JunchuanYu commited on
Commit
82a00d8
1 Parent(s): 19edaa0

Upload 5 files

Browse files
Files changed (5) hide show
  1. app.py +92 -0
  2. chat_func.py +300 -0
  3. presets.py +46 -0
  4. requirements.txt +7 -0
  5. utils.py +301 -0
app.py ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import os
3
+ import sys
4
+ import argparse
5
+ from utils import *
6
+ from presets import *
7
+ from chat_func import *
8
+ my_api_key = "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" # 在这里输入你的 API 密钥
9
+
10
+ if my_api_key == "":
11
+ my_api_key = os.environ.get('my_api_key')
12
+
13
+ if my_api_key == "empty":
14
+ print("Please give a api key!")
15
+ sys.exit(1)
16
+
17
+ gr.Chatbot.postprocess = postprocess
18
+
19
+ # css = """
20
+ # #col-container {max-width: 80%; margin-left: auto; margin-right: auto;}
21
+ # #chatbox {min-height: 150px}
22
+ # #header {text-align: center;font-size: 2.8em}
23
+ # #prompt_template_preview {padding: 1em; border-width: 1px; border-style: solid; border-color: #e0e0e0; border-radius: 4px;}
24
+ # #submit {text-align: center; background-color: #e0e0e0;}
25
+ # #label {font-size: 0.8em; padding: 0.5em; margin: 0;}
26
+ # """
27
+ with gr.Blocks() as demo:
28
+ history = gr.State([])
29
+ token_count = gr.State([])
30
+ promptTemplates = gr.State(load_template(get_template_names(plain=True)[0], mode=2))
31
+ user_api_key = gr.State(my_api_key)
32
+ TRUECOMSTANT = gr.State(True)
33
+ FALSECONSTANT = gr.State(False)
34
+ gr.Markdown(title)
35
+
36
+ with gr.Accordion("Build by [45度科研人](WeChat Public Accounts)", open=False):
37
+ gr.Markdown(description)
38
+
39
+ with gr.Row(scale=1).style(equal_height=True):
40
+ with gr.Column(scale=5):
41
+ with gr.Column():
42
+ chatbot = gr.Chatbot().style(color_map=("blue", "green"))
43
+ user_input = gr.Textbox(show_label=False, placeholder="Enter text and press submit", visible=True).style(container=False)
44
+ submitBtn = gr.Button("Submit", variant="primary").style(container=False)
45
+ emptyBtn = gr.Button("Restart Conversation")
46
+ status_display = gr.Markdown("")
47
+
48
+ with gr.Column():
49
+ with gr.Column(min_width=50):
50
+ with gr.Tab(label="ChatGPT"):
51
+ with gr.Column():
52
+ with gr.Row():
53
+ keyTxt = gr.Textbox(show_label=False, placeholder=f"You can input your own openAI API-key",value=hide_middle_chars(my_api_key),visible=True, type="password", label="API-Key")
54
+ # keyTxt = gr.Textbox(show_label=False, placeholder=f"You can input your own openAI API-key",value=my_api_key,visible=True, type="password", label="API-Key")
55
+ systemPromptTxt = gr.Textbox(show_label=True,placeholder=f"Set a custom insruction for the chatbot: You are a helpful assistant.",label="Custom prompt",value=initial_prompt,lines=10,).style(container=False)
56
+
57
+ with gr.Row():
58
+ templateSelectDropdown = gr.Dropdown(label="load from template",choices=load_template(get_template_names(plain=True)[0], mode=1),
59
+ multiselect=False,value=load_template(get_template_names(plain=True)[0], mode=1)[0],).style(container=False)
60
+
61
+ with gr.Tab(label="Settings"):
62
+ with gr.Column():
63
+ with gr.Row():
64
+ with gr.Column(scale=3):
65
+ saveFileName = gr.Textbox(show_label=True, placeholder=f"output file name...",label='Save conversation history', value="").style(container=False)
66
+ with gr.Column(scale=1):
67
+ exportMarkdownBtn = gr.Button("Save")
68
+ with gr.Row():
69
+ with gr.Column(scale=1):
70
+ downloadFile = gr.File(interactive=False)
71
+ gr.Markdown("""
72
+ <div align=center>you can follow the WeChat public account [45度科研人] and leave me a message!
