Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Commit
•
a51e754
1
Parent(s):
85095bb
GitHub 94adb4f
Browse files- app.py +339 -58
- chat_func.py +423 -0
- custom.css +188 -0
- llama_func.py +192 -0
- overwrites.py +40 -0
- presets.py +59 -69
- requirements.txt +3 -0
- utils.py +100 -261
app.py
CHANGED
@@ -1,18 +1,24 @@
|
|
1 |
# -*- coding:utf-8 -*-
|
2 |
-
import gradio as gr
|
3 |
import os
|
4 |
import logging
|
5 |
import sys
|
6 |
-
|
|
|
|
|
7 |
from utils import *
|
8 |
from presets import *
|
|
|
|
|
9 |
|
10 |
-
logging.basicConfig(
|
|
|
|
|
|
|
11 |
|
12 |
-
my_api_key = ""
|
13 |
|
14 |
-
#if we are running in Docker
|
15 |
-
if os.environ.get(
|
16 |
dockerflag = True
|
17 |
else:
|
18 |
dockerflag = False
|
@@ -20,17 +26,21 @@ else:
|
|
20 |
authflag = False
|
21 |
|
22 |
if dockerflag:
|
23 |
-
my_api_key = os.environ.get(
|
24 |
if my_api_key == "empty":
|
25 |
logging.error("Please give a api key!")
|
26 |
sys.exit(1)
|
27 |
-
#auth
|
28 |
-
username = os.environ.get(
|
29 |
-
password = os.environ.get(
|
30 |
if not (isinstance(username, type(None)) or isinstance(password, type(None))):
|
31 |
authflag = True
|
32 |
else:
|
33 |
-
if
|
|
|
|
|
|
|
|
|
34 |
with open("api_key.txt", "r") as f:
|
35 |
my_api_key = f.read().strip()
|
36 |
if os.path.exists("auth.json"):
|
@@ -42,11 +52,78 @@ else:
|
|
42 |
authflag = True
|
43 |
|
44 |
gr.Chatbot.postprocess = postprocess
|
|
|
|
|
|
|
|
|
45 |
|
46 |
-
with gr.Blocks(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
history = gr.State([])
|
48 |
token_count = gr.State([])
|
49 |
promptTemplates = gr.State(load_template(get_template_names(plain=True)[0], mode=2))
|
|
|
50 |
TRUECOMSTANT = gr.State(True)
|
51 |
FALSECONSTANT = gr.State(False)
|
52 |
topic = gr.State("未命名对话历史记录")
|
@@ -58,114 +135,318 @@ with gr.Blocks(css=customCSS) as demo:
|
|
58 |
with gr.Row(scale=1).style(equal_height=True):
|
59 |
with gr.Column(scale=5):
|
60 |
with gr.Row(scale=1):
|
61 |
-
chatbot = gr.Chatbot().style(height=
|
62 |
with gr.Row(scale=1):
|
63 |
with gr.Column(scale=12):
|
64 |
-
user_input = gr.Textbox(
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
68 |
with gr.Row(scale=1):
|
69 |
-
emptyBtn = gr.Button(
|
|
|
|
|
70 |
retryBtn = gr.Button("🔄 重新生成")
|
71 |
delLastBtn = gr.Button("🗑️ 删除一条对话")
|
72 |
reduceTokenBtn = gr.Button("♻️ 总结对话")
|
73 |
|
74 |
with gr.Column():
|
75 |
-
with gr.Column(min_width=50,scale=1):
|
76 |
with gr.Tab(label="ChatGPT"):
|
77 |
-
keyTxt = gr.Textbox(
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
use_websearch_checkbox = gr.Checkbox(label="使用在线搜索", value=False)
|
|
|
86 |
|
87 |
with gr.Tab(label="Prompt"):
|
88 |
-
systemPromptTxt = gr.Textbox(
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
with gr.Accordion(label="加载Prompt模板", open=True):
|
90 |
with gr.Column():
|
91 |
with gr.Row():
|
92 |
with gr.Column(scale=6):
|
93 |
-
templateFileSelectDropdown = gr.Dropdown(
|
|
|
|
|
|
|
|
|
|
|
94 |
with gr.Column(scale=1):
|
95 |
templateRefreshBtn = gr.Button("🔄 刷新")
|
96 |
with gr.Row():
|
97 |
with gr.Column():
|
98 |
-
templateSelectDropdown = gr.Dropdown(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
|
100 |
with gr.Tab(label="保存/加载"):
|
101 |
with gr.Accordion(label="保存/加载对话历史记录", open=True):
|
102 |
-
gr.Markdown("对话历史默认保存在history文件夹中。")
|
103 |
with gr.Column():
|
104 |
with gr.Row():
|
105 |
with gr.Column(scale=6):
|
106 |
-
|
107 |
-
|
|
|
|
|
|
|
|
|
108 |
with gr.Column(scale=1):
|
109 |
-
|
110 |
-
exportMarkdownBtn = gr.Button("📝 导出为Markdown")
|
111 |
with gr.Row():
|
112 |
with gr.Column(scale=6):
|
113 |
-
|
|
|
|
|
|
|
|
|
|
|
114 |
with gr.Column(scale=1):
|
115 |
-
|
|
|
|
|
116 |
with gr.Row():
|
117 |
with gr.Column():
|
118 |
downloadFile = gr.File(interactive=True)
|
119 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
gr.Markdown(description)
|
121 |
|
|
|
122 |
# Chatbot
|
123 |
-
user_input.submit(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
user_input.submit(reset_textbox, [], [user_input])
|
125 |
|
126 |
-
submitBtn.click(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
127 |
submitBtn.click(reset_textbox, [], [user_input])
|
128 |
|
129 |
-
emptyBtn.click(
|
|
|
|
|
|
|
|
|
130 |
|
131 |
-
retryBtn.click(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
|
133 |
-
delLastBtn.click(
|
134 |
-
|
|
|
|
|
|
|
|
|
135 |
|
136 |
-
reduceTokenBtn.click(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
137 |
|
138 |
# Template
|
139 |
templateRefreshBtn.click(get_template_names, None, [templateFileSelectDropdown])
|
140 |
-
templateFileSelectDropdown.change(
|
141 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
|
143 |
# S&L
|
144 |
-
saveHistoryBtn.click(
|
|
|
|
|
|
|
|
|
|
|
145 |
saveHistoryBtn.click(get_history_names, None, [historyFileSelectDropdown])
|
146 |
-
exportMarkdownBtn.click(
|
|
|
|
|
|
|
|
|
|
|
147 |
historyRefreshBtn.click(get_history_names, None, [historyFileSelectDropdown])
|
148 |
-
historyFileSelectDropdown.change(
|
149 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
151 |
|
152 |
-
logging.info(
|
|
|
|
|
|
|
|
|
153 |
# 默认开启本地服务器,默认可以直接从IP访问,默认不创建公开分享链接
|
154 |
demo.title = "川虎ChatGPT 🚀"
|
155 |
|
156 |
if __name__ == "__main__":
|
157 |
-
#if running in Docker
|
158 |
if dockerflag:
|
159 |
if authflag:
|
160 |
-
demo.queue().launch(
|
|
|
|
|
161 |
else:
|
162 |
demo.queue().launch(server_name="0.0.0.0", server_port=7860, share=False)
|
163 |
-
#if not running in Docker
|
164 |
else:
|
165 |
if authflag:
|
166 |
demo.queue().launch(share=False, auth=(username, password))
|
167 |
else:
|
168 |
-
demo.queue().launch(share=False)
|
169 |
-
#demo.queue().launch(server_name="0.0.0.0", server_port=7860, share=False) # 可自定义端口
|
170 |
-
#demo.queue().launch(server_name="0.0.0.0", server_port=7860,auth=("在这里填写用户名", "在这里填写密码")) # 可设置用户名与密码
|
171 |
-
#demo.queue().launch(auth=("在这里填写用户名", "在这里填写密码")) # 适合Nginx反向代理
|
|
|
1 |
# -*- coding:utf-8 -*-
|
|
|
2 |
import os
|
3 |
import logging
|
4 |
import sys
|
5 |
+
|
6 |
+
import gradio as gr
|
7 |
+
|
8 |
from utils import *
|
9 |
from presets import *
|
10 |
+
from overwrites import *
|
11 |
+
from chat_func import *
|
12 |
|
13 |
+
logging.basicConfig(
|
14 |
+
level=logging.DEBUG,
|
15 |
+
format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s",
|
16 |
+
)
|
17 |
|
18 |
+
my_api_key = "" # 在这里输入你的 API 密钥
|
19 |
|
20 |
+
# if we are running in Docker
|
21 |
+
if os.environ.get("dockerrun") == "yes":
|
22 |
dockerflag = True
|
23 |
else:
|
24 |
dockerflag = False
|
|
|
26 |
authflag = False
|
27 |
|
28 |
if dockerflag:
|
29 |
+
my_api_key = os.environ.get("my_api_key")
|
30 |
if my_api_key == "empty":
|
31 |
logging.error("Please give a api key!")
