ChatGLM3-6B MLU370-M8运行手册来咯！！_chatglm 手册

---

前言

提示：不废话直接干货

---

一、智算中心平台环境选择

驱动：建议MLU370 -5.10.22

镜像：

二、镜像内部环境准备

1.transformers

git clone -b v4.33.0 https://githubfast.com/huggingface/transformers.git
python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i transformers/
pip install -e ./transformers_mlu/
1
2
3

2.accelerate

git clone -b v0.22.0 https://githubfast.com/huggingface/accelerate.git
python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i accelerate/
pip install -e ./accelerate_mlu/
1
2
3

3.deepspeed

wget https://sdk.cambricon.com/static/Basis/MLU370_X86_ubuntu20.04/deepspeed_mlu-0.9.0-py3-none-any.whl
python /torch/src/catch/tools/torch_gpu2mlu/torch_gpu2mlu.py -i accelerate/
pip install deepspeed_mlu-0.9.0-py3-none-any.whl
1
2
3

常规安装

pip install modelscope sentencepiece
1

三、模型下载

from modelscope import snapshot_download
model_dir = snapshot_download("ZhipuAI/chatglm3-6b", revision = "v1.0.0")
1
2

提示：下载好的模型记得mv到存储卷中，我们需要修改部分源码

四、模型文件修改

modeling_chatglm.py

32-36注释

279行

++attention_mask = attention_mask.cpu()
attention_mask.tril_()
++attention_mask = attention_mask.mlu()
1
2
3

五、直接运行

提示：模型改成自己得

import os
import platform
from transformers import AutoTokenizer, AutoModel
MODEL_PATH = os.environ.get('MODEL_PATH', '/workspace/volume/gpt/zhouguojun/GLM3/chatglm3-6b')
TOKENIZER_PATH = os.environ.get("TOKENIZER_PATH", MODEL_PATH)

tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_PATH, trust_remote_code=True)
model = AutoModel.from_pretrained(MODEL_PATH, trust_remote_code=True, device_map="auto").eval()

os_name = platform.system()
clear_command = 'cls' if os_name == 'Windows' else 'clear'
stop_stream = False

welcome_prompt = "欢迎使用 ChatGLM3-6B 模型，输入内容即可进行对话，clear 清空对话历史，stop 终止程序"


def build_prompt(history):
    prompt = welcome_prompt
    for query, response in history:
        prompt += f"\n\n用户：{query}"
        prompt += f"\n\nChatGLM3-6B：{response}"
    return prompt


def main():
    past_key_values, history = None, []
    global stop_stream
    print(welcome_prompt)
    while True:
        query = input("\n用户：")
        if query.strip() == "stop":
            break
        if query.strip() == "clear":
            past_key_values, history = None, []
            os.system(clear_command)
            print(welcome_prompt)
            continue
        print("\nChatGLM：", end="")
        current_length = 0
        for response, history, past_key_values in model.stream_chat(tokenizer, query, history=history, top_p=1,
                                                                    temperature=0.01,
                                                                    past_key_values=past_key_values,
                                                                    return_past_key_values=True):
            if stop_stream:
                stop_stream = False
                break
            else:
                print(response[current_length:], end="", flush=True)
                current_length = len(response)
        print("")


if __name__ == "__main__":
    main()
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54

cnmon可查看显存，下期见拜拜！