73
+ <div align=center><img width = '200' height ='200' src ="https://dunazo.oss-cn-beijing.aliyuncs.com/blog/wechat-simple.png"/></div>""")
74
+ keyTxt.submit(submit_key, keyTxt, [user_api_key, status_display])
75
+ keyTxt.change(submit_key, keyTxt, [user_api_key, status_display])
76
+ # Chatbot
77
+ user_input.submit(predict,[user_api_key,systemPromptTxt,history,user_input,chatbot,token_count,],[chatbot, history, status_display, token_count],show_progress=True)
78
+ user_input.submit(reset_textbox, [], [user_input])
79
+
80
+ submitBtn.click(predict,[user_api_key,systemPromptTxt,history,user_input,chatbot,token_count,],[chatbot, history, status_display, token_count],show_progress=True)
81
+ submitBtn.click(reset_textbox, [], [user_input])
82
+
83
+ emptyBtn.click(reset_state,outputs=[chatbot, history, token_count, status_display],show_progress=True,)
84
+
85
+ templateSelectDropdown.change(get_template_content,[promptTemplates, templateSelectDropdown, systemPromptTxt],[systemPromptTxt],show_progress=True,)
86
+ exportMarkdownBtn.click(export_markdown,[saveFileName, systemPromptTxt, history, chatbot],downloadFile,show_progress=True,)
87
+ downloadFile.change(load_chat_history,[downloadFile, systemPromptTxt, history, chatbot],[saveFileName, systemPromptTxt, history, chatbot],)
88
+
89
+ # demo.title = "Sydney-AI 2.0"
90
+
91
+ if __name__ == "__main__":
92
+ demo.queue().launch(debug=False,show_api=False,share=True)
chat_func.py ADDED
@@ -0,0 +1,300 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # -*- coding:utf-8 -*-
2
+ from __future__ import annotations
3
+ from typing import TYPE_CHECKING, List
4
+
5
+ import logging
6
+ import json
7
+ import os
8
+ import requests
9
+
10
+ from tqdm import tqdm
11
+
12
+ from presets import *
13
+ # from llama_func import *
14
+ from utils import *
15
+
16
+ # logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s")
17
+
18
+ if TYPE_CHECKING:
19
+ from typing import TypedDict
20
+
21
+ class DataframeData(TypedDict):
22
+ headers: List[str]
23
+ data: List[List[str | int | bool]]
24
+
25
+
26
+ initial_prompt = "You are a helpful assistant."
