File size: 12,912 Bytes
a5c122b
 
7d5aaa5
 
 
 
 
 
 
 
 
 
 
27d01c0
dcaa7a1
27d01c0
 
 
 
 
 
889b719
 
9bd8511
 
 
27d01c0
2bf30d8
 
27d01c0
ac2c8ca
f04d975
 
 
ac2c8ca
 
 
 
 
 
e39c511
 
0666fec
61b4ea6
3eef2d5
 
 
 
 
 
 
 
 
 
 
61b4ea6
dcaa7a1
0666fec
61b4ea6
 
 
 
9bd8511
61b4ea6
 
 
 
 
 
 
 
 
 
 
50dfccc
 
 
 
61b4ea6
 
5e8eb62
 
 
 
 
 
 
61b4ea6
 
3aa446c
 
e6cf553
dcaa7a1
 
 
 
 
 
cd6a1fd
61b4ea6
5e8eb62
dcaa7a1
61b4ea6
 
 
0666fec
f04d975
3eef2d5
 
 
 
 
 
 
f04d975
9bd8511
3725122
 
 
 
9bd8511
3725122
9bd8511
3725122
 
27d01c0
dcaa7a1
 
06f8094
dcaa7a1
 
27d01c0
5549e58
 
 
 
 
 
 
 
 
 
 
 
fa22df9
27d01c0
248e18e
 
 
fa22df9
9bd8511
248e18e
 
 
 
 
0666fec
248e18e
27d01c0
fa22df9
 
 
27d01c0
 
 
 
fa22df9
 
f238a34
fa22df9
 
27d01c0
 
 
fa22df9
 
27d01c0
fa22df9
1fa9a79
 
fa22df9
 
 
1fa9a79
0666fec
1fa9a79
 
 
0666fec
ac2c8ca
 
 
dcaa7a1
5b9de09
e341596
dcaa7a1
5b9de09
 
d26fa46
 
5b9de09
 
0666fec
ac2c8ca
 
0666fec
f04d975
3eef2d5
f04d975
9bd8511
3725122
 
9bd8511
 
fa22df9
 
9bd8511
fa22df9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9bd8511
fa22df9
0666fec
 
fa22df9
 
 
 
 
fc33168
270889a
fc33168
 
fa22df9
 
ac2c8ca
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
# 借鉴了 https://github.com/GaiZhenbiao/ChuanhuChatGPT 项目

"""
    该文件中主要包含三个函数

    不具备多线程能力的函数:
    1. predict: 正常对话时使用,具备完备的交互功能,不可多线程

    具备多线程调用能力的函数
    2. predict_no_ui:高级实验性功能模块调用,不会实时显示在界面上,参数简单,可以多线程并行,方便实现复杂的功能逻辑
    3. predict_no_ui_long_connection:在实验过程中发现调用predict_no_ui处理长文档时,和openai的连接容易断掉,这个函数用stream的方式解决这个问题,同样支持多线程
"""

import json
import time
import gradio as gr
import logging
import traceback
import requests
import importlib

# config_private.py放自己的秘密如API和代理网址
# 读取时首先看是否存在私密的config_private配置文件(不受git管控),如果有,则覆盖原config文件
from toolbox import get_conf, update_ui, is_any_api_key, select_api_key
proxies, API_KEY, TIMEOUT_SECONDS, MAX_RETRY = \
    get_conf('proxies', 'API_KEY', 'TIMEOUT_SECONDS', 'MAX_RETRY')

timeout_bot_msg = '[Local Message] Request timeout. Network error. Please check proxy settings in config.py.' + \
                  '网络错误,检查代理服务器是否可用,以及代理设置的格式是否正确,格式须是[协议]://[地址]:[端口],缺一不可。'

def get_full_error(chunk, stream_response):
    """
        获取完整的从Openai返回的报错
    """
    while True:
        try:
            chunk += next(stream_response)
        except:
            break
    return chunk


def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=None, console_slience=False):
    """
    发送至chatGPT,等待回复,一次性完成,不显示中间过程。但内部用stream的方法避免中途网线被掐。
    inputs:
        是本次问询的输入
    sys_prompt:
        系统静默prompt
    llm_kwargs:
        chatGPT的内部调优参数
    history:
        是之前的对话列表
    observe_window = None:
        用于负责跨越线程传递已经输出的部分,大部分时候仅仅为了fancy的视觉效果,留空即可。observe_window[0]:观测窗。observe_window[1]:看门狗
    """
    watch_dog_patience = 5 # 看门狗的耐心, 设置5秒即可
    headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt=sys_prompt, stream=True)
    retry = 0
    while True:
        try:
            # make a POST request to the API endpoint, stream=False
            response = requests.post(llm_kwargs['endpoint'], headers=headers, proxies=proxies,
                                    json=payload, stream=True, timeout=TIMEOUT_SECONDS); break
        except requests.exceptions.ReadTimeout as e:
            retry += 1
            traceback.print_exc()
            if retry > MAX_RETRY: raise TimeoutError
            if MAX_RETRY!=0: print(f'请求超时,正在重试 ({retry}/{MAX_RETRY}) ……')

