XVERSE

# 升级pip
python -m pip install --upgrade pip
# 更换 pypi 源加速库的安装
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple

pip install -r requirement.txt

import torch
from modelscope import snapshot_download, AutoModel, AutoTokenizer
import os
model_dir = snapshot_download('xverse/XVERSE-7B-Chat', cache_dir='/root/autodl-tmp', revision='master')

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers import GenerationConfig


# 加载预训练的分词器和模型
model_path = "xverse/XVERSE-7B-Chat"
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16, trust_remote_code=True).cuda()
model.generation_config = GenerationConfig.from_pretrained(model_path)

# 使用 INT8、INT4 进行量化推理 
# model = model.quantize(8).cuda()
model = model.quantize(4).cuda()

model = model.eval()

print("=============Welcome to XVERSE chatbot, type 'exit' to exit.=============")


# 设置多轮对话
while True:
    user_input = input("\n帅哥美女请输入: ")
    if user_input.lower() == "exit":
        break
    # 创建消息
    history = [{"role": "user", "content": user_input}]
    response = model.chat(tokenizer, history)
    print("\nXVERSE-7B-Chat: {}".format(response))

    # 添加回答到历史
    history.append({"role": "assistant", "content": response})

# 升级pip
python -m pip install --upgrade pip
# 更换 pypi 源加速库的安装
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple

pip install -r requirement.txt

import torch
from modelscope import snapshot_download, AutoModel, AutoTokenizer
import os
model_dir = snapshot_download('xverse/XVERSE-7B-Chat', cache_dir='/root/autodl-tmp', revision='master')

from fastapi import FastAPI, Request
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
import uvicorn
import json
import datetime
import torch

# 设置设备参数
DEVICE = "cuda"  # 使用CUDA
DEVICE_ID = "0"  # CUDA设备ID，如果未设置则为空
CUDA_DEVICE = f"{DEVICE}:{DEVICE_ID}" if DEVICE_ID else DEVICE  # 组合CUDA设备信息

# 清理GPU内存函数
def torch_gc():
    if torch.cuda.is_available():  # 检查是否可用CUDA
        with torch.cuda.device(CUDA_DEVICE):  # 指定CUDA设备
            torch.cuda.empty_cache()  # 清空CUDA缓存
            torch.cuda.ipc_collect()  # 收集CUDA内存碎片

# 创建FastAPI应用
app = FastAPI()

# 处理POST请求的端点
@app.post("/")
async def create_item(request: Request):
    global model, tokenizer  # 声明全局变量以便在函数内部使用模型和分词器
    json_post_raw = await request.json()  # 获取POST请求的JSON数据
    json_post = json.dumps(json_post_raw)  # 将JSON数据转换为字符串
    json_post_list = json.loads(json_post)  # 将字符串转换为Python对象
    prompt = json_post_list.get('prompt')  # 获取请求中的提示

    # 构建消息
    history = [{"role": "user", "content": prompt}]
    
    response = model.chat(tokenizer, history)
    
    now = datetime.datetime.now()  # 获取当前时间
    time = now.strftime("%Y-%m-%d %H:%M:%S")  # 格式化时间为字符串
    # 构建响应JSON
    answer = {
        "response": response,
        "status": 200,
        "time": time
    }
    # 构建日志信息
    log = "[" + time + "] " + '", prompt:"' + prompt + '", response:"' + repr(response) + '"'
    print(log)  # 打印日志
    torch_gc()  # 执行GPU内存清理
    return answer  # 返回响应

# 主函数入口
if __name__ == '__main__':
    # 加载预训练的分词器和模型
    model_path = "xverse/XVERSE-7B-Chat"
    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16, trust_remote_code=True).cuda()
    model.generation_config = GenerationConfig.from_pretrained(model_path)
    model = model.eval()

    # 启动FastAPI应用
    # 用6006端口可以将autodl的端口映射到本地，从而在本地使用api
    uvicorn.run(app, host='0.0.0.0', port=6006, workers=1)  # 在指定端口和主机上启动应用

cd /root/autodl-tmp
python api.py

curl -X POST "http://127.0.0.1:6006" \
     -H 'Content-Type: application/json' \
     -d '{"prompt": "你好，你是谁？"}'

import requests
import json

def get_completion(prompt):
    headers = {'Content-Type': 'application/json'}
    data = {"prompt": prompt}
    response = requests.post(url='http://127.0.0.1:6006', headers=headers, data=json.dumps(data))
    return response.json()['response']

if __name__ == '__main__':
    print(get_completion('你好，你是谁？'))