27
+ API_URL = "https://api.openai.com/v1/chat/completions"
28
+ TEMPLATES_DIR = "templates"
29
+
30
+ def get_response(
31
+ openai_api_key, system_prompt, history, stream, selected_model
32
+ ):
33
+ headers = {
34
+ "Content-Type": "application/json",
35
+ "Authorization": f"Bearer {openai_api_key}",
36
+ }
37
+
38
+ history = [construct_system(system_prompt), *history]
39
+
40
+ payload = {
41
+ "model": selected_model,
42
+ "messages": history, # [{"role": "user", "content": f"{inputs}"}],
43
+ "temperature": 1.0, # 1.0,
44
+ "top_p": 1.0, # 1.0,
45
+ "n": 1,
46
+ "stream": stream,
47
+ "presence_penalty": 0,
48
+ "frequency_penalty": 0,
49
+ }
50
+ if stream:
51
+ timeout = timeout_streaming
52
+ else:
53
+ timeout = timeout_all
54
+
55
+ # 获取环境变量中的代理设置
56
+ http_proxy = os.environ.get("HTTP_PROXY") or os.environ.get("http_proxy")
57
+ https_proxy = os.environ.get("HTTPS_PROXY") or os.environ.get("https_proxy")
58
+
59
+ # 如果存在代理设置,使用它们
60
+ proxies = {}
61
+ if http_proxy:
62
+ logging.info(f"Using HTTP proxy: {http_proxy}")
63
+ proxies["http"] = http_proxy
64
+ if https_proxy:
65
+ logging.info(f"Using HTTPS proxy: {https_proxy}")
66
+ proxies["https"] = https_proxy
67
+
68
+ # 如果有代理,使用代理发送请求,否则使用默认设置发送请求
69
+ if proxies:
70
+ response = requests.post(
71
+ API_URL,
72
+ headers=headers,
73
+ json=payload,
74
+ stream=True,
75
+ timeout=timeout,
76
+ proxies=proxies,
77
+ )
78
+ else:
79
+ response = requests.post(
80
+ API_URL,
81
+ headers=headers,
82
+ json=payload,
83
+ stream=True,
84
+ timeout=timeout,
85
+ )
86
+ return response
87
+
88
+
89
+ def stream_predict(
90
+ openai_api_key,
91
+ system_prompt,
92
+ history,
93
+ inputs,
94
+ chatbot,
95
+ all_token_counts,
96
+ selected_model,
97
+ fake_input=None,
98
+ display_append=""
99
+ ):
100
+ def get_return_value():
101
+ return chatbot, history, status_text, all_token_counts
102
+ # logging.info("实时回答模式")
103
+ partial_words = ""
104
+ counter = 0
105
+ status_text = "answering……"
106
+ history.append(construct_user(inputs))
107
+ history.append(construct_assistant(""))
108
+ if fake_input:
109
+ chatbot.append((fake_input, ""))
110
+ else:
111
+ chatbot.append((inputs, ""))
112
+ user_token_count = 0
113
+ if len(all_token_counts) == 0:
114
+ system_prompt_token_count = count_token(construct_system(system_prompt))
115
+ user_token_count = (
116
+ count_token(construct_user(inputs)) + system_prompt_token_count
117
+ )
118
+ else:
119
+ user_token_count = count_token(construct_user(inputs))
120
+ all_token_counts.append(user_token_count)
121
+ logging.info(f"input token count: {user_token_count}")
122
+ yield get_return_value()
123
+ try:
124
+ response = get_response(
125
+ openai_api_key,
126
+ system_prompt,
127
+ history,
128
+ True,
129
+ selected_model,
130
+ )
131
+ except requests.exceptions.ConnectTimeout:
132
+ status_text = (
133
+ standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
134
+ )
135
+ yield get_return_value()
136
+ return
137
+ except requests.exceptions.ReadTimeout:
138
+ status_text = standard_error_msg + read_timeout_prompt + error_retrieve_prompt
139
+ yield get_return_value()
140
+ return
141
+
142
+ yield get_return_value()
143
+ error_json_str = ""
144
+
145
+ for chunk in tqdm(response.iter_lines()):
146
+ if counter == 0:
147
+ counter += 1
148
+ continue
149
+ counter += 1
150
+ # check whether each line is non-empty
151
+ if chunk:
152
+ chunk = chunk.decode()
153
+ chunklength = len(chunk)
154
+ try:
155
+ chunk = json.loads(chunk[6:])
156
+ except json.JSONDecodeError:
157
+ logging.info(chunk)
158
+ error_json_str += chunk
159
+ status_text = f"JSON file parsing error. Please reset the conversation. received content: {error_json_str}"
160
+ yield get_return_value()
161
+ continue
162
+ # decode each line as response data is in bytes
163
+ if chunklength > 6 and "delta" in chunk["choices"][0]:
164
+ finish_reason = chunk["choices"][0]["finish_reason"]
165
+ status_text = construct_token_message(
166
+ sum(all_token_counts), stream=True
167
+ )
168
+ if finish_reason == "stop":
169
+ yield get_return_value()
170
+ break
171
+ try:
172
+ partial_words = (
173
+ partial_words + chunk["choices"][0]["delta"]["content"]
174
+ )
175