|
32 |
sys.exit(1)
|
33 |
+
# auth
|
34 |
+
username = os.environ.get("USERNAME")
|
35 |
+
password = os.environ.get("PASSWORD")
|
36 |
if not (isinstance(username, type(None)) or isinstance(password, type(None))):
|
37 |
authflag = True
|
38 |
else:
|
39 |
+
if (
|
40 |
+
not my_api_key
|
41 |
+
and os.path.exists("api_key.txt")
|
42 |
+
and os.path.getsize("api_key.txt")
|
43 |
+
):
|
44 |
with open("api_key.txt", "r") as f:
|
45 |
my_api_key = f.read().strip()
|
46 |
if os.path.exists("auth.json"):
|
|
|
52 |
authflag = True
|
53 |
|
54 |
gr.Chatbot.postprocess = postprocess
|
55 |
+
PromptHelper.compact_text_chunks = compact_text_chunks
|
56 |
+
|
57 |
+
with open("custom.css", "r", encoding="utf-8") as f:
|
58 |
+
customCSS = f.read()
|
59 |
|
60 |
+
with gr.Blocks(
|
61 |
+
css=customCSS,
|
62 |
+
theme=gr.themes.Soft(
|
63 |
+
primary_hue=gr.themes.Color(
|
64 |
+
c50="#02C160",
|
65 |
+
c100="rgba(2, 193, 96, 0.2)",
|
66 |
+
c200="#02C160",
|
67 |
+
c300="rgba(2, 193, 96, 0.32)",
|
68 |
+
c400="rgba(2, 193, 96, 0.32)",
|
69 |
+
c500="rgba(2, 193, 96, 1.0)",
|
70 |
+
c600="rgba(2, 193, 96, 1.0)",
|
71 |
+
c700="rgba(2, 193, 96, 0.32)",
|
72 |
+
c800="rgba(2, 193, 96, 0.32)",
|
73 |
+
c900="#02C160",
|
74 |
+
c950="#02C160",
|
75 |
+
),
|
76 |
+
secondary_hue=gr.themes.Color(
|
77 |
+
c50="#576b95",
|
78 |
+
c100="#576b95",
|
79 |
+
c200="#576b95",
|
80 |
+
c300="#576b95",
|
81 |
+
c400="#576b95",
|
82 |
+
c500="#576b95",
|
83 |
+
c600="#576b95",
|
84 |
+
c700="#576b95",
|
85 |
+
c800="#576b95",
|
86 |
+
c900="#576b95",
|
87 |
+
c950="#576b95",
|
88 |
+
),
|
89 |
+
neutral_hue=gr.themes.Color(
|
90 |
+
name="gray",
|
91 |
+
c50="#f9fafb",
|
92 |
+
c100="#f3f4f6",
|
93 |
+
c200="#e5e7eb",
|
94 |
+
c300="#d1d5db",
|
95 |
+
c400="#B2B2B2",
|
96 |
+
c500="#808080",
|
97 |
+
c600="#636363",
|
98 |
+
c700="#515151",
|
99 |
+
c800="#393939",
|
100 |
+
c900="#272727",
|
101 |
+
c950="#171717",
|
102 |
+
),
|
103 |
+
radius_size=gr.themes.sizes.radius_sm,
|
104 |
+
).set(
|
105 |
+
button_primary_background_fill="#06AE56",
|
106 |
+
button_primary_background_fill_dark="#06AE56",
|
107 |
+
button_primary_background_fill_hover="#07C863",
|
108 |
+
button_primary_border_color="#06AE56",
|
109 |
+
button_primary_border_color_dark="#06AE56",
|
110 |
+
button_primary_text_color="#FFFFFF",
|
111 |
+
button_primary_text_color_dark="#FFFFFF",
|
112 |
+
button_secondary_background_fill="#F2F2F2",
|
113 |
+
button_secondary_background_fill_dark="#2B2B2B",
|
114 |
+
button_secondary_text_color="#393939",
|
115 |
+
button_secondary_text_color_dark="#FFFFFF",
|
116 |
+
# background_fill_primary="#F7F7F7",
|
117 |
+
# background_fill_primary_dark="#1F1F1F",
|
118 |
+
block_title_text_color="*primary_500",
|
119 |
+
block_title_background_fill = "*primary_100",
|
120 |
+
input_background_fill="#F6F6F6",
|
121 |
+
),
|
122 |
+
) as demo:
|
123 |
history = gr.State([])
|
124 |
token_count = gr.State([])
|
125 |
promptTemplates = gr.State(load_template(get_template_names(plain=True)[0], mode=2))
|
126 |
+
user_api_key = gr.State(my_api_key)
|
127 |
TRUECOMSTANT = gr.State(True)
|
128 |
FALSECONSTANT = gr.State(False)
|
129 |
topic = gr.State("未命名对话历史记录")
|
|
|
135 |
with gr.Row(scale=1).style(equal_height=True):
|
136 |
with gr.Column(scale=5):
|
137 |
with gr.Row(scale=1):
|
138 |
+
chatbot = gr.Chatbot(elem_id="chuanhu_chatbot").style(height="100%")
|
139 |
with gr.Row(scale=1):
|
140 |
with gr.Column(scale=12):
|
141 |
+
user_input = gr.Textbox(
|
142 |
+
show_label=False, placeholder="在这里输入"
|
143 |
+
).style(container=False)
|
144 |
+
with gr.Column(min_width=70, scale=1):
|
145 |
+
submitBtn = gr.Button("发送", variant="primary")
|
146 |
with gr.Row(scale=1):
|
147 |
+
emptyBtn = gr.Button(
|
148 |
+
"🧹 新的对话",
|
149 |
+
)
|
150 |
retryBtn = gr.Button("🔄 重新生成")
|
151 |
delLastBtn = gr.Button("🗑️ 删除一条对话")
|
152 |
reduceTokenBtn = gr.Button("♻️ 总结对话")
|
153 |
|
154 |
with gr.Column():
|
155 |
+
with gr.Column(min_width=50, scale=1):
|
156 |
with gr.Tab(label="ChatGPT"):
|
157 |
+
keyTxt = gr.Textbox(
|
158 |
+
show_label=True,
|
159 |
+
placeholder=f"OpenAI API-key...",
|
160 |
+
value=hide_middle_chars(my_api_key),
|
161 |
+
type="password",
|
162 |
+
visible=not HIDE_MY_KEY,
|
163 |
+
label="API-Key(按Enter提交)",
|
164 |
+
)
|
165 |
+
model_select_dropdown = gr.Dropdown(
|
166 |
+
label="选择模型", choices=MODELS, multiselect=False, value=MODELS[0]
|
167 |
+
)
|
168 |
+
use_streaming_checkbox = gr.Checkbox(
|
169 |
+
label="实时传输回答", value=True, visible=enable_streaming_option
|
170 |
+
)
|
171 |
use_websearch_checkbox = gr.Checkbox(label="使用在线搜索", value=False)
|
172 |
+
index_files = gr.Files(label="上传索引文件", type="file", multiple=True)
|
173 |
|
174 |
with gr.Tab(label="Prompt"):
|
175 |
+
systemPromptTxt = gr.Textbox(
|
176 |
+
show_label=True,
|
177 |
+
placeholder=f"在这里输入System Prompt...",
|
178 |
+
label="System prompt",
|
179 |
+
value=initial_prompt,
|
180 |
+
lines=10,
|
181 |
+
).style(container=True)
|
182 |
with gr.Accordion(label="加载Prompt模板", open=True):
|
183 |
with gr.Column():
|
184 |
with gr.Row():
|
185 |
with gr.Column(scale=6):
|
186 |
+
templateFileSelectDropdown = gr.Dropdown(
|
187 |
+
label="选择Prompt模板集合文件",
|
188 |
+
choices=get_template_names(plain=True),
|
189 |
+
multiselect=False,
|
190 |
+
value=get_template_names(plain=True)[0],
|
191 |
+
)
|
192 |
with gr.Column(scale=1):
|
193 |
templateRefreshBtn = gr.Button("🔄 刷新")
|
194 |
with gr.Row():
|
195 |
with gr.Column():
|
196 |
+
templateSelectDropdown = gr.Dropdown(
|
197 |
+
label="从Prompt模板中加载",
|
198 |
+
choices=load_template(
|
199 |
+
get_template_names(plain=True)[0], mode=1
|
200 |
+
),
|
201 |
+
multiselect=False,
|
202 |
+
value=load_template(
|
203 |
+
get_template_names(plain=True)[0], mode=1
|
204 |
+
)[0],
|
205 |
+
)
|
206 |
|
207 |
with gr.Tab(label="保存/加载"):
|
208 |
with gr.Accordion(label="保存/加载对话历史记录", open=True):
|
|
|
209 |
with gr.Column():
|
210 |
with gr.Row():
|
211 |
with gr.Column(scale=6):
|
212 |
+
historyFileSelectDropdown = gr.Dropdown(
|
213 |
+
label="从列表中加载对话",
|
214 |
+
choices=get_history_names(plain=True),
|
215 |
+
multiselect=False,
|
216 |
+
value=get_history_names(plain=True)[0],
|
217 |
+
)
|
218 |
with gr.Column(scale=1):
|
219 |
+
historyRefreshBtn = gr.Button("🔄 刷新")
|
|
|
220 |
with gr.Row():
|
221 |
with gr.Column(scale=6):
|
222 |
+
saveFileName = gr.Textbox(
|
223 |
+
show_label=True,
|
224 |
+
placeholder=f"设置文件名: 默认为.json,可选为.md",
|
225 |
+
label="设置保存文件名",
|
226 |
+
value="对话历史记录",
|
227 |
+
).style(container=True)
|
228 |
with gr.Column(scale=1):
|
229 |
+
saveHistoryBtn = gr.Button("💾 保存对话")
|
230 |
+
exportMarkdownBtn = gr.Button("📝 导出为Markdown")
|
231 |
+
gr.Markdown("默认保存于history文件夹")
|
232 |
with gr.Row():
|
233 |
with gr.Column():
|
234 |
downloadFile = gr.File(interactive=True)
|
235 |
|
236 |
+
with gr.Tab(label="高级"):
|
237 |
+
default_btn = gr.Button("🔙 恢复默认设置")
|
238 |
+
gr.Markdown("# ⚠️ 务必谨慎更改 ⚠️\n\n如果无法使用请恢复默认设置")
|
239 |
+
|
240 |
+
with gr.Accordion("参数", open=False):
|
241 |
+
top_p = gr.Slider(
|
242 |
+
minimum=-0,
|
243 |
+
maximum=1.0,
|
244 |
+
value=1.0,
|
245 |
+
step=0.05,
|
246 |
+
interactive=True,
|
247 |
+
label="Top-p (nucleus sampling)",
|
248 |
+
)
|
249 |
+
temperature = gr.Slider(
|
250 |
+
minimum=-0,
|
251 |
+
maximum=2.0,
|
252 |
+
value=1.0,
|
253 |
+
step=0.1,
|
254 |
+
interactive=True,
|
255 |
+
label="Temperature",
|
256 |
+
)
|
257 |
+
|
258 |
+
apiurlTxt = gr.Textbox(
|
259 |
+
show_label=True,
|
260 |
+
placeholder=f"在这里输入API地址...",
|
261 |
+
label="API地址",
|
262 |
+
value="https://api.openai.com/v1/chat/completions",
|
263 |
+
lines=2,
|
264 |
+
)
|
265 |
+
changeAPIURLBtn = gr.Button("🔄 切换API地址")
|
266 |
+
proxyTxt = gr.Textbox(
|
267 |
+
show_label=True,
|
268 |
+
placeholder=f"在这里输入代理地址...",
|
269 |
+
label="代理地址(示例:http://127.0.0.1:10809)",
|
270 |
+
value="",
|
271 |
+
lines=2,
|
272 |
+
)
|
273 |
+
changeProxyBtn = gr.Button("🔄 设置代理地址")
|
274 |
+
|
275 |
gr.Markdown(description)