{"response":"我是元象公司开发的一个人工智能语言模型，名为XChat。我是基于Transformer架构训练出来的。我可以理解和生成自然语言，协助回答问题，提供信息，参与对话，创作文本等。请注意，我并非具有自我意识或个人情感，我只是一个能理解和生成文本的工具。我的知识截止于 2023 年 7 月，所以有关更近期事件和发展的信息可能不会包括在内。如果有任何问题，我会尽力提供帮助！","status":200,"time":"2024-04-18 12:07:16"}

# 升级pip
python -m pip install --upgrade pip
# 更换 pypi 源加速库的安装
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple

pip install -r requirement.txt

import torch
from modelscope import snapshot_download, AutoModel, AutoTokenizer
import os
model_dir = snapshot_download('xverse/XVERSE-7B-Chat', cache_dir='/root/autodl-tmp', revision='master')

from langchain.llms.base import LLM
from typing import Any, List, Optional
from langchain.callbacks.manager import CallbackManagerForLLMRun
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig, LlamaTokenizerFast
import torch

class XVERSE_LLM(LLM):
    # 基于本地 XVERSE-7B-Chat 自定义 LLM 类
    tokenizer: AutoTokenizer = None
    model: AutoModelForCausalLM = None
        
    def __init__(self, mode_name_or_path :str):

        super().__init__()
        print("正在从本地加载模型...")
        self.tokenizer = AutoTokenizer.from_pretrained(mode_name_or_path, trust_remote_code=True)
        self.model = AutoModelForCausalLM.from_pretrained(mode_name_or_path, torch_dtype=torch.float16, trust_remote_code=True).cuda()
        self.model = self.model.eval()
        print("完成本地模型的加载")
        
    def _call(self, prompt : str, stop: Optional[List[str]] = None,
                run_manager: Optional[CallbackManagerForLLMRun] = None,
                **kwargs: Any):

        # 构建消息
        history = [{"role": "user", "content": prompt}]

        response = self.model.chat(self.tokenizer, history)
        
        return response
    @property
    def _llm_type(self) -> str:
        return "XVERSE_LLM"

from LLM import XVERSE_LLM

llm = XVERSE_LLM(mode_name_or_path = "/root/autodl-tmp/xverse/XVERSE-7B-Chat")
llm("你好呀，你能帮助我干什么啊？")

'当然可以，我可以帮助您做很多事情。例如：\n\n1. 回答问题：无论是关于天气、新闻、历史、科学、数学、艺术等各种领域的问题，我都会尽力提供准确的信息。\n2. 设置提醒和闹钟：告诉我你需要提醒或设定的时间，我会在指定的时间通知你。\n3. 发送消息：如果你想给某人发送消息，只需告诉我收件人的名字或者联系方式，我就可以帮你完成。\n4. 查询路线：输入你的目的地，我可以为你规划出最佳的行驶路线。\n5. 播放音乐和电影：只要告诉我你想听哪首歌或者看哪部电影，我就可以为你播放。\n6. 学习新知识：如果你有任何学习需求，比如学习新的语言、技能等，我也可以提供相关的学习资源和建议。\n7. 进行简单的计算：如果你需要进行复杂的数学运算，我可能无法直接解答，但我可以尝试提供一些基本的计算方法。\n8. 提供新闻更新：我可以为你提供实时的新闻更新，让你随时了解世界大事。\n9. 其他许多任务：只要是你需要的，我都会尽力去做。'

# 升级pip
python -m pip install --upgrade pip
# 更换 pypi 源加速库的安装
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple

# 安装python依赖
pip install -r requirement.txt

from modelscope import snapshot_download
model_dir = snapshot_download("xverse/XVERSE-7B-Chat", cache_dir='/root/autodl-tmp', revision="master")

import argparse
import torch
import gradio as gr
import json
from datetime import datetime
from transformers import AutoModelForCausalLM, AutoTokenizer,GenerationConfig

tokenizer, model = None, None

def init_model(args):
    global tokenizer, model
    tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_path, truncation_side="left", padding_side="left")
    model = AutoModelForCausalLM.from_pretrained(args.model_path, trust_remote_code=True, torch_dtype=torch.bfloat16,
                                                 low_cpu_mem_usage=True, device_map='auto')
    model.generation_config = GenerationConfig.from_pretrained(args.model_path)
    model = model.eval()

def chat(message, history, request: gr.Request):
    global tokenizer, model
    history = history or []
    history.append({"role": "user", "content": message})