+ except KeyError:
176
+ status_text = (
177
+ standard_error_msg
178
+ + "Token count has reached the maxtoken limit. Please reset the conversation. Current Token Count: "
179
+ + str(sum(all_token_counts))
180
+ )
181
+ yield get_return_value()
182
+ break
183
+ history[-1] = construct_assistant(partial_words)
184
+ chatbot[-1] = (chatbot[-1][0], partial_words+display_append)
185
+ all_token_counts[-1] += 1
186
+ yield get_return_value()
187
+
188
+
189
+ def predict_all(
190
+ openai_api_key,
191
+ system_prompt,
192
+ history,
193
+ inputs,
194
+ chatbot,
195
+ all_token_counts,
196
+ selected_model,
197
+ fake_input=None,
198
+ display_append=""
199
+ ):
200
+ # logging.info("一次性回答模式")
201
+ history.append(construct_user(inputs))
202
+ history.append(construct_assistant(""))
203
+ if fake_input:
204
+ chatbot.append((fake_input, ""))
205
+ else:
206
+ chatbot.append((inputs, ""))
207
+ all_token_counts.append(count_token(construct_user(inputs)))
208
+ try:
209
+ response = get_response(
210
+ openai_api_key,
211
+ system_prompt,
212
+ history,
213
+ False,
214
+ selected_model,
215
+ )
216
+ except requests.exceptions.ConnectTimeout:
217
+ status_text = (
218
+ standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
219
+ )
220
+ return chatbot, history, status_text, all_token_counts
221
+ except requests.exceptions.ProxyError:
222
+ status_text = standard_error_msg + proxy_error_prompt + error_retrieve_prompt
223
+ return chatbot, history, status_text, all_token_counts
224
+ except requests.exceptions.SSLError:
225
+ status_text = standard_error_msg + ssl_error_prompt + error_retrieve_prompt
226
+ return chatbot, history, status_text, all_token_counts
227
+ response = json.loads(response.text)
228
+ content = response["choices"][0]["message"]["content"]
229
+ history[-1] = construct_assistant(content)
230
+ chatbot[-1] = (chatbot[-1][0], content+display_append)
231
+ total_token_count = response["usage"]["total_tokens"]
232
+ all_token_counts[-1] = total_token_count - sum(all_token_counts)
233
+ status_text = construct_token_message(total_token_count)
234
+ return chatbot, history, status_text, all_token_counts
235
+
236
+
237
+ def predict(
238
+ openai_api_key,
239
+ system_prompt,
240
+ history,
241
+ inputs,
242
+ chatbot,
243
+ all_token_counts,
244
+ stream=True,
245
+ selected_model=MODELS[0],
246
+ use_websearch=False,
247
+ files = None,
248
+ should_check_token_count=True,
249
+ ): # repetition_penalty, top_k
250
+
251
+ old_inputs = ""
252
+ link_references = ""
253
+
254
+ if len(openai_api_key) != 51:
255
+ status_text = standard_error_msg + no_apikey_msg
256
+ logging.info(status_text)
257
+ chatbot.append((inputs, ""))
258
+ if len(history) == 0:
259
+ history.append(construct_user(inputs))
260
+ history.append("")
261
+ all_token_counts.append(0)
262
+ else:
263
+ history[-2] = construct_user(inputs)
264
+ yield chatbot, history, status_text, all_token_counts
265
+ return
266
+
267
+ yield chatbot, history, "answering……", all_token_counts
268
+
269
+ if stream:
270
+ # logging.info("使用流式传输")
271
+ iter = stream_predict(
272
+ openai_api_key,
273
+ system_prompt,
274
+ history,
275
+ inputs,
276
+ chatbot,
277
+ all_token_counts,
278
+ selected_model,
279
+ fake_input=old_inputs,
280
+ display_append=link_references
281
+ )
282
+ for chatbot, history, status_text, all_token_counts in iter:
283
+ yield chatbot, history, status_text, all_token_counts
284
+ else:
285
+ # logging.info("不使用流式传输")
286
+ chatbot, history, status_text, all_token_counts = predict_all(
287
+ openai_api_key,
288
+ system_prompt,
289
+ history,
290
+ inputs,
291
+ chatbot,
292
+ all_token_counts,
293
+ selected_model,
294
+ fake_input=old_inputs,
295
+ display_append=link_references
296
+ )
297
+ yield chatbot, history, status_text, all_token_counts
298
+
299
+ logging.info(f"The current token count is{all_token_counts}")
300
+
presets.py ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # -*- coding:utf-8 -*-
2
+ use_websearch_checkbox=False
3
+ use_streaming_checkbox=True
4
+ model_select_dropdown="gpt-3.5-turbo"
5
+ # top_p=1
6
+ dockerflag = True
7
+ authflag = False
8
+
9
+ # ChatGPT 设置
10
+ initial_prompt = "You are a helpful assistant."
