File size: 7,736 Bytes
f81fd9e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176

from transformers import AutoModel, AutoTokenizer
import time
import threading
import importlib
from toolbox import update_ui, get_conf
from multiprocessing import Process, Pipe

load_message = "jittorllms尚未加载,加载需要一段时间。注意,请避免混用多种jittor模型,否则可能导致显存溢出而造成卡顿,取决于`config.py`的配置,jittorllms消耗大量的内存(CPU)或显存(GPU),也许会导致低配计算机卡死 ……"

#################################################################################
class GetGLMHandle(Process):
    def __init__(self):
        super().__init__(daemon=True)
        self.parent, self.child = Pipe()
        self.jittorllms_model = None
        self.info = ""
        self.local_history = []
        self.success = True
        self.check_dependency()
        self.start()
        self.threadLock = threading.Lock()
        
    def check_dependency(self):
        try:
            import pandas
            self.info = "依赖检测通过"
            self.success = True
        except:
            from toolbox import trimmed_format_exc
            self.info = r"缺少jittorllms的依赖,如果要使用jittorllms,除了基础的pip依赖以外,您还需要运行`pip install -r request_llm/requirements_jittorllms.txt -i https://pypi.jittor.org/simple -I`"+\
                        r"和`git clone https://gitlink.org.cn/jittor/JittorLLMs.git --depth 1 request_llm/jittorllms`两个指令来安装jittorllms的依赖(在项目根目录运行这两个指令)。" +\
                        r"警告:安装jittorllms依赖后将完全破坏现有的pytorch环境,建议使用docker环境!" + trimmed_format_exc()
            self.success = False

    def ready(self):
        return self.jittorllms_model is not None

    def run(self):
        # 子进程执行
        # 第一次运行,加载参数
        def validate_path():
            import os, sys
            dir_name = os.path.dirname(__file__)
            env = os.environ.get("PATH", "")
            os.environ["PATH"] = env.replace('/cuda/bin', '/x/bin')
            root_dir_assume = os.path.abspath(os.path.dirname(__file__) +  '/..')
            os.chdir(root_dir_assume + '/request_llm/jittorllms')
            sys.path.append(root_dir_assume + '/request_llm/jittorllms')
        validate_path() # validate path so you can run from base directory

        def load_model():
            import types
            try:
                if self.jittorllms_model is None:
                    device, = get_conf('LOCAL_MODEL_DEVICE')
                    from .jittorllms.models import get_model
                    # availabel_models = ["chatglm", "pangualpha", "llama", "chatrwkv"]
                    args_dict = {'model': 'llama'}
                    print('self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))')
                    self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))
                    print('done get model')
            except:
                self.child.send('[Local Message] Call jittorllms fail 不能正常加载jittorllms的参数。')
                raise RuntimeError("不能正常加载jittorllms的参数!")
        print('load_model')
        load_model()

        # 进入任务等待状态
        print('进入任务等待状态')
        while True:
            # 进入任务等待状态
            kwargs = self.child.recv()
            query = kwargs['query']
            history = kwargs['history']
            # 是否重置
            if len(self.local_history) > 0 and len(history)==0:
                print('触发重置')
                self.jittorllms_model.reset()
            self.local_history.append(query)

            print('收到消息,开始请求')
            try:
                for response in self.jittorllms_model.stream_chat(query, history):
                    print(response)
                    self.child.send(response)
            except:
                from toolbox import trimmed_format_exc
                print(trimmed_format_exc())
                self.child.send('[Local Message] Call jittorllms fail.')
            # 请求处理结束,开始下一个循环
            self.child.send('[Finish]')

    def stream_chat(self, **kwargs):
        # 主进程执行
        self.threadLock.acquire()
        self.parent.send(kwargs)
        while True:
            res = self.parent.recv()
            if res != '[Finish]':
                yield res
            else:
                break
        self.threadLock.release()
    
global llama_glm_handle
llama_glm_handle = None
#################################################################################
def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=[], console_slience=False):
    """
        多线程方法
        函数的说明请见 request_llm/bridge_all.py
    """
    global llama_glm_handle
    if llama_glm_handle is None:
        llama_glm_handle = GetGLMHandle()
        if len(observe_window) >= 1: observe_window[0] = load_message + "\n\n" + llama_glm_handle.info
        if not llama_glm_handle.success: 
            error = llama_glm_handle.info
            llama_glm_handle = None
            raise RuntimeError(error)

    # jittorllms 没有 sys_prompt 接口,因此把prompt加入 history
    history_feedin = []
    for i in range(len(history)//2):
        history_feedin.append([history[2*i], history[2*i+1]] )

    watch_dog_patience = 5 # 看门狗 (watchdog) 的耐心, 设置5秒即可
    response = ""
    for response in llama_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=sys_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
        print(response)
        if len(observe_window) >= 1:  observe_window[0] = response
        if len(observe_window) >= 2:  
            if (time.time()-observe_window[1]) > watch_dog_patience:
                raise RuntimeError("程序终止。")
    return response



def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
    """
        单线程方法
        函数的说明请见 request_llm/bridge_all.py
    """
    chatbot.append((inputs, ""))

    global llama_glm_handle
    if llama_glm_handle is None:
        llama_glm_handle = GetGLMHandle()
        chatbot[-1] = (inputs, load_message + "\n\n" + llama_glm_handle.info)
        yield from update_ui(chatbot=chatbot, history=[])
        if not llama_glm_handle.success: 
            llama_glm_handle = None
            return

    if additional_fn is not None:
        from core_functional import handle_core_functionality
        inputs, history = handle_core_functionality(additional_fn, inputs, history, chatbot)

    # 处理历史信息
    history_feedin = []
    for i in range(len(history)//2):
        history_feedin.append([history[2*i], history[2*i+1]] )

    # 开始接收jittorllms的回复
    response = "[Local Message]: 等待jittorllms响应中 ..."
    for response in llama_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=system_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
        chatbot[-1] = (inputs, response)
        yield from update_ui(chatbot=chatbot, history=history)

    # 总结输出
    if response == "[Local Message]: 等待jittorllms响应中 ...":
        response = "[Local Message]: jittorllms响应异常 ..."
    history.extend([inputs, response])
    yield from update_ui(chatbot=chatbot, history=history)