|
276 |
|
277 |
+
keyTxt.submit(submit_key, keyTxt, [user_api_key, status_display])
|
278 |
# Chatbot
|
279 |
+
user_input.submit(
|
280 |
+
predict,
|
281 |
+
[
|
282 |
+
user_api_key,
|
283 |
+
systemPromptTxt,
|
284 |
+
history,
|
285 |
+
user_input,
|
286 |
+
chatbot,
|
287 |
+
token_count,
|
288 |
+
top_p,
|
289 |
+
temperature,
|
290 |
+
use_streaming_checkbox,
|
291 |
+
model_select_dropdown,
|
292 |
+
use_websearch_checkbox,
|
293 |
+
index_files
|
294 |
+
],
|
295 |
+
[chatbot, history, status_display, token_count],
|
296 |
+
show_progress=True,
|
297 |
+
)
|
298 |
user_input.submit(reset_textbox, [], [user_input])
|
299 |
|
300 |
+
submitBtn.click(
|
301 |
+
predict,
|
302 |
+
[
|
303 |
+
user_api_key,
|
304 |
+
systemPromptTxt,
|
305 |
+
history,
|
306 |
+
user_input,
|
307 |
+
chatbot,
|
308 |
+
token_count,
|
309 |
+
top_p,
|
310 |
+
temperature,
|
311 |
+
use_streaming_checkbox,
|
312 |
+
model_select_dropdown,
|
313 |
+
use_websearch_checkbox,
|
314 |
+
index_files
|
315 |
+
],
|
316 |
+
[chatbot, history, status_display, token_count],
|
317 |
+
show_progress=True,
|
318 |
+
)
|
319 |
submitBtn.click(reset_textbox, [], [user_input])
|
320 |
|
321 |
+
emptyBtn.click(
|
322 |
+
reset_state,
|
323 |
+
outputs=[chatbot, history, token_count, status_display],
|
324 |
+
show_progress=True,
|
325 |
+
)
|
326 |
|
327 |
+
retryBtn.click(
|
328 |
+
retry,
|
329 |
+
[
|
330 |
+
user_api_key,
|
331 |
+
systemPromptTxt,
|
332 |
+
history,
|
333 |
+
chatbot,
|
334 |
+
token_count,
|
335 |
+
top_p,
|
336 |
+
temperature,
|
337 |
+
use_streaming_checkbox,
|
338 |
+
model_select_dropdown,
|
339 |
+
],
|
340 |
+
[chatbot, history, status_display, token_count],
|
341 |
+
show_progress=True,
|
342 |
+
)
|
343 |
|
344 |
+
delLastBtn.click(
|
345 |
+
delete_last_conversation,
|
346 |
+
[chatbot, history, token_count],
|
347 |
+
[chatbot, history, token_count, status_display],
|
348 |
+
show_progress=True,
|
349 |
+
)
|
350 |
|
351 |
+
reduceTokenBtn.click(
|
352 |
+
reduce_token_size,
|
353 |
+
[
|
354 |
+
user_api_key,
|
355 |
+
systemPromptTxt,
|
356 |
+
history,
|
357 |
+
chatbot,
|
358 |
+
token_count,
|
359 |
+
top_p,
|
360 |
+
temperature,
|
361 |
+
use_streaming_checkbox,
|
362 |
+
model_select_dropdown,
|
363 |
+
],
|
364 |
+
[chatbot, history, status_display, token_count],
|
365 |
+
show_progress=True,
|
366 |
+
)
|
367 |
|
368 |
# Template
|
369 |
templateRefreshBtn.click(get_template_names, None, [templateFileSelectDropdown])
|
370 |
+
templateFileSelectDropdown.change(
|
371 |
+
load_template,
|
372 |
+
[templateFileSelectDropdown],
|
373 |
+
[promptTemplates, templateSelectDropdown],
|
374 |
+
show_progress=True,
|
375 |
+
)
|
376 |
+
templateSelectDropdown.change(
|
377 |
+
get_template_content,
|
378 |
+
[promptTemplates, templateSelectDropdown, systemPromptTxt],
|
379 |
+
[systemPromptTxt],
|
380 |
+
show_progress=True,
|
381 |
+
)
|
382 |
|
383 |
# S&L
|
384 |
+
saveHistoryBtn.click(
|
385 |
+
save_chat_history,
|
386 |
+
[saveFileName, systemPromptTxt, history, chatbot],
|
387 |
+
downloadFile,
|
388 |
+
show_progress=True,
|
389 |
+
)
|
390 |
saveHistoryBtn.click(get_history_names, None, [historyFileSelectDropdown])
|
391 |
+
exportMarkdownBtn.click(
|
392 |
+
export_markdown,
|
393 |
+
[saveFileName, systemPromptTxt, history, chatbot],
|
394 |
+
downloadFile,
|
395 |
+
show_progress=True,
|
396 |
+
)
|
397 |
historyRefreshBtn.click(get_history_names, None, [historyFileSelectDropdown])
|
398 |
+
historyFileSelectDropdown.change(
|
399 |
+
load_chat_history,
|
400 |
+
[historyFileSelectDropdown, systemPromptTxt, history, chatbot],
|
401 |
+
[saveFileName, systemPromptTxt, history, chatbot],
|
402 |
+
show_progress=True,
|
403 |
+
)
|
404 |
+
downloadFile.change(
|
405 |
+
load_chat_history,
|
406 |
+
[downloadFile, systemPromptTxt, history, chatbot],
|
407 |
+
[saveFileName, systemPromptTxt, history, chatbot],
|
408 |
+
)
|
409 |
|
410 |
+
# Advanced
|
411 |
+
default_btn.click(
|
412 |
+
reset_default, [], [apiurlTxt, proxyTxt, status_display], show_progress=True
|
413 |
+
)
|
414 |
+
changeAPIURLBtn.click(
|
415 |
+
change_api_url,
|
416 |
+
[apiurlTxt],
|
417 |
+
[status_display],
|
418 |
+
show_progress=True,
|
419 |
+
)
|
420 |
+
changeProxyBtn.click(
|
421 |
+
change_proxy,
|
422 |
+
[proxyTxt],
|
423 |
+
[status_display],
|
424 |
+
show_progress=True,
|
425 |
+
)
|
426 |
|
427 |
+
logging.info(
|
428 |
+
colorama.Back.GREEN
|
429 |
+
+ "\n川虎的温馨提示:访问 http://localhost:7860 查看界面"
|
430 |
+
+ colorama.Style.RESET_ALL
|
431 |
+
)
|
432 |
# 默认开启本地服务器,默认可以直接从IP访问,默认不创建公开分享链接
|
433 |
demo.title = "川虎ChatGPT 🚀"
|
434 |
|
435 |
if __name__ == "__main__":
|
436 |
+
# if running in Docker
|
437 |
if dockerflag:
|
438 |
if authflag:
|
439 |
+
demo.queue().launch(
|
440 |
+
server_name="0.0.0.0", server_port=7860, auth=(username, password)
|
441 |
+
)
|
442 |
else:
|
443 |
demo.queue().launch(server_name="0.0.0.0", server_port=7860, share=False)
|
444 |
+
# if not running in Docker
|
445 |
else:
|
446 |
if authflag:
|
447 |
demo.queue().launch(share=False, auth=(username, password))
|
448 |
else:
|
449 |
+
demo.queue().launch(share=False) # 改为 share=True 可以创建公开分享链接
|
450 |
+
# demo.queue().launch(server_name="0.0.0.0", server_port=7860, share=False) # 可自定义端口
|
451 |
+
# demo.queue().launch(server_name="0.0.0.0", server_port=7860,auth=("在这里填写用户名", "在这里填写密码")) # 可设置用户名与密码
|
452 |
+
# demo.queue().launch(auth=("在这里填写用户名", "在这里填写密码")) # 适合Nginx反向代理
|
chat_func.py
ADDED
@@ -0,0 +1,423 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
import colorama
|
12 |
+
from duckduckgo_search import ddg
|
13 |
+
|
14 |
+
from presets import *
|
15 |
+
from llama_func import *
|
16 |
+
from utils import *
|
17 |
+
|
18 |
+
# logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s")
|
19 |
+
|
20 |
+
if TYPE_CHECKING:
|
21 |
+
from typing import TypedDict
|
22 |
+
|
23 |
+
class DataframeData(TypedDict):
|
24 |
+
headers: List[str]
|
25 |
+
data: List[List[str | int | bool]]
|
26 |
+
|
27 |
+
|
28 |
+
initial_prompt = "You are a helpful assistant."
|
29 |
+
API_URL = "https://api.openai.com/v1/chat/completions"
|
30 |
+
HISTORY_DIR = "history"
|
31 |
+
TEMPLATES_DIR = "templates"
|
32 |
+
|
33 |
+
def get_response(
|
34 |
+
openai_api_key, system_prompt, history, temperature, top_p, stream, selected_model
|
35 |
+
):
|
36 |
+
headers = {
|
37 |
+
"Content-Type": "application/json",
|
38 |
+
"Authorization": f"Bearer {openai_api_key}",
|
39 |
+
}
|
40 |
+
|
41 |
+
history = [construct_system(system_prompt), *history]
|
42 |
+
|
43 |
+
payload = {
|
44 |
+
"model": selected_model,
|
45 |
+
"messages": history, # [{"role": "user", "content": f"{inputs}"}],
|
46 |
+
"temperature": temperature, # 1.0,
|
47 |
+
"top_p": top_p, # 1.0,
|
48 |
+
"n": 1,
|
49 |
+
"stream": stream,
|
50 |
+
"presence_penalty": 0,
|
51 |
+
"frequency_penalty": 0,
|
52 |
+
}
|
53 |
+
if stream:
|
54 |
+
timeout = timeout_streaming
|
55 |
+
else:
|
56 |
+
timeout = timeout_all
|
57 |
+
|
58 |
+
# 获取环境变量中的代理设置
|
59 |
+
http_proxy = os.environ.get("HTTP_PROXY") or os.environ.get("http_proxy")
|
60 |
+
https_proxy = os.environ.get("HTTPS_PROXY") or os.environ.get("https_proxy")
|
61 |
+
|
62 |
+
# 如果存在代理设置,使用它们
|
63 |
+
proxies = {}
|
64 |
+
if http_proxy:
|
65 |
+
logging.info(f"Using HTTP proxy: {http_proxy}")
|
66 |
+
proxies["http"] = http_proxy
|
67 |
+
if https_proxy:
|
68 |
+
logging.info(f"Using HTTPS proxy: {https_proxy}")
|
69 |
+
proxies["https"] = https_proxy
|
70 |
+
|
71 |
+
# 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
|
72 |
+
if proxies:
|
73 |
+
response = requests.post(
|
74 |
+
API_URL,
|
75 |
+
headers=headers,
|
76 |
+
json=payload,
|
77 |
+
stream=True,
|
78 |
+
timeout=timeout,
|
79 |
+
proxies=proxies,
|
80 |
+
)
|
81 |
+
else:
|
82 |
+
response = requests.post(
|
83 |
+
API_URL,
|
84 |
+
headers=headers,
|
85 |
+
json=payload,
|
86 |
+
stream=True,
|
87 |
+
timeout=timeout,
|
88 |
+
)
|
89 |
+
return response
|
90 |
+
|
91 |
+
|