11
+ API_URL = "https://api.openai.com/v1/chat/completions"
12
+ HISTORY_DIR = "history"
13
+ TEMPLATES_DIR = "templates"
14
+
15
+ # 错误信息
16
+ standard_error_msg = "Error:" # 错误信息的标准前缀
17
+ error_retrieve_prompt = "Please check the network connection and the API-Key" # 获取对话时发生错误
18
+ connection_timeout_prompt = "Time out" # 连接超时
19
+ read_timeout_prompt = "Time out" # 读取超时
20
+ proxy_error_prompt = "Proxy error" # 代理错误
21
+ ssl_error_prompt = "SSL error" # SSL 错误
22
+ no_apikey_msg = "please check whether the input is correct" # API key 长度不足 51 位
23
+
24
+ max_token_streaming = 3500 # 流式对话时的最大 token 数
25
+ timeout_streaming = 5 # 流式对话时的超时时间
26
+ max_token_all = 3500 # 非流式对话时的最大 token 数
27
+ timeout_all = 200 # 非流式对话时的超时时间
28
+ enable_streaming_option = True # 是否启用选择选择是否实时显示回答的勾选框
29
+ HIDE_MY_KEY = True # 如果你想在UI中隐藏你的 API 密钥,将此值设置为 True
30
+
31
+ SIM_K = 5
32
+ INDEX_QUERY_TEMPRATURE = 1.0
33
+ title= """\
34
+ # <p align="center">Sydney-AI 2.0<b>"""
35
+
36
+ description = """\
37
+ <p>
38
+ <p>
39
+
40
+ 本应用是一款基于最新OpenAI API“gpt-3.5-turbo”开发的智能在线聊天应用。 该应用程序的运营成本由“45 Degrees Research Fellows”赞助。 目前,token 限制为 3500。如果你想取消这个限制,你可以输入你自己的 OpenAI API 密钥。 <p>
41
+ App默认角色为ChatGPT原版助手,但您也可以从模板提供的角色中进行选择。 如果您对自定义Prompt有好的建议,请联系我们!<p>
42
+ This app is an intelligent online chat app developed based on the newly released OpenAI API "gpt-3.5-turbo". The app's operating costs are sponsored by "45度科研人". Currently, the tokens is limited to 3500. If you want to remove this restriction, you can input your own OpenAI API key.<p>
43
+ The default model role of the app is the original assistant of ChatGPT, but you can also choose from the provided roles. If you have good suggestions for customizing Prompt, please contact us!<p>
44
+ """
45
+
46
+ MODELS = ["gpt-3.5-turbo", "gpt-3.5-turbo-0301",]
requirements.txt ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ gradio==3.19.1
2
+ mdtex2html
3
+ pypinyin
4
+ tiktoken
5
+ tqdm
6
+ Pygments
7
+ markdown
utils.py ADDED
@@ -0,0 +1,301 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # -*- coding:utf-8 -*-
2
+ from __future__ import annotations
3
+ from typing import TYPE_CHECKING, Any, Callable, Dict, List, Tuple, Type
4
+ import logging
5
+ import json
6
+ import os
7
+
8
+ import csv
9
+ import requests
10
+ import re
11
+
12
+ import gradio as gr
13
+ from pypinyin import lazy_pinyin
14
+ import tiktoken
15
+ import mdtex2html
16
+ from markdown import markdown
17