    stream_response =  response.iter_lines()
    result = ''
    while True:
        try: chunk = next(stream_response).decode()
        except StopIteration: 
            break
        except requests.exceptions.ConnectionError:
            chunk = next(stream_response).decode() # 失败了,重试一次?再失败就没办法了。
        if len(chunk)==0: continue
        if not chunk.startswith('data:'): 
            error_msg = get_full_error(chunk.encode('utf8'), stream_response).decode()
            if "reduce the length" in error_msg:
                raise ConnectionAbortedError("OpenAI拒绝了请求:" + error_msg)
            else:
                raise RuntimeError("OpenAI拒绝了请求:" + error_msg)
        json_data = json.loads(chunk.lstrip('data:'))['choices'][0]
        delta = json_data["delta"]
        if len(delta) == 0: break
        if "role" in delta: continue
        if "content" in delta: 
            result += delta["content"]
            if not console_slience: print(delta["content"], end='')
            if observe_window is not None: 
                # 观测窗,把已经获取的数据显示出去
                if len(observe_window) >= 1: observe_window[0] += delta["content"]
                # 看门狗,如果超过期限没有喂狗,则终止
                if len(observe_window) >= 2:  
                    if (time.time()-observe_window[1]) > watch_dog_patience:
                        raise RuntimeError("用户取消了程序。")
        else: raise RuntimeError("意外Json结构:"+delta)
    if json_data['finish_reason'] == 'length':
        raise ConnectionAbortedError("正常结束,但显示Token不足,导致输出不完整,请削减单次输入的文本量。")
    return result


def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
    """
    发送至chatGPT,流式获取输出。
    用于基础的对话功能。
    inputs 是本次问询的输入
    top_p, temperature是chatGPT的内部调优参数
    history 是之前的对话列表(注意无论是inputs还是history,内容太长了都会触发token数量溢出的错误)
    chatbot 为WebUI中显示的对话列表,修改它,然后yeild出去,可以直接修改对话界面内容
    additional_fn代表点击的哪个按钮,按钮见functional.py
    """
    if is_any_api_key(inputs):
        chatbot._cookies['api_key'] = inputs
        chatbot.append(("输入已识别为openai的api_key", "api_key已导入"))
        yield from update_ui(chatbot=chatbot, history=history, msg="api_key已导入") # 刷新界面
        return
    elif not is_any_api_key(chatbot._cookies['api_key']):
        chatbot.append((inputs, "缺少api_key。\n\n1. 临时解决方案:直接在输入区键入api_key,然后回车提交。\n\n2. 长效解决方案:在config.py中配置。"))
        yield from update_ui(chatbot=chatbot, history=history, msg="缺少api_key") # 刷新界面
        return

    if additional_fn is not None:
        import core_functional
        importlib.reload(core_functional)    # 热更新prompt
        core_functional = core_functional.get_core_functions()
        if "PreProcess" in core_functional[additional_fn]: inputs = core_functional[additional_fn]["PreProcess"](inputs)  # 获取预处理函数(如果有的话)
        inputs = core_functional[additional_fn]["Prefix"] + inputs + core_functional[additional_fn]["Suffix"]

    raw_input = inputs
    logging.info(f'[raw_input] {raw_input}')
    chatbot.append((inputs, ""))
    yield from update_ui(chatbot=chatbot, history=history, msg="等待响应") # 刷新界面

    try:
        headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt, stream)
    except RuntimeError as e:
        chatbot[-1] = (inputs, f"您提供的api-key不满足要求,不包含任何可用于{llm_kwargs['llm_model']}的api-key。")
        yield from update_ui(chatbot=chatbot, history=history, msg="api-key不满足要求") # 刷新界面
        return
        
    history.append(inputs); history.append(" ")

    retry = 0
    while True:
        try:
            # make a POST request to the API endpoint, stream=True
            response = requests.post(llm_kwargs['endpoint'], headers=headers, proxies=proxies,
                                    json=payload, stream=True, timeout=TIMEOUT_SECONDS);break
        except:
            retry += 1
            chatbot[-1] = ((chatbot[-1][0], timeout_bot_msg))
            retry_msg = f",正在重试 ({retry}/{MAX_RETRY}) ……" if MAX_RETRY > 0 else ""
            yield from update_ui(chatbot=chatbot, history=history, msg="请求超时"+retry_msg) # 刷新界面
            if retry > MAX_RETRY: raise TimeoutError