92 |
+
def stream_predict(
|
93 |
+
openai_api_key,
|
94 |
+
system_prompt,
|
95 |
+
history,
|
96 |
+
inputs,
|
97 |
+
chatbot,
|
98 |
+
all_token_counts,
|
99 |
+
top_p,
|
100 |
+
temperature,
|
101 |
+
selected_model,
|
102 |
+
):
|
103 |
+
def get_return_value():
|
104 |
+
return chatbot, history, status_text, all_token_counts
|
105 |
+
|
106 |
+
logging.info("实时回答模式")
|
107 |
+
partial_words = ""
|
108 |
+
counter = 0
|
109 |
+
status_text = "开始实时传输回答……"
|
110 |
+
history.append(construct_user(inputs))
|
111 |
+
history.append(construct_assistant(""))
|
112 |
+
chatbot.append((parse_text(inputs), ""))
|
113 |
+
user_token_count = 0
|
114 |
+
if len(all_token_counts) == 0:
|
115 |
+
system_prompt_token_count = count_token(construct_system(system_prompt))
|
116 |
+
user_token_count = (
|
117 |
+
count_token(construct_user(inputs)) + system_prompt_token_count
|
118 |
+
)
|
119 |
+
else:
|
120 |
+
user_token_count = count_token(construct_user(inputs))
|
121 |
+
all_token_counts.append(user_token_count)
|
122 |
+
logging.info(f"输入token计数: {user_token_count}")
|
123 |
+
yield get_return_value()
|
124 |
+
try:
|
125 |
+
response = get_response(
|
126 |
+
openai_api_key,
|
127 |
+
system_prompt,
|
128 |
+
history,
|
129 |
+
temperature,
|
130 |
+
top_p,
|
131 |
+
True,
|
132 |
+
selected_model,
|
133 |
+
)
|
134 |
+
except requests.exceptions.ConnectTimeout:
|
135 |
+
status_text = (
|
136 |
+
standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
|
137 |
+
)
|
138 |
+
yield get_return_value()
|
139 |
+
return
|
140 |
+
except requests.exceptions.ReadTimeout:
|
141 |
+
status_text = standard_error_msg + read_timeout_prompt + error_retrieve_prompt
|
142 |
+
yield get_return_value()
|
143 |
+
return
|
144 |
+
|
145 |
+
yield get_return_value()
|
146 |
+
error_json_str = ""
|
147 |
+
|
148 |
+
for chunk in tqdm(response.iter_lines()):
|
149 |
+
if counter == 0:
|
150 |
+
counter += 1
|
151 |
+
continue
|
152 |
+
counter += 1
|
153 |
+
# check whether each line is non-empty
|
154 |
+
if chunk:
|
155 |
+
chunk = chunk.decode()
|
156 |
+
chunklength = len(chunk)
|
157 |
+
try:
|
158 |
+
chunk = json.loads(chunk[6:])
|
159 |
+
except json.JSONDecodeError:
|
160 |
+
logging.info(chunk)
|
161 |
+
error_json_str += chunk
|
162 |
+
status_text = f"JSON解析错误。请重置对话。收到的内容: {error_json_str}"
|
163 |
+
yield get_return_value()
|
164 |
+
continue
|
165 |
+
# decode each line as response data is in bytes
|
166 |
+
if chunklength > 6 and "delta" in chunk["choices"][0]:
|
167 |
+
finish_reason = chunk["choices"][0]["finish_reason"]
|
168 |
+
status_text = construct_token_message(
|
169 |
+
sum(all_token_counts), stream=True
|
170 |
+
)
|
171 |
+
if finish_reason == "stop":
|
172 |
+
yield get_return_value()
|
173 |
+
break
|
174 |
+
try:
|
175 |
+
partial_words = (
|
176 |
+
partial_words + chunk["choices"][0]["delta"]["content"]
|
177 |
+
)
|
178 |
+
except KeyError:
|
179 |
+
status_text = (
|
180 |
+
standard_error_msg
|
181 |
+
+ "API回复中找不到内容。很可能是Token计数达到上限了。请重置对话。当前Token计数: "
|
182 |
+
+ str(sum(all_token_counts))
|
183 |
+
)
|
184 |
+
yield get_return_value()
|
185 |
+
break
|
186 |
+
history[-1] = construct_assistant(partial_words)
|
187 |
+
chatbot[-1] = (parse_text(inputs), parse_text(partial_words))
|
188 |
+
all_token_counts[-1] += 1
|
189 |
+
yield get_return_value()
|
190 |
+
|
191 |
+
|
192 |
+
def predict_all(
|
193 |
+
openai_api_key,
|
194 |
+
system_prompt,
|
195 |
+
history,
|
196 |
+
inputs,
|
197 |
+
chatbot,
|
198 |
+
all_token_counts,
|
199 |
+
top_p,
|
200 |
+
temperature,
|
201 |
+
selected_model,
|
202 |
+
):
|
203 |
+
logging.info("一次性回答模式")
|
204 |
+
history.append(construct_user(inputs))
|
205 |
+
history.append(construct_assistant(""))
|
206 |
+
chatbot.append((parse_text(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 |
+
temperature,
|
214 |
+
top_p,
|
215 |
+
False,
|
216 |
+
selected_model,
|
217 |
+
)
|
218 |
+
except requests.exceptions.ConnectTimeout:
|
219 |
+
status_text = (
|
220 |
+
standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
|
221 |
+
)
|
222 |
+
return chatbot, history, status_text, all_token_counts
|
223 |
+
except requests.exceptions.ProxyError:
|
224 |
+
status_text = standard_error_msg + proxy_error_prompt + error_retrieve_prompt
|
225 |
+
return chatbot, history, status_text, all_token_counts
|
226 |
+
except requests.exceptions.SSLError:
|
227 |
+
status_text = standard_error_msg + ssl_error_prompt + error_retrieve_prompt
|
228 |
+
return chatbot, history, status_text, all_token_counts
|
229 |
+
response = json.loads(response.text)
|
230 |
+
content = response["choices"][0]["message"]["content"]
|
231 |
+
history[-1] = construct_assistant(content)
|
232 |
+
chatbot[-1] = (parse_text(inputs), parse_text(content))
|
233 |
+
total_token_count = response["usage"]["total_tokens"]
|
234 |
+
all_token_counts[-1] = total_token_count - sum(all_token_counts)
|
235 |
+
status_text = construct_token_message(total_token_count)
|
236 |
+
return chatbot, history, status_text, all_token_counts
|
237 |
+
|
238 |
+
|
239 |
+
def predict(
|
240 |
+
openai_api_key,
|
241 |
+
system_prompt,
|
242 |
+
history,
|
243 |
+
inputs,
|
244 |
+
chatbot,
|
245 |
+
all_token_counts,
|
246 |
+
top_p,
|
247 |
+
temperature,
|
248 |
+
stream=False,
|
249 |
+
selected_model=MODELS[0],
|
250 |
+
use_websearch_checkbox=False,
|
251 |
+
files = None,
|
252 |
+
should_check_token_count=True,
|
253 |
+
): # repetition_penalty, top_k
|
254 |
+
logging.info("输入为:" + colorama.Fore.BLUE + f"{inputs}" + colorama.Style.RESET_ALL)
|
255 |
+
if files:
|
256 |
+
msg = "构建索引中……(这可能需要比较久的时间)"
|
257 |
+
logging.info(msg)
|
258 |
+
yield chatbot, history, msg, all_token_counts
|
259 |
+
index = construct_index(openai_api_key, file_src=files)
|
260 |
+
msg = "索引构建完成,获取回答中……"
|
261 |
+
yield chatbot, history, msg, all_token_counts
|
262 |
+
history, chatbot, status_text = chat_ai(openai_api_key, index, inputs, history, chatbot)
|
263 |
+
yield chatbot, history, status_text, all_token_counts
|
264 |
+
return
|
265 |
+
if use_websearch_checkbox:
|
266 |
+
results = ddg(inputs, max_results=3)
|
267 |
+
web_results = []
|
268 |
+
for idx, result in enumerate(results):
|
269 |
+
logging.info(f"搜索结果{idx + 1}:{result}")
|
270 |
+
web_results.append(f'[{idx+1}]"{result["body"]}"\nURL: {result["href"]}')
|
271 |
+
web_results = "\n\n".join(web_results)
|
272 |
+
inputs = (
|
273 |
+
replace_today(WEBSEARCH_PTOMPT_TEMPLATE)
|
274 |
+
.replace("{query}", inputs)
|
275 |
+
.replace("{web_results}", web_results)
|
276 |
+
)
|
277 |
+
if len(openai_api_key) != 51:
|
278 |
+
status_text = standard_error_msg + no_apikey_msg
|
279 |
+
logging.info(status_text)
|
280 |
+
chatbot.append((parse_text(inputs), ""))
|
281 |
+
if len(history) == 0:
|
282 |
+
history.append(construct_user(inputs))
|
283 |
+
history.append("")
|
284 |
+
all_token_counts.append(0)
|
285 |
+
else:
|
286 |
+
history[-2] = construct_user(inputs)
|
287 |
+
yield chatbot, history, status_text, all_token_counts
|
288 |
+
return
|
289 |
+
if stream:
|
290 |
+
yield chatbot, history, "开始生成回答……", all_token_counts
|
291 |
+
if stream:
|
292 |
+
logging.info("使用流式传输")
|
293 |
+
iter = stream_predict(
|
294 |
+
openai_api_key,
|
295 |
+
system_prompt,
|
296 |
+
history,
|
297 |
+
inputs,
|
298 |
+
chatbot,
|
299 |
+
all_token_counts,
|
300 |
+
top_p,
|
301 |
+
temperature,
|
302 |
+
selected_model,
|
303 |
+
)
|
304 |
+
for chatbot, history, status_text, all_token_counts in iter:
|
305 |
+
yield chatbot, history, status_text, all_token_counts
|
306 |
+
else:
|
307 |
+
logging.info("不使用流式传输")
|
308 |
+
chatbot, history, status_text, all_token_counts = predict_all(
|
309 |
+
openai_api_key,
|
310 |
+
system_prompt,
|
311 |
+
history,
|
312 |
+
inputs,
|
313 |
+
chatbot,
|
314 |
+
all_token_counts,
|
315 |
+
top_p,
|
316 |
+
temperature,
|
317 |
+
selected_model,
|
318 |
+
)
|
319 |
+
yield chatbot, history, status_text, all_token_counts
|
320 |
+
logging.info(f"传输完毕。当前token计数为{all_token_counts}")