    # init
    history.append({"role": "assistant", "content": ""})
    utter_history = []
    for i in range(0, len(history), 2):
        utter_history.append([history[i]["content"], history[i+1]["content"]])

    # chat with stream
    for next_text in model.chat(tokenizer, history[:-1], stream=True):
        utter_history[-1][1] += next_text
        history[-1]["content"] += next_text
        if torch.backends.mps.is_available():
            torch.mps.empty_cache()
        yield utter_history, history

    # log
    current_time = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
    print(f'{current_time} request_ip:{request.client.host}\nquery: {message}\nhistory: {json.dumps(history, ensure_ascii=False)}\nanswer: {json.dumps(utter_history[-1][1], ensure_ascii=False)}')

# 增加配置，添加模型地址
def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument("--port", type=int, default=6006,
                       help="server port")
    parser.add_argument("--model_path", type=str, default="/root/autodl-tmp/xverse/XVERSE-7B-Chat",
                        help="model path")
    parser.add_argument("--tokenizer_path", type=str, default="/root/autodl-tmp/xverse/XVERSE-7B-Chat",
                        help="Path to the tokenizer.")
    args = parser.parse_args()
    return args

if __name__ == "__main__":
    args = get_args()
    # 初始化模型
    init_model(args)

    # 构建demo应用
    with gr.Blocks(theme=gr.themes.Soft()) as demo:
        gr.Markdown("""
                        # <center>💬 XVERSE-7B-Chat</center>
                        ## <center>🚀 A Gradio chatbot powered by Self-LLM</center>
                        ### <center>✨ 感兴趣的小伙伴可以去看我们的开源项目哦——[开源大模型食用指南 self-llm](https://github.com/datawhalechina/self-llm.git)，教你一杯奶茶跑通所有主流大模型😀。</center>
                    """)
        chatbot = gr.Chatbot(label="Chat history", height=500)
        state = gr.State([])

        with gr.Row():
            text_box = gr.Textbox(label="Message", show_label=False, placeholder="请输入你的消息并回车")

        with gr.Row():
            submit_btn = gr.Button(value="Send", variant="secondary")
            reset_btn = gr.Button(value="Reset")

        text_box.submit(fn=chat,
                        inputs=[text_box, state],
                        outputs=[chatbot, state],
                        api_name="chat")
        submit_btn.click(fn=chat,
                         inputs=[text_box, state],
                         outputs=[chatbot, state])

        # 用于清空text_box
        def clear_textbox():
            return gr.update(value="")
        text_box.submit(fn=clear_textbox, inputs=None, outputs=[text_box])
        submit_btn.click(fn=clear_textbox, inputs=None, outputs=[text_box])

        # 用于清空页面和重置state
        def reset():
            return None, []
        reset_btn.click(fn=reset, inputs=None, outputs=[chatbot, state])

    demo.launch(server_name="0.0.0.0", server_port=args.port)

python chatBot.py

cd code
pip install -r requirement.txt

{
  "instruction": "解释什么是人工智能。\n",
  "input": "",
  "output": "人工智能是一种利用计算机程序和算法创造出类似人类智能的技术，可以让计算机在解决问题、学习、推理和自然语言处理等方面表现出类似人类的能力。"
}

{
    "inputs": "Human:解释什么是人工智能。\n Assistant:", 
    "targets": "人工智能是一种利用计算机程序和算法创造出类似人类智能的技术，可以让计算机在解决问题、学习、推理和自然语言处理等方面表现出类似人类的能力。"}

def process_func(example):
        MAX_LENGTH = 384
        input_ids = []
        labels = []

        instruction = tokenizer(text=f"Human:现在你要扮演皇帝身边的女人--甄嬛\n\n {example['instruction']}{example['input']}Assistant:", add_special_tokens=False)
        response = tokenizer(text=f"{example['output']}", add_special_tokens=False)
        input_ids = [tokenizer.bos_token_id] + instruction["input_ids"] + response["input_ids"] + [tokenizer.eos_token_id]
        labels = [tokenizer.bos_token_id] + [-100] * len(instruction["input_ids"]) + response["input_ids"] + [tokenizer.eos_token_id]
        if len(input_ids) > MAX_LENGTH:
            input_ids = input_ids[:MAX_LENGTH]
            labels = labels[:MAX_LENGTH]

        return {
            "input_ids": input_ids,
            "labels": labels
        }