    gpt_replying_buffer = ""
    
    is_head_of_the_stream = True
    if stream:
        stream_response =  response.iter_lines()
        while True:
            chunk = next(stream_response)
            # print(chunk.decode()[6:])
            if is_head_of_the_stream:
                # 数据流的第一帧不携带content
                is_head_of_the_stream = False; continue
            
            if chunk:
                try:
                    if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
                        # 判定为数据流的结束,gpt_replying_buffer也写完了
                        logging.info(f'[response] {gpt_replying_buffer}')
                        break
                    # 处理数据流的主体
                    chunkjson = json.loads(chunk.decode()[6:])
                    status_text = f"finish_reason: {chunkjson['choices'][0]['finish_reason']}"
                    # 如果这里抛出异常,一般是文本过长,详情见get_full_error的输出
                    gpt_replying_buffer = gpt_replying_buffer + json.loads(chunk.decode()[6:])['choices'][0]["delta"]["content"]
                    history[-1] = gpt_replying_buffer
                    chatbot[-1] = (history[-2], history[-1])
                    yield from update_ui(chatbot=chatbot, history=history, msg=status_text) # 刷新界面

                except Exception as e:
                    traceback.print_exc()
                    yield from update_ui(chatbot=chatbot, history=history, msg="Json解析不合常规") # 刷新界面
                    chunk = get_full_error(chunk, stream_response)
                    error_msg = chunk.decode()
                    if "reduce the length" in error_msg:
                        chatbot[-1] = (chatbot[-1][0], "[Local Message] Reduce the length. 本次输入过长,或历史数据过长. 历史缓存数据现已释放,您可以请再次尝试.")
                        history = []    # 清除历史
                    elif "Incorrect API key" in error_msg:
                        chatbot[-1] = (chatbot[-1][0], "[Local Message] Incorrect API key. OpenAI以提供了不正确的API_KEY为由,拒绝服务.")
                    elif "exceeded your current quota" in error_msg:
                        chatbot[-1] = (chatbot[-1][0], "[Local Message] You exceeded your current quota. OpenAI以账户额度不足为由,拒绝服务.")
                    else:
                        from toolbox import regular_txt_to_markdown
                        tb_str = '```\n' + traceback.format_exc() + '```'
                        chatbot[-1] = (chatbot[-1][0], f"[Local Message] 异常 \n\n{tb_str} \n\n{regular_txt_to_markdown(chunk.decode()[4:])}")
                    yield from update_ui(chatbot=chatbot, history=history, msg="Json异常" + error_msg) # 刷新界面
                    return

def generate_payload(inputs, llm_kwargs, history, system_prompt, stream):
    """
    整合所有信息,选择LLM模型,生成http请求,为发送请求做准备
    """
    if not is_any_api_key(llm_kwargs['api_key']):
        raise AssertionError("你提供了错误的API_KEY。\n\n1. 临时解决方案:直接在输入区键入api_key,然后回车提交。\n\n2. 长效解决方案:在config.py中配置。")

    api_key = select_api_key(llm_kwargs['api_key'], llm_kwargs['llm_model'])

    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {api_key}"
    }

    conversation_cnt = len(history) // 2

    messages = [{"role": "system", "content": system_prompt}]
    if conversation_cnt:
        for index in range(0, 2*conversation_cnt, 2):
            what_i_have_asked = {}
            what_i_have_asked["role"] = "user"
            what_i_have_asked["content"] = history[index]
            what_gpt_answer = {}
            what_gpt_answer["role"] = "assistant"
            what_gpt_answer["content"] = history[index+1]
            if what_i_have_asked["content"] != "":
                if what_gpt_answer["content"] == "": continue
                if what_gpt_answer["content"] == timeout_bot_msg: continue
                messages.append(what_i_have_asked)
                messages.append(what_gpt_answer)
            else:
                messages[-1]['content'] = what_gpt_answer['content']

    what_i_ask_now = {}
    what_i_ask_now["role"] = "user"
    what_i_ask_now["content"] = inputs
    messages.append(what_i_ask_now)

    payload = {
        "model": llm_kwargs['llm_model'].strip('api2d-'),
        "messages": messages, 
        "temperature": llm_kwargs['temperature'],  # 1.0,
        "top_p": llm_kwargs['top_p'],  # 1.0,
        "n": 1,
        "stream": stream,
        "presence_penalty": 0,
        "frequency_penalty": 0,
    }
    try:
        print(f" {llm_kwargs['llm_model']} : {conversation_cnt} : {inputs[:100]} ..........")
    except:
        print('输入中可能存在乱码。')
    return headers,payload