|
321 |
+
if len(history) > 1 and history[-1]["content"] != inputs:
|
322 |
+
logging.info(
|
323 |
+
"回答为:"
|
324 |
+
+ colorama.Fore.BLUE
|
325 |
+
+ f"{history[-1]['content']}"
|
326 |
+
+ colorama.Style.RESET_ALL
|
327 |
+
)
|
328 |
+
if stream:
|
329 |
+
max_token = max_token_streaming
|
330 |
+
else:
|
331 |
+
max_token = max_token_all
|
332 |
+
if sum(all_token_counts) > max_token and should_check_token_count:
|
333 |
+
status_text = f"精简token中{all_token_counts}/{max_token}"
|
334 |
+
logging.info(status_text)
|
335 |
+
yield chatbot, history, status_text, all_token_counts
|
336 |
+
iter = reduce_token_size(
|
337 |
+
openai_api_key,
|
338 |
+
system_prompt,
|
339 |
+
history,
|
340 |
+
chatbot,
|
341 |
+
all_token_counts,
|
342 |
+
top_p,
|
343 |
+
temperature,
|
344 |
+
stream=False,
|
345 |
+
selected_model=selected_model,
|
346 |
+
hidden=True,
|
347 |
+
)
|
348 |
+
for chatbot, history, status_text, all_token_counts in iter:
|
349 |
+
status_text = f"Token 达到上限,已自动降低Token计数至 {status_text}"
|
350 |
+
yield chatbot, history, status_text, all_token_counts
|
351 |
+
|
352 |
+
|
353 |
+
def retry(
|
354 |
+
openai_api_key,
|
355 |
+
system_prompt,
|
356 |
+
history,
|
357 |
+
chatbot,
|
358 |
+
token_count,
|
359 |
+
top_p,
|
360 |
+
temperature,
|
361 |
+
stream=False,
|
362 |
+
selected_model=MODELS[0],
|
363 |
+
):
|
364 |
+
logging.info("重试中……")
|
365 |
+
if len(history) == 0:
|
366 |
+
yield chatbot, history, f"{standard_error_msg}上下文是空的", token_count
|
367 |
+
return
|
368 |
+
history.pop()
|
369 |
+
inputs = history.pop()["content"]
|
370 |
+
token_count.pop()
|
371 |
+
iter = predict(
|
372 |
+
openai_api_key,
|
373 |
+
system_prompt,
|
374 |
+
history,
|
375 |
+
inputs,
|
376 |
+
chatbot,
|
377 |
+
token_count,
|
378 |
+
top_p,
|
379 |
+
temperature,
|
380 |
+
stream=stream,
|
381 |
+
selected_model=selected_model,
|
382 |
+
)
|
383 |
+
logging.info("重试完毕")
|
384 |
+
for x in iter:
|
385 |
+
yield x
|
386 |
+
|
387 |
+
|
388 |
+
def reduce_token_size(
|
389 |
+
openai_api_key,
|
390 |
+
system_prompt,
|
391 |
+
history,
|
392 |
+
chatbot,
|
393 |
+
token_count,
|
394 |
+
top_p,
|
395 |
+
temperature,
|
396 |
+
stream=False,
|
397 |
+
selected_model=MODELS[0],
|
398 |
+
hidden=False,
|
399 |
+
):
|
400 |
+
logging.info("开始减少token数量……")
|
401 |
+
iter = predict(
|
402 |
+
openai_api_key,
|
403 |
+
system_prompt,
|
404 |
+
history,
|
405 |
+
summarize_prompt,
|
406 |
+
chatbot,
|
407 |
+
token_count,
|
408 |
+
top_p,
|
409 |
+
temperature,
|
410 |
+
stream=stream,
|
411 |
+
selected_model=selected_model,
|
412 |
+
should_check_token_count=False,
|
413 |
+
)
|
414 |
+
logging.info(f"chatbot: {chatbot}")
|
415 |
+
for chatbot, history, status_text, previous_token_count in iter:
|
416 |
+
history = history[-2:]
|
417 |
+
token_count = previous_token_count[-1:]
|
418 |
+
if hidden:
|
419 |
+
chatbot.pop()
|
420 |
+
yield chatbot, history, construct_token_message(
|
421 |
+
sum(token_count), stream=stream
|
422 |
+
), token_count
|
423 |
+
logging.info("减少token数量完毕")
|
custom.css
ADDED
@@ -0,0 +1,188 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
/* status_display */
|
2 |
+
#status_display {
|
3 |
+
display: flex;
|
4 |
+
min-height: 2.5em;
|
5 |
+
align-items: flex-end;
|
6 |
+
justify-content: flex-end;
|
7 |
+
}
|
8 |
+
#status_display p {
|
9 |
+
font-size: .85em;
|
10 |
+
font-family: monospace;
|
11 |
+
color: var(--text-color-subdued) !important;
|
12 |
+
}
|
13 |
+
/* chatbot */
|
14 |
+
:root {
|
15 |
+
--bg-color-light: #F3F3F3;
|
16 |
+
--bg-color-dark: #121111;
|
17 |
+
}
|
18 |
+
|
19 |
+
@media (prefers-color-scheme: light) {
|
20 |
+
#chuanhu_chatbot {
|
21 |
+
background-color: var(--bg-color-light) !important;
|
22 |
+
}
|
23 |
+
[data-testid = "bot"] {
|
24 |
+
background-color: #FFFFFF !important;
|
25 |
+
}
|
26 |
+
[data-testid = "user"] {
|
27 |
+
background-color: #95EC69 !important;
|
28 |
+
}
|
29 |
+
}
|
30 |
+
|
31 |
+
@media (prefers-color-scheme: dark) {
|
32 |
+
#chuanhu_chatbot {
|
33 |
+
background-color: var(--bg-color-dark) !important;
|
34 |
+
}
|
35 |
+
[data-testid = "bot"] {
|
36 |
+
background-color: #2C2C2C !important;
|
37 |
+
}
|
38 |
+
[data-testid = "user"] {
|
39 |
+
background-color: #26B561 !important;
|
40 |
+
}
|
41 |
+
}
|
42 |
+
|
43 |
+
/* 屏幕宽度大于等于500px的设备 */
|
44 |
+
@media (min-width: 500px) {
|
45 |
+
#chuanhu_chatbot {
|
46 |
+
height: calc(100vh - 200px);
|
47 |
+
}
|
48 |
+
#chuanhu_chatbot .wrap {
|
49 |
+
max-height: calc(100vh - 200px - var(--line-sm)*1rem - 2*var(--block-label-margin) );
|
50 |
+
}
|
51 |
+
}
|
52 |
+
/* 屏幕宽度小于500px的设备 */
|
53 |
+
@media (max-width: 499px) {
|
54 |
+
#chuanhu_chatbot {
|
55 |
+
height: calc(100vh - 140px);
|
56 |
+
}
|
57 |
+
#chuanhu_chatbot .wrap {
|
58 |
+
max-height: calc(100vh - 140 - var(--line-sm)*1rem - 2*var(--block-label-margin) );
|
59 |
+
}
|
60 |
+
}
|
61 |
+
/* 对话气泡 */
|
62 |
+
[class *= "message"] {
|
63 |
+
border-radius: var(--radius-xl) !important;
|
64 |
+
border: none;
|
65 |
+
padding: var(--spacing-xl) !important;
|
66 |
+
font-size: var(--text-md) !important;
|
67 |
+
line-height: var(--line-md) !important;
|
68 |
+
}
|
69 |
+
[data-testid = "bot"] {
|
70 |
+
max-width: 85%;
|
71 |
+
border-bottom-left-radius: 0 !important;
|
72 |
+
}
|
73 |
+
[data-testid = "user"] {
|
74 |
+
max-width: 85%;
|
75 |
+
width: auto !important;
|
76 |
+
border-bottom-right-radius: 0 !important;
|
77 |
+
}
|
78 |
+
/* 表格 */
|
79 |
+
table {
|
80 |
+
margin: 1em 0;
|
81 |
+
border-collapse: collapse;
|
82 |
+
empty-cells: show;
|
83 |
+
}
|
84 |
+
td,th {
|
85 |
+
border: 1.2px solid var(--color-border-primary) !important;
|
86 |
+
padding: 0.2em;
|
87 |
+
}
|
88 |
+
thead {
|
89 |
+
background-color: rgba(175,184,193,0.2);
|
90 |
+
}
|
91 |
+
thead th {
|
92 |
+
padding: .5em .2em;
|
93 |
+
}
|
94 |
+
/* 行内代码 */
|
95 |
+
code {
|
96 |
+
display: inline;
|
97 |
+
white-space: break-spaces;
|
98 |
+
border-radius: 6px;
|
99 |
+
margin: 0 2px 0 2px;
|
100 |
+
padding: .2em .4em .1em .4em;
|
101 |
+
background-color: rgba(175,184,193,0.2);
|
102 |
+
}
|
103 |
+
/* 代码块 */
|
104 |
+
pre code {
|
105 |
+
display: block;
|
106 |
+
white-space: pre;
|
107 |
+
background-color: hsla(0, 0%, 0%, 80%)!important;
|
108 |
+
border-radius: 10px;
|
109 |
+
padding: 1rem 1.2rem 1rem;
|
110 |
+
margin: 1.2em 2em 1.2em 0.5em;
|
111 |
+
color: #FFF;
|
112 |
+
box-shadow: 6px 6px 16px hsla(0, 0%, 0%, 0.2);
|
113 |
+
}
|
114 |
+
/* 代码高亮样式 */
|
115 |
+
.codehilite .hll { background-color: #49483e }
|
116 |
+
.codehilite .c { color: #75715e } /* Comment */
|
117 |
+
.codehilite .err { color: #960050; background-color: #1e0010 } /* Error */
|
118 |
+
.codehilite .k { color: #66d9ef } /* Keyword */
|
119 |
+
.codehilite .l { color: #ae81ff } /* Literal */
|
120 |
+
.codehilite .n { color: #f8f8f2 } /* Name */
|
121 |
+
.codehilite .o { color: #f92672 } /* Operator */
|
122 |
+
.codehilite .p { color: #f8f8f2 } /* Punctuation */
|
123 |
+
.codehilite .ch { color: #75715e } /* Comment.Hashbang */
|
124 |
+
.codehilite .cm { color: #75715e } /* Comment.Multiline */
|
125 |
+
.codehilite .cp { color: #75715e } /* Comment.Preproc */
|
126 |
+
.codehilite .cpf { color: #75715e } /* Comment.PreprocFile */
|
127 |
+
.codehilite .c1 { color: #75715e } /* Comment.Single */
|
128 |
+
.codehilite .cs { color: #75715e } /* Comment.Special */
|
129 |
+
.codehilite .gd { color: #f92672 } /* Generic.Deleted */
|
130 |
+
.codehilite .ge { font-style: italic } /* Generic.Emph */
|
131 |
+
.codehilite .gi { color: #a6e22e } /* Generic.Inserted */
|
132 |
+
.codehilite .gs { font-weight: bold } /* Generic.Strong */
|
133 |
+
.codehilite .gu { color: #75715e } /* Generic.Subheading */
|
134 |
+
.codehilite .kc { color: #66d9ef } /* Keyword.Constant */
|
135 |
+
.codehilite .kd { color: #66d9ef } /* Keyword.Declaration */
|
136 |
+
.codehilite .kn { color: #f92672 } /* Keyword.Namespace */
|
137 |
+
.codehilite .kp { color: #66d9ef } /* Keyword.Pseudo */
|