'<|startoftext|>Human:现在你要扮演皇帝身边的女人--甄嬛\n\n 这个温太医啊，也是古怪，谁不知太医不得皇命不能为皇族以外的人请脉诊病，他倒好，十天半月便往咱们府里跑。Assistant:你们俩话太多了，我该和温太医要一剂药，好好治治你们。<|endoftext|>'

import torch

model = AutoModelForCausalLM.from_pretrained('xverse/XVERSE-7B-Chat', trust_remote_code=True, torch_dtype=torch.half, device_map="auto")
model.generation_config = GenerationConfig.from_pretrained('xverse/XVERSE-7B-Chat')

config = LoraConfig(
    task_type=TaskType.CAUSAL_LM, 
    target_modules=["c_attn", "c_proj", "w1", "w2"],
    inference_mode=False, # 训练模式
    r=8, # Lora 秩
    lora_alpha=32, # Lora alaph，具体作用参见 Lora 原理
    lora_dropout=0.1# Dropout 比例
)

args = TrainingArguments(
    output_dir="./output/BlueLM",
    per_device_train_batch_size=8,
    gradient_accumulation_steps=2,
    logging_steps=10,
    num_train_epochs=3,
    gradient_checkpointing=True,
    save_steps=100,
    learning_rate=1e-4,
    save_on_each_node=True
)

trainer = Trainer(
    model=model,
    args=args,
    train_dataset=tokenized_id,
    data_collator=DataCollatorForSeq2Seq(tokenizer=tokenizer, padding=True),
)
trainer.train()

model.eval()
text = "小姐，别的秀女都在求中选，唯有咱们小姐想被撂牌子，菩萨一定记得真真儿的——"
inputs = tokenizer(f"Human:{text} Assistant:", return_tensors="pt", return_token_type_ids=False, )
outputs = model.generate(**inputs.to(model.device), max_new_tokens=100)
result = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(result)

# 因为涉及到访问github因此最好打开autodl的学术镜像加速
source /etc/network_turbo
# 升级pip
python -m pip install --upgrade pip
# 更换 pypi 源加速库的安装
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
# 从transformers的github仓库中安装包含XVERSE-MoE的新版本
# 如果安装不上可以使用 pip install git+https://github.moeyy.xyz/https://github.com/huggingface/transformers
pip install git+https://github.com/huggingface/transformers
# 安装需要的python包
pip install modelscope sentencepiece accelerate fastapi uvicorn requests streamlit transformers_stream_generator
# 安装flash-attention
# 这个也是不行使用 pip install https://github.moeyy.xyz/https://github.com/Dao-AILab/flash-attention/releases/download/v2.4.2/flash_attn-2.4.2+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install https://github.com/Dao-AILab/flash-attention/releases/download/v2.4.2/flash_attn-2.4.2+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

import torch
from modelscope import snapshot_download, AutoModel, AutoTokenizer
import os
model_dir = snapshot_download('xverse/XVERSE-MoE-A4.2B', cache_dir='/root/autodl-tmp', revision='master')

import torch  # 导入torch库，用于深度学习相关操作
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig  # 三个类分别用于加载分词器、加载因果语言模型和加载生成配置

# 将模型路径设置为刚刚下载的模型路径
model_name = "/root/autodl-tmp/xverse/XVERSE-MoE-A4.2B"

# 加载语言模型，设置数据类型为bfloat16即混合精度格式以优化性能并减少显存使用，将推理设备设置为`auto`自动选择最佳的设备进行推理，如果没有可用的GPU，它可能会回退到CPU
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")

# 加载分词器
tokenizer = AutoTokenizer.from_pretrained(model_name)

# 定义input字符串
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": prompt}
]
# 使用分词器的apply_chat_template方法来处理messages，转换格式
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True # 在消息前添加生成提示
)
# 将text变量中的文本转换为模型输入的格式，指定返回的张量为PyTorch张量（"pt"）
model_inputs = tokenizer([text], return_tensors="pt").to(device)
# 使用模型的generate方法来生成文本
generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=512
)
# 从生成的ID中提取出除了原始输入之外的新生成的token
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
# 使用分词器的batch_decode方法将生成的token ID转换回文本
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
# 显示生成的回答
print(response)