+ from pygments import highlight
18
+ from pygments.lexers import get_lexer_by_name
19
+ from pygments.formatters import HtmlFormatter
20
+
21
+ from presets import *
22
+
23
+ # logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s")
24
+
25
+ if TYPE_CHECKING:
26
+ from typing import TypedDict
27
+
28
+ class DataframeData(TypedDict):
29
+ headers: List[str]
30
+ data: List[List[str | int | bool]]
31
+
32
+
33
+ def count_token(message):
34
+ encoding = tiktoken.get_encoding("cl100k_base")
35
+ input_str = f"role: {message['role']}, content: {message['content']}"
36
+ length = len(encoding.encode(input_str))
37
+ return length
38
+
39
+
40
+ def markdown_to_html_with_syntax_highlight(md_str):
41
+ def replacer(match):
42
+ lang = match.group(1) or "text"
43
+ code = match.group(2)
44
+
45
+ try:
46
+ lexer = get_lexer_by_name(lang, stripall=True)
47
+ except ValueError:
48
+ lexer = get_lexer_by_name("text", stripall=True)
49
+
50
+ formatter = HtmlFormatter()
51
+ highlighted_code = highlight(code, lexer, formatter)
52
+
53
+ return f'<pre><code class="{lang}">{highlighted_code}</code></pre>'
54
+
55
+ code_block_pattern = r"```(\w+)?\n([\s\S]+?)\n```"
56
+ md_str = re.sub(code_block_pattern, replacer, md_str, flags=re.MULTILINE)
57
+
58
+ html_str = markdown(md_str)
59
+ return html_str
60
+
61
+
62
+ def normalize_markdown(md_text: str) -> str:
63
+ lines = md_text.split("\n")
64
+ normalized_lines = []
65
+ inside_list = False
66
+
67
+ for i, line in enumerate(lines):
68
+ if re.match(r"^(\d+\.|-|\*|\+)\s", line.strip()):
69
+ if not inside_list and i > 0 and lines[i - 1].strip() != "":
70
+ normalized_lines.append("")
71
+ inside_list = True
72
+ normalized_lines.append(line)
73
+ elif inside_list and line.strip() == "":
74
+ if i < len(lines) - 1 and not re.match(
75
+ r"^(\d+\.|-|\*|\+)\s", lines[i + 1].strip()
76
+ ):
77
+ normalized_lines.append(line)
78
+ continue
79
+ else:
80
+ inside_list = False
81
+ normalized_lines.append(line)
82
+
83
+ return "\n".join(normalized_lines)
84
+
85
+ def postprocess(
86
+ self, y: List[Tuple[str | None, str | None]]
87
+ ) -> List[Tuple[str | None, str | None]]:
88
+ """
89
+ Parameters:
90
+ y: List of tuples representing the message and response pairs. Each message and response should be a string, which may be in Markdown format.
91
+ Returns:
92
+ List of tuples representing the message and response. Each message and response will be a string of HTML.