138 |
+
.codehilite .kr { color: #66d9ef } /* Keyword.Reserved */
|
139 |
+
.codehilite .kt { color: #66d9ef } /* Keyword.Type */
|
140 |
+
.codehilite .ld { color: #e6db74 } /* Literal.Date */
|
141 |
+
.codehilite .m { color: #ae81ff } /* Literal.Number */
|
142 |
+
.codehilite .s { color: #e6db74 } /* Literal.String */
|
143 |
+
.codehilite .na { color: #a6e22e } /* Name.Attribute */
|
144 |
+
.codehilite .nb { color: #f8f8f2 } /* Name.Builtin */
|
145 |
+
.codehilite .nc { color: #a6e22e } /* Name.Class */
|
146 |
+
.codehilite .no { color: #66d9ef } /* Name.Constant */
|
147 |
+
.codehilite .nd { color: #a6e22e } /* Name.Decorator */
|
148 |
+
.codehilite .ni { color: #f8f8f2 } /* Name.Entity */
|
149 |
+
.codehilite .ne { color: #a6e22e } /* Name.Exception */
|
150 |
+
.codehilite .nf { color: #a6e22e } /* Name.Function */
|
151 |
+
.codehilite .nl { color: #f8f8f2 } /* Name.Label */
|
152 |
+
.codehilite .nn { color: #f8f8f2 } /* Name.Namespace */
|
153 |
+
.codehilite .nx { color: #a6e22e } /* Name.Other */
|
154 |
+
.codehilite .py { color: #f8f8f2 } /* Name.Property */
|
155 |
+
.codehilite .nt { color: #f92672 } /* Name.Tag */
|
156 |
+
.codehilite .nv { color: #f8f8f2 } /* Name.Variable */
|
157 |
+
.codehilite .ow { color: #f92672 } /* Operator.Word */
|
158 |
+
.codehilite .w { color: #f8f8f2 } /* Text.Whitespace */
|
159 |
+
.codehilite .mb { color: #ae81ff } /* Literal.Number.Bin */
|
160 |
+
.codehilite .mf { color: #ae81ff } /* Literal.Number.Float */
|
161 |
+
.codehilite .mh { color: #ae81ff } /* Literal.Number.Hex */
|
162 |
+
.codehilite .mi { color: #ae81ff } /* Literal.Number.Integer */
|
163 |
+
.codehilite .mo { color: #ae81ff } /* Literal.Number.Oct */
|
164 |
+
.codehilite .sa { color: #e6db74 } /* Literal.String.Affix */
|
165 |
+
.codehilite .sb { color: #e6db74 } /* Literal.String.Backtick */
|
166 |
+
.codehilite .sc { color: #e6db74 } /* Literal.String.Char */
|
167 |
+
.codehilite .dl { color: #e6db74 } /* Literal.String.Delimiter */
|
168 |
+
.codehilite .sd { color: #e6db74 } /* Literal.String.Doc */
|
169 |
+
.codehilite .s2 { color: #e6db74 } /* Literal.String.Double */
|
170 |
+
.codehilite .se { color: #ae81ff } /* Literal.String.Escape */
|
171 |
+
.codehilite .sh { color: #e6db74 } /* Literal.String.Heredoc */
|
172 |
+
.codehilite .si { color: #e6db74 } /* Literal.String.Interpol */
|
173 |
+
.codehilite .sx { color: #e6db74 } /* Literal.String.Other */
|
174 |
+
.codehilite .sr { color: #e6db74 } /* Literal.String.Regex */
|
175 |
+
.codehilite .s1 { color: #e6db74 } /* Literal.String.Single */
|
176 |
+
.codehilite .ss { color: #e6db74 } /* Literal.String.Symbol */
|
177 |
+
.codehilite .bp { color: #f8f8f2 } /* Name.Builtin.Pseudo */
|
178 |
+
.codehilite .fm { color: #a6e22e } /* Name.Function.Magic */
|
179 |
+
.codehilite .vc { color: #f8f8f2 } /* Name.Variable.Class */
|
180 |
+
.codehilite .vg { color: #f8f8f2 } /* Name.Variable.Global */
|
181 |
+
.codehilite .vi { color: #f8f8f2 } /* Name.Variable.Instance */
|
182 |
+
.codehilite .vm { color: #f8f8f2 } /* Name.Variable.Magic */
|
183 |
+
.codehilite .il { color: #ae81ff } /* Literal.Number.Integer.Long */
|
184 |
+
|
185 |
+
/* 全局元素 */
|
186 |
+
* {
|
187 |
+
transition: all 0.6s;
|
188 |
+
}
|
llama_func.py
ADDED
@@ -0,0 +1,192 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import logging
|
3 |
+
|
4 |
+
from llama_index import GPTSimpleVectorIndex
|
5 |
+
from llama_index import download_loader
|
6 |
+
from llama_index import (
|
7 |
+
Document,
|
8 |
+
LLMPredictor,
|
9 |
+
PromptHelper,
|
10 |
+
QuestionAnswerPrompt,
|
11 |
+
RefinePrompt,
|
12 |
+
)
|
13 |
+
from langchain.llms import OpenAI
|
14 |
+
import colorama
|
15 |
+
|
16 |
+
|
17 |
+
from presets import *
|
18 |
+
from utils import *
|
19 |
+
|
20 |
+
|
21 |
+
def get_documents(file_src):
|
22 |
+
documents = []
|
23 |
+
index_name = ""
|
24 |
+
logging.debug("Loading documents...")
|
25 |
+
logging.debug(f"file_src: {file_src}")
|
26 |
+
for file in file_src:
|
27 |
+
logging.debug(f"file: {file.name}")
|
28 |
+
index_name += file.name
|
29 |
+
if os.path.splitext(file.name)[1] == ".pdf":
|
30 |
+
logging.debug("Loading PDF...")
|
31 |
+
CJKPDFReader = download_loader("CJKPDFReader")
|
32 |
+
loader = CJKPDFReader()
|
33 |
+
documents += loader.load_data(file=file.name)
|
34 |
+
elif os.path.splitext(file.name)[1] == ".docx":
|
35 |
+
logging.debug("Loading DOCX...")
|
36 |
+
DocxReader = download_loader("DocxReader")
|
37 |
+
loader = DocxReader()
|
38 |
+
documents += loader.load_data(file=file.name)
|
39 |
+
elif os.path.splitext(file.name)[1] == ".epub":
|
40 |
+
logging.debug("Loading EPUB...")
|
41 |
+
EpubReader = download_loader("EpubReader")
|
42 |
+
loader = EpubReader()
|
43 |
+
documents += loader.load_data(file=file.name)
|
44 |
+
else:
|
45 |
+
logging.debug("Loading text file...")
|
46 |
+
with open(file.name, "r", encoding="utf-8") as f:
|
47 |
+
text = add_space(f.read())
|
48 |
+
documents += [Document(text)]
|
49 |
+
index_name = sha1sum(index_name)
|
50 |
+
return documents, index_name
|
51 |
+
|
52 |
+
|
53 |
+
def construct_index(
|
54 |
+
api_key,
|
55 |
+
file_src,
|
56 |
+
max_input_size=4096,
|
57 |
+
num_outputs=1,
|
58 |
+
max_chunk_overlap=20,
|
59 |
+
chunk_size_limit=600,
|
60 |
+
embedding_limit=None,
|
61 |
+
separator=" ",
|
62 |
+
num_children=10,
|
63 |
+
max_keywords_per_chunk=10,
|
64 |
+
):
|
65 |
+
os.environ["OPENAI_API_KEY"] = api_key
|
66 |
+
chunk_size_limit = None if chunk_size_limit == 0 else chunk_size_limit
|
67 |
+
embedding_limit = None if embedding_limit == 0 else embedding_limit
|
68 |
+
separator = " " if separator == "" else separator
|
69 |
+
|
70 |
+
llm_predictor = LLMPredictor(
|
71 |
+
llm=OpenAI(model_name="gpt-3.5-turbo-0301", openai_api_key=api_key)
|
72 |
+
)
|
73 |
+
prompt_helper = PromptHelper(
|
74 |
+
max_input_size,
|
75 |
+
num_outputs,
|
76 |
+
max_chunk_overlap,
|
77 |
+
embedding_limit,
|
78 |
+
chunk_size_limit,
|
79 |
+
separator=separator,
|
80 |
+
)
|
81 |
+
documents, index_name = get_documents(file_src)
|
82 |
+
if os.path.exists(f"./index/{index_name}.json"):
|
83 |
+
logging.info("找到了缓存的索引文件,加载中……")
|
84 |
+
return GPTSimpleVectorIndex.load_from_disk(f"./index/{index_name}.json")
|
85 |
+
else:
|
86 |
+
try:
|
87 |
+
logging.debug("构建索引中……")
|
88 |
+
index = GPTSimpleVectorIndex(
|
89 |
+
documents, llm_predictor=llm_predictor, prompt_helper=prompt_helper
|
90 |
+
)
|
91 |
+
os.makedirs("./index", exist_ok=True)
|
92 |
+
index.save_to_disk(f"./index/{index_name}.json")
|
93 |
+
return index
|
94 |
+
except Exception as e:
|
95 |
+
print(e)
|
96 |
+
return None
|
97 |
+
|
98 |
+
|
99 |
+
def chat_ai(
|
100 |
+
api_key,
|
101 |
+
index,
|
102 |
+
question,
|
103 |
+
context,
|
104 |
+
chatbot,
|
105 |
+
):
|
106 |
+
os.environ["OPENAI_API_KEY"] = api_key
|
107 |
+
|
108 |
+
logging.info(f"Question: {question}")
|
109 |
+
|
110 |
+
response, chatbot_display, status_text = ask_ai(
|
111 |
+
api_key,
|
112 |
+
index,
|
113 |
+
question,
|
114 |
+
replace_today(PROMPT_TEMPLATE),
|
115 |
+
REFINE_TEMPLATE,
|
116 |
+
SIM_K,
|
117 |
+
INDEX_QUERY_TEMPRATURE,
|
118 |
+
context,
|
119 |
+
)
|
120 |
+
if response is None:
|
121 |
+
status_text = "查询失败,请换个问法试试"
|
122 |
+
return context, chatbot
|
123 |
+
response = response
|
124 |
+
|
125 |
+
context.append({"role": "user", "content": question})
|
126 |
+
context.append({"role": "assistant", "content": response})
|
127 |
+
chatbot.append((question, chatbot_display))
|
128 |
+
|
129 |
+
os.environ["OPENAI_API_KEY"] = ""
|
130 |
+
return context, chatbot, status_text
|
131 |
+
|
132 |
+
|
133 |
+
def ask_ai(
|
134 |
+
api_key,
|
135 |
+
index,
|
136 |
+
question,
|
137 |
+
prompt_tmpl,
|
138 |
+
refine_tmpl,
|
139 |
+
sim_k=1,
|
140 |
+
temprature=0,
|
141 |
+
prefix_messages=[],
|
142 |
+
):
|
143 |
+
os.environ["OPENAI_API_KEY"] = api_key
|
144 |
+
|
145 |
+
logging.debug("Index file found")
|
146 |
+
logging.debug("Querying index...")
