    from threading import Thread

    streamer = TextIteratorStreamer(llm.tokenizer, skip_prompt=True)
    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=4000, temperature=0.6)
    thread = Thread(target=llm.model.generate, kwargs=generation_kwargs)
    thread.start()
    
    generated_text = ""
    for new_text in streamer:
        generated_text += new_text
    
    logging.warning(f"Document analysis - Question: {question}")
    logging.warning(f"Answer: {generated_text}")
    
    return "**以下为思考过程**\n" + generated_text.replace("</think>", "**以下为回答内容**\n").replace("<｜end▁of▁sentence｜>","")

def checkForQuestion(msgs):
    return {"code": 100, "animation": "no", "msg": get_reply(msgs)}

def checkMessage(msgs):
    question = checkForQuestion(msgs)
    if question:
        return question
    else:
        return {"code": 101, "animation": "inlove", "msg": "I didnt understand that, im a pretty dumb boto"}

@route('/', method='GET')
def index():
    return template("index.html")

@route("/chat/<username>", method=['OPTIONS', 'POST'])
def chat(username):
    user_messages = request.json['msg']["messages"]
    logging.warning('get new msg:')
    logging.warning(user_messages)
    return json.dumps(checkMessage(user_messages))

@route("/analyze", method=['OPTIONS', 'POST'])
def analyze():
    try:
        data = request.json
        doc_id = data.get('docId')
        question = data.get('question')
        content = data.get('content')
        
        if not all([doc_id, question, content]):
            return json.dumps({
                "code": 400,
                "error": "Missing required parameters"
            })
        
        answer = analyze_document(content, question)
        
        return json.dumps({
            "code": 200,
            "answer": answer
        })
    except Exception as e:
        logging.error(f"Analysis error: {str(e)}")
        return json.dumps({
            "code": 500,
            "error": "Internal server error"
        })

@route("/test", method='POST')
def test_chat():
    user_message = request.POST.get('msg')
    return json.dumps({"animation": "inlove", "msg": user_message})

@route('/<file_type:re:(js|css|images)>/<filename:re:.*\.(js|css|jpg|png|gif|ico)>', method='GET')
def serve_static(file_type, filename):
    root_dirs = {
        'js': 'js',
        'css': 'css',
        'images': 'images'
    }
    root = root_dirs.get(file_type)
    if root:
        return static_file(filename, root=root)
    return "File type not supported", 404

def main():
    app.run(host='0.0.0.0', port=8181, server='paste')

if __name__ == '__main__':
    main()