93
+ """
94
+ if y is None or y == []:
95
+ return []
96
+ tag_regex = re.compile(r"^<\w+>[^<]+</\w+>")
97
+ if tag_regex.search(y[-1][1]):
98
+ y[-1] = (convert_user(y[-1][0]), y[-1][1])
99
+ else:
100
+ y[-1] = (convert_user(y[-1][0]), convert_mdtext(y[-1][1]))
101
+ return y
102
+
103
+ def convert_mdtext(md_text):
104
+ code_block_pattern = re.compile(r"```(.*?)(?:```|$)", re.DOTALL)
105
+ inline_code_pattern = re.compile(r"`(.*?)`", re.DOTALL)
106
+ code_blocks = code_block_pattern.findall(md_text)
107
+ non_code_parts = code_block_pattern.split(md_text)[::2]
108
+
109
+ result = []
110
+ for non_code, code in zip(non_code_parts, code_blocks + [""]):
111
+ if non_code.strip():
112
+ non_code = normalize_markdown(non_code)
113
+ if inline_code_pattern.search(non_code):
114
+ result.append(markdown(non_code, extensions=["tables"]))
115
+ else:
116
+ result.append(mdtex2html.convert(non_code, extensions=["tables"]))
117
+ if code.strip():
118
+ # _, code = detect_language(code) # 暂时去除代码高亮功能,因为在大段代码的情况下会出现问题
119
+ # code = code.replace("\n\n", "\n") # 暂时去除代码中的空行,因为在大段代码的情况下会出现问题
120
+ code = f"```{code}\n\n```"
121
+ code = markdown_to_html_with_syntax_highlight(code)
122
+ result.append(code)
123
+ result = "".join(result)
124
+ return result
125
+
126
+ def convert_user(userinput):
127
+ userinput = userinput.replace("\n", "<br>")
128
+ return f"<pre>{userinput}</pre>"
129
+
130
+ def construct_text(role, text):
131
+ return {"role": role, "content": text}
132
+
133
+
134
+ def construct_user(text):
135
+ return construct_text("user", text)
136
+
137
+
138
+ def construct_system(text):
139
+ return construct_text("system", text)
140
+
141
+
142
+ def construct_assistant(text):
143
+ return construct_text("assistant", text)
144
+
145
+
146
+ def construct_token_message(token, stream=False):
147
+ return f"Token count: {token}"
148
+
149
+
150
+ def save_file(filename, system, history, chatbot):
151
+ logging.info("saving......")
152
+ os.makedirs(HISTORY_DIR, exist_ok=True)
153
+ if filename.endswith(".json"):
154
+ json_s = {"system": system, "history": history, "chatbot": chatbot}
155
+ print(json_s)
156
+ with open(os.path.join(HISTORY_DIR, filename), "w") as f:
157
+ json.dump(json_s, f)
158
+ elif filename.endswith(".md"):
159
+ md_s = f"system: \n- {system} \n"
160
+ for data in history:
161
+ md_s += f"\n{data['role']}: \n- {data['content']} \n"
162
+ with open(os.path.join(HISTORY_DIR, filename), "w", encoding="utf8") as f:
163
+ f.write(md_s)
164
+ # logging.info("保存对话历史完毕")
165
+ return os.path.join(HISTORY_DIR, filename)
166
+
167
+
168
+ def save_chat_history(filename, system, history, chatbot):
169
+ if filename == "":
170
+ return
171
+ if not filename.endswith(".json"):
172
+ filename += ".json"
173
+ return save_file(filename, system, history, chatbot)
174
+
175
+
176
+ def export_markdown(filename, system, history, chatbot):
177
+ if filename == "":
178
+ return
179
+ if not filename.endswith(".md"):
180
+ filename += ".md"
181
+ return save_file(filename, system, history, chatbot)
182
+
183
+
184
+ def load_chat_history(filename, system, history, chatbot):
185
+ # logging.info("加载对话历史中……")
186
+ if type(filename) != str:
187
+ filename = filename.name
188
+ try:
189
+ with open(os.path.join(HISTORY_DIR, filename), "r") as f:
190