|
147 |
+
llm_predictor = LLMPredictor(
|
148 |
+
llm=OpenAI(
|
149 |
+
temperature=temprature,
|
150 |
+
model_name="gpt-3.5-turbo-0301",
|
151 |
+
prefix_messages=prefix_messages,
|
152 |
+
)
|
153 |
+
)
|
154 |
+
|
155 |
+
response = None # Initialize response variable to avoid UnboundLocalError
|
156 |
+
qa_prompt = QuestionAnswerPrompt(prompt_tmpl)
|
157 |
+
rf_prompt = RefinePrompt(refine_tmpl)
|
158 |
+
response = index.query(
|
159 |
+
question,
|
160 |
+
llm_predictor=llm_predictor,
|
161 |
+
similarity_top_k=sim_k,
|
162 |
+
text_qa_template=qa_prompt,
|
163 |
+
refine_template=rf_prompt,
|
164 |
+
response_mode="compact",
|
165 |
+
)
|
166 |
+
|
167 |
+
if response is not None:
|
168 |
+
logging.info(f"Response: {response}")
|
169 |
+
ret_text = response.response
|
170 |
+
nodes = []
|
171 |
+
for index, node in enumerate(response.source_nodes):
|
172 |
+
brief = node.source_text[:25].replace("\n", "")
|
173 |
+
nodes.append(
|
174 |
+
f"<details><summary>[{index+1}]\t{brief}...</summary><p>{node.source_text}</p></details>"
|
175 |
+
)
|
176 |
+
new_response = ret_text + "\n----------\n" + "\n\n".join(nodes)
|
177 |
+
logging.info(
|
178 |
+
f"Response: {colorama.Fore.BLUE}{ret_text}{colorama.Style.RESET_ALL}"
|
179 |
+
)
|
180 |
+
os.environ["OPENAI_API_KEY"] = ""
|
181 |
+
return ret_text, new_response, f"查询消耗了{llm_predictor.last_token_usage} tokens"
|
182 |
+
else:
|
183 |
+
logging.warning("No response found, returning None")
|
184 |
+
os.environ["OPENAI_API_KEY"] = ""
|
185 |
+
return None
|
186 |
+
|
187 |
+
|
188 |
+
def add_space(text):
|
189 |
+
punctuations = {",": ", ", "。": "。 ", "?": "? ", "!": "! ", ":": ": ", ";": "; "}
|
190 |
+
for cn_punc, en_punc in punctuations.items():
|
191 |
+
text = text.replace(cn_punc, en_punc)
|
192 |
+
return text
|
overwrites.py
ADDED
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from __future__ import annotations
|
2 |
+
import logging
|
3 |
+
|
4 |
+
from llama_index import Prompt
|
5 |
+
from typing import List, Tuple
|
6 |
+
import mdtex2html
|
7 |
+
|
8 |
+
from presets import *
|
9 |
+
from llama_func import *
|
10 |
+
|
11 |
+
|
12 |
+
def compact_text_chunks(self, prompt: Prompt, text_chunks: List[str]) -> List[str]:
|
13 |
+
logging.debug("Compacting text chunks...🚀🚀🚀")
|
14 |
+
combined_str = [c.strip() for c in text_chunks if c.strip()]
|
15 |
+
combined_str = [f"[{index+1}] {c}" for index, c in enumerate(combined_str)]
|
16 |
+
combined_str = "\n\n".join(combined_str)
|
17 |
+
# resplit based on self.max_chunk_overlap
|
18 |
+
text_splitter = self.get_text_splitter_given_prompt(prompt, 1, padding=1)
|
19 |
+
return text_splitter.split_text(combined_str)
|
20 |
+
|
21 |
+
|
22 |
+
def postprocess(
|
23 |
+
self, y: List[Tuple[str | None, str | None]]
|
24 |
+
) -> List[Tuple[str | None, str | None]]:
|
25 |
+
"""
|
26 |
+
Parameters:
|
27 |
+
y: List of tuples representing the message and response pairs. Each message and response should be a string, which may be in Markdown format.
|
28 |
+
Returns:
|
29 |
+
List of tuples representing the message and response. Each message and response will be a string of HTML.
|
30 |
+
"""
|
31 |
+
if y is None:
|
32 |
+
return []
|
33 |
+
for i, (message, response) in enumerate(y):
|
34 |
+
y[i] = (
|
35 |
+
# None if message is None else markdown.markdown(message),
|
36 |
+
# None if response is None else markdown.markdown(response),
|
37 |
+
None if message is None else message,
|
38 |
+
None if response is None else mdtex2html.convert(response, extensions=['fenced_code','codehilite','tables']),
|
39 |
+
)
|
40 |
+
return y
|
presets.py
CHANGED
@@ -1,6 +1,26 @@
|
|
1 |
# -*- coding:utf-8 -*-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
title = """<h1 align="left" style="min-width:200px; margin-top:0;">川虎ChatGPT 🚀</h1>"""
|
3 |
-
description = """
|
|
|
4 |
|
5 |
由Bilibili [土川虎虎虎](https://space.bilibili.com/29125536) 和 [明昭MZhao](https://space.bilibili.com/24807452)开发
|
6 |
|
@@ -9,62 +29,21 @@ description = """<div align="center" style="margin:16px 0">
|
|
9 |
此App使用 `gpt-3.5-turbo` 大语言模型
|
10 |
</div>
|
11 |
"""
|
12 |
-
customCSS = """
|
13 |
-
#status_display {
|
14 |
-
display: flex;
|
15 |
-
min-height: 2.5em;
|
16 |
-
align-items: flex-end;
|
17 |
-
justify-content: flex-end;
|
18 |
-
}
|
19 |
-
#status_display p {
|
20 |
-
font-size: .85em;
|
21 |
-
font-family: monospace;
|
22 |
-
color: var(--text-color-subdued) !important;
|
23 |
-
}
|
24 |
-
[class *= "message"] {
|
25 |
-
border-radius: var(--radius-xl) !important;
|
26 |
-
border: none;
|
27 |
-
padding: var(--spacing-xl) !important;
|
28 |
-
font-size: var(--text-md) !important;
|
29 |
-
line-height: var(--line-md) !important;
|
30 |
-
}
|
31 |
-
[data-testid = "bot"] {
|
32 |
-
max-width: 85%;
|
33 |
-
border-bottom-left-radius: 0 !important;
|
34 |
-
}
|
35 |
-
[data-testid = "user"] {
|
36 |
-
max-width: 85%;
|
37 |
-
width: auto !important;
|
38 |
-
border-bottom-right-radius: 0 !important;
|
39 |
-
}
|
40 |
-
code {
|
41 |
-
display: inline;
|
42 |
-
white-space: break-spaces;
|
43 |
-
border-radius: 6px;
|
44 |
-
margin: 0 2px 0 2px;
|
45 |
-
padding: .2em .4em .1em .4em;
|
46 |
-
background-color: rgba(175,184,193,0.2);
|
47 |
-
}
|
48 |
-
pre code {
|
49 |
-
display: block;
|
50 |
-
white-space: pre;
|
51 |
-
background-color: hsla(0, 0%, 0%, 72%);
|
52 |
-
border: solid 5px var(--color-border-primary) !important;
|
53 |
-
border-radius: 10px;
|
54 |
-
padding: 0 1.2rem 1.2rem;
|
55 |
-
margin-top: 1em !important;
|
56 |
-
color: #FFF;
|
57 |
-
box-shadow: inset 0px 8px 16px hsla(0, 0%, 0%, .2)
|
58 |
-
}
|
59 |
|
60 |
-
|
61 |
-
transition: all 0.6s;
|
62 |
-
}
|
63 |
-
"""
|
64 |
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
{web_results}
|
70 |
Current date: {current_date}
|
@@ -73,18 +52,29 @@ Instructions: Using the provided web search results, write a comprehensive reply
|
|
73 |
Query: {query}
|
74 |
Reply in 中文"""
|
75 |
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
|
|
|
|
|
|
|
|
|
|
84 |
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# -*- coding:utf-8 -*-
|
2 |
+
# 错误信息
|
3 |
+
standard_error_msg = "☹️发生了错误:" # 错误信息的标准前缀
|
4 |
+
error_retrieve_prompt = "请检查网络连接,或者API-Key是否有效。" # 获取对话时发生错误
|
5 |
+
connection_timeout_prompt = "连接超时,无法获取对话。" # 连接超时
|
6 |
+
read_timeout_prompt = "读取超时,无法获取对话。" # 读取超时
|
7 |
+
proxy_error_prompt = "代理错误,无法获取对话。" # 代理错误
|
8 |
+
ssl_error_prompt = "SSL错误,无法获取对话。" # SSL 错误
|
9 |
+
no_apikey_msg = "API key长度不是51位,请检查是否输入正确。" # API key 长度不足 51 位
|
10 |
+
|
11 |
+
max_token_streaming = 3500 # 流式对话时的最大 token 数
|
12 |
+
timeout_streaming = 30 # 流式对话时的超时时间
|
13 |
+
max_token_all = 3500 # 非流式对话时的最大 token 数
|
14 |
+
timeout_all = 200 # 非流式对话时的超时时间
|
15 |
+
enable_streaming_option = True # 是否启用选择选择是否实时显示回答的勾选框
|
16 |
+
HIDE_MY_KEY = False # 如果你想在UI中隐藏你的 API 密钥,将此值设置为 True
|
17 |
+
|
18 |
+
SIM_K = 5
|
19 |
+
INDEX_QUERY_TEMPRATURE = 1.0
|
20 |
+
|
21 |
title = """<h1 align="left" style="min-width:200px; margin-top:0;">川虎ChatGPT 🚀</h1>"""
|
22 |
+
description = """\
|
23 |
+
<div align="center" style="margin:16px 0">
|
24 |
|
25 |
由Bilibili [土川虎虎虎](https://space.bilibili.com/29125536) 和 [明昭MZhao](https://space.bilibili.com/24807452)开发
|
26 |
|
|
|
29 |
此App使用 `gpt-3.5-turbo` 大语言模型
|
30 |
</div>
|
31 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
+
summarize_prompt = "你是谁?我们刚才聊了什么?" # 总结对话时的 prompt
|
|
|
|
|
|
|
34 |
|
35 |
+
MODELS = [
|
36 |
+
"gpt-3.5-turbo",
|
37 |
+
"gpt-3.5-turbo-0301",
|
38 |
+
"gpt-4",
|
39 |
+
"gpt-4-0314",
|
40 |
+
"gpt-4-32k",
|
41 |
+
"gpt-4-32k-0314",
|
42 |
+
] # 可选的模型
|
43 |
+
|
44 |
+
|
45 |
+
WEBSEARCH_PTOMPT_TEMPLATE = """\
|
46 |
+
Web search results:
|
47 |
|
48 |
{web_results}
|
49 |
Current date: {current_date}
|
|
|
52 |
Query: {query}
|
53 |
Reply in 中文"""
|
54 |
|
55 |
+
PROMPT_TEMPLATE = """\
|
56 |
+
Context information is below.
|
57 |
+
---------------------
|
58 |
+
{context_str}
|
59 |
+
---------------------
|
60 |
+
Current date: {current_date}.
|
61 |
+
Using the provided context information, write a comprehensive reply to the given query.
|
62 |
+
Make sure to cite results using [number] notation after the reference.