+ json_s = json.load(f)
191
+ try:
192
+ if type(json_s["history"][0]) == str:
193
+ # logging.info("历史记录格式为旧版,正在转换……")
194
+ new_history = []
195
+ for index, item in enumerate(json_s["history"]):
196
+ if index % 2 == 0:
197
+ new_history.append(construct_user(item))
198
+ else:
199
+ new_history.append(construct_assistant(item))
200
+ json_s["history"] = new_history
201
+ logging.info(new_history)
202
+ except:
203
+ # 没有对话历史
204
+ pass
205
+ # logging.info("加载对话历史完毕")
206
+ return filename, json_s["system"], json_s["history"], json_s["chatbot"]
207
+ except FileNotFoundError:
208
+ # logging.info("没有找到对话历史文件,不执行任何操作")
209
+ return filename, system, history, chatbot
210
+
211
+
212
+ def sorted_by_pinyin(list):
213
+ return sorted(list, key=lambda char: lazy_pinyin(char)[0][0])
214
+
215
+
216
+ def get_file_names(dir, plain=False, filetypes=[".json"]):
217
+ # logging.info(f"获取文件名列表,目录为{dir},文件类型为{filetypes},是否为纯文本列表{plain}")
218
+ files = []
219
+ try:
220
+ for type in filetypes:
221
+ files += [f for f in os.listdir(dir) if f.endswith(type)]
222
+ except FileNotFoundError:
223
+ files = []
224
+ files = sorted_by_pinyin(files)
225
+ if files == []:
226
+ files = [""]
227
+ if plain:
228
+ return files
229
+ else:
230
+ return gr.Dropdown.update(choices=files)
231
+
232
+
233
+ def get_history_names(plain=False):
234
+ # logging.info("获取历史记录文件名列表")
235
+ return get_file_names(HISTORY_DIR, plain)
236
+
237
+
238
+ def load_template(filename, mode=0):
239
+ # logging.info(f"加载模板文件{filename},模式为{mode}(0为返回字典和下拉菜单,1为返回下拉菜单,2为返回字典)")
240
+ lines = []
241
+ logging.info("Loading template...")
242
+ # filename='中文Prompts.json'
243
+ if filename.endswith(".json"):
244
+ with open(os.path.join(TEMPLATES_DIR, filename), "r", encoding="utf8") as f:
245
+ lines = json.load(f)
246
+ lines = [[i["act"], i["prompt"]] for i in lines]
247
+ else:
248
+ with open(
249
+ os.path.join(TEMPLATES_DIR, filename), "r", encoding="utf8") as csvfile:
250
+ reader = csv.reader(csvfile)
251
+ lines = list(reader)
252
+ lines = lines[1:]
253
+ if mode == 1:
254
+ return sorted_by_pinyin([row[0] for row in lines])
255
+ elif mode == 2:
256
+ return {row[0]: row[1] for row in lines}
257
+ else:
258
+ choices = sorted_by_pinyin([row[0] for row in lines])
259
+ return {row[0]: row[1] for row in lines}, gr.Dropdown.update(
260
+ choices=choices, value=choices[0])
261
+
262
+
263
+ def get_template_names(plain=False):
264
+ # logging.info("获取模板文件名列表")
265
+ return get_file_names(TEMPLATES_DIR, plain, filetypes=[".csv", "json"])
266
+
267
+
268
+ def get_template_content(templates, selection, original_system_prompt):
269
+ logging.info(f"Prompt: {selection}")
270
+ try:
271
+ return templates[selection]
272
+ except:
273
+ return original_system_prompt
274
+
275
+
276
+ def reset_state():
277
+ logging.info("Reset")
278
+ return [], [], [], construct_token_message(0)
279
+
280
+
281
+ def reset_textbox():
282
+ return gr.update(value="")
283
+
284
+
285
+ def hide_middle_chars(s):
286
+ if len(s) <= 8:
287
+ return s
288
+ else:
289
+ head = s[:4]
290
+ tail = s[-4:]
291
+ hidden = "*" * (len(s) - 8)
292
+ return head + hidden + tail
293
+
294
+ def submit_key(key):
295
+ key = key.strip()
296
+ msg = f"API-Key: {hide_middle_chars(key)}"
297
+ logging.info(msg)
298
+ return key, msg
299
+
300
+
301
+