|
63 |
+
If the provided context information refer to multiple subjects with the same name, write separate answers for each subject.
|
64 |
+
Use prior knowledge only if the given context didn't provide enough information.
|
65 |
+
Answer the question: {query_str}
|
66 |
+
Reply in 中文
|
67 |
+
"""
|
68 |
|
69 |
+
REFINE_TEMPLATE = """\
|
70 |
+
The original question is as follows: {query_str}
|
71 |
+
We have provided an existing answer: {existing_answer}
|
72 |
+
We have the opportunity to refine the existing answer
|
73 |
+
(only if needed) with some more context below.
|
74 |
+
------------
|
75 |
+
{context_msg}
|
76 |
+
------------
|
77 |
+
Given the new context, refine the original answer to better
|
78 |
+
Answer in the same language as the question, such as English, 中文, 日本語, Español, Français, or Deutsch.
|
79 |
+
If the context isn't useful, return the original answer.
|
80 |
+
"""
|
requirements.txt
CHANGED
@@ -6,3 +6,6 @@ socksio
|
|
6 |
tqdm
|
7 |
colorama
|
8 |
duckduckgo_search
|
|
|
|
|
|
|
|
6 |
tqdm
|
7 |
colorama
|
8 |
duckduckgo_search
|
9 |
+
Pygments
|
10 |
+
llama_index
|
11 |
+
langchain
|
utils.py
CHANGED
@@ -3,21 +3,16 @@ from __future__ import annotations
|
|
3 |
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Tuple, Type
|
4 |
import logging
|
5 |
import json
|
6 |
-
import gradio as gr
|
7 |
-
# import openai
|
8 |
import os
|
9 |
-
import
|
10 |
-
import
|
11 |
-
# import markdown
|
12 |
import csv
|
13 |
-
|
|
|
14 |
from pypinyin import lazy_pinyin
|
15 |
-
from presets import *
|
16 |
import tiktoken
|
17 |
-
|
18 |
-
import
|
19 |
-
from duckduckgo_search import ddg
|
20 |
-
import datetime
|
21 |
|
22 |
# logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s")
|
23 |
|
@@ -28,30 +23,12 @@ if TYPE_CHECKING:
|
|
28 |
headers: List[str]
|
29 |
data: List[List[str | int | bool]]
|
30 |
|
|
|
31 |
initial_prompt = "You are a helpful assistant."
|
32 |
API_URL = "https://api.openai.com/v1/chat/completions"
|
33 |
HISTORY_DIR = "history"
|
34 |
TEMPLATES_DIR = "templates"
|
35 |
|
36 |
-
def postprocess(
|
37 |
-
self, y: List[Tuple[str | None, str | None]]
|
38 |
-
) -> List[Tuple[str | None, str | None]]:
|
39 |
-
"""
|
40 |
-
Parameters:
|
41 |
-
y: List of tuples representing the message and response pairs. Each message and response should be a string, which may be in Markdown format.
|
42 |
-
Returns:
|
43 |
-
List of tuples representing the message and response. Each message and response will be a string of HTML.
|
44 |
-
"""
|
45 |
-
if y is None:
|
46 |
-
return []
|
47 |
-
for i, (message, response) in enumerate(y):
|
48 |
-
y[i] = (
|
49 |
-
# None if message is None else markdown.markdown(message),
|
50 |
-
# None if response is None else markdown.markdown(response),
|
51 |
-
None if message is None else message,
|
52 |
-
None if response is None else mdtex2html.convert(response),
|
53 |
-
)
|
54 |
-
return y
|
55 |
|
56 |
def count_token(message):
|
57 |
encoding = tiktoken.get_encoding("cl100k_base")
|
@@ -59,251 +36,43 @@ def count_token(message):
|
|
59 |
length = len(encoding.encode(input_str))
|
60 |
return length
|
61 |
|
|
|
62 |
def parse_text(text):
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
lines[i] = f'<pre><code class="language-{items[-1]}">'
|
72 |
-
else:
|
73 |
-
lines[i] = f'<br></code></pre>'
|
74 |
else:
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
line = line.replace(">", ">")
|
80 |
-
line = line.replace(" ", " ")
|
81 |
-
line = line.replace("*", "*")
|
82 |
-
line = line.replace("_", "_")
|
83 |
-
line = line.replace("-", "-")
|
84 |
-
line = line.replace(".", ".")
|
85 |
-
line = line.replace("!", "!")
|
86 |
-
line = line.replace("(", "(")
|
87 |
-
line = line.replace(")", ")")
|
88 |
-
line = line.replace("$", "$")
|
89 |
-
lines[i] = "<br>"+line
|
90 |
-
text = "".join(lines)
|
91 |
return text
|
92 |
|
|
|
93 |
def construct_text(role, text):
|
94 |
return {"role": role, "content": text}
|
95 |
|
|
|
96 |
def construct_user(text):
|
97 |
return construct_text("user", text)
|
98 |
|
|
|
99 |
def construct_system(text):
|
100 |
return construct_text("system", text)
|
101 |
|
|
|
102 |
def construct_assistant(text):
|
103 |
return construct_text("assistant", text)
|
104 |
|
|
|
105 |
def construct_token_message(token, stream=False):
|
106 |
return f"Token 计数: {token}"
|
107 |
|
108 |
-
def get_response(openai_api_key, system_prompt, history, temperature, top_p, stream, selected_model):
|
109 |
-
headers = {
|
110 |
-
"Content-Type": "application/json",
|
111 |
-
"Authorization": f"Bearer {openai_api_key}"
|
112 |
-
}
|
113 |
-
|
114 |
-
history = [construct_system(system_prompt), *history]
|
115 |
-
|
116 |
-
payload = {
|
117 |
-
"model": selected_model,
|
118 |
-
"messages": history, # [{"role": "user", "content": f"{inputs}"}],
|
119 |
-
"temperature": temperature, # 1.0,
|
120 |
-
"top_p": top_p, # 1.0,
|
121 |
-
"n": 1,
|
122 |
-
"stream": stream,
|
123 |
-
"presence_penalty": 0,
|
124 |
-
"frequency_penalty": 0,
|
125 |
-
}
|
126 |
-
if stream:
|
127 |
-
timeout = timeout_streaming
|
128 |
-
else:
|
129 |
-
timeout = timeout_all
|
130 |
-
response = requests.post(API_URL, headers=headers, json=payload, stream=True, timeout=timeout)
|
131 |
-
return response
|
132 |
-
|
133 |
-
def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model):
|
134 |
-
def get_return_value():
|
135 |
-
return chatbot, history, status_text, all_token_counts
|
136 |
-
|
137 |
-
logging.info("实时回答模式")
|
138 |
-
partial_words = ""
|
139 |
-
counter = 0
|
140 |
-
status_text = "开始实时传输回答……"
|
141 |
-
history.append(construct_user(inputs))
|
142 |
-
history.append(construct_assistant(""))
|
143 |
-
chatbot.append((parse_text(inputs), ""))
|
144 |
-
user_token_count = 0
|
145 |
-
if len(all_token_counts) == 0:
|
146 |
-
system_prompt_token_count = count_token(construct_system(system_prompt))
|
147 |
-
user_token_count = count_token(construct_user(inputs)) + system_prompt_token_count
|
148 |
-
else:
|
149 |
-
user_token_count = count_token(construct_user(inputs))
|
150 |
-
all_token_counts.append(user_token_count)
|
151 |
-
logging.info(f"输入token计数: {user_token_count}")
|
152 |
-
yield get_return_value()
|
153 |
-
try:
|
154 |
-
response = get_response(openai_api_key, system_prompt, history, temperature, top_p, True, selected_model)
|
155 |
-
except requests.exceptions.ConnectTimeout:
|
156 |
-
status_text = standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
|
157 |
-
yield get_return_value()
|
158 |
-
return
|
159 |
-
except requests.exceptions.ReadTimeout:
|
160 |
-
status_text = standard_error_msg + read_timeout_prompt + error_retrieve_prompt
|
161 |
-
yield get_return_value()
|
162 |
-
return
|
163 |
-
|
164 |
-
yield get_return_value()
|
165 |
-
error_json_str = ""
|
166 |
-
|
167 |
-
for chunk in tqdm(response.iter_lines()):
|
168 |
-
if counter == 0:
|
169 |
-
counter += 1
|
170 |
-
continue
|
171 |
-
counter += 1
|
172 |
-
# check whether each line is non-empty
|
173 |
-
if chunk:
|
174 |
-
chunk = chunk.decode()
|
175 |
-
chunklength = len(chunk)
|
176 |
-
try:
|
177 |
-
chunk = json.loads(chunk[6:])
|
178 |
-
except json.JSONDecodeError:
|
179 |
-
logging.info(chunk)
|
180 |
-
error_json_str += chunk
|
181 |
-
status_text = f"JSON解析错误。请重置对话。收到的内容: {error_json_str}"
|
182 |
-
yield get_return_value()
|
183 |
-
continue
|
184 |
-
# decode each line as response data is in bytes
|
185 |
-
if chunklength > 6 and "delta" in chunk['choices'][0]:
|
186 |
-
finish_reason = chunk['choices'][0]['finish_reason']
|
187 |
-
status_text = construct_token_message(sum(all_token_counts), stream=True)
|
188 |
-
if finish_reason == "stop":
|
189 |
-
yield get_return_value()
|
190 |
-
break
|
191 |
-
try:
|
192 |
-
partial_words = partial_words + chunk['choices'][0]["delta"]["content"]
|
193 |
-
except KeyError:
|
194 |
-
status_text = standard_error_msg + "API回复中找不到内容。很可能是Token计数达到上限了。请重置对话。当前Token计数: " + str(sum(all_token_counts))
|
195 |
-
yield get_return_value()
|
196 |
-
break
|
197 |
-
history[-1] = construct_assistant(partial_words)
|
198 |
-
chatbot[-1] = (parse_text(inputs), parse_text(partial_words))
|
199 |
-
all_token_counts[-1] += 1
|
200 |
-
yield get_return_value()
|
201 |
-
|
202 |
-
|
203 |
-
def predict_all(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model):
|
204 |
-
logging.info("一次性回答模式")
|
205 |
-
history.append(construct_user(inputs))
|
206 |
-
history.append(construct_assistant(""))
|
207 |
-
chatbot.append((parse_text(inputs), ""))
|
208 |
-
all_token_counts.append(count_token(construct_user(inputs)))
|
209 |
-
try:
|
210 |
-
response = get_response(openai_api_key, system_prompt, history, temperature, top_p, False, selected_model)
|
211 |
-
except requests.exceptions.ConnectTimeout:
|
212 |
-
status_text = standard_error_msg + connection_timeout_prompt + error_retrieve_prompt
|
213 |
-
return chatbot, history, status_text, all_token_counts
|
214 |
-
except requests.exceptions.ProxyError:
|
215 |
-
status_text = standard_error_msg + proxy_error_prompt + error_retrieve_prompt
|
216 |
-
return chatbot, history, status_text, all_token_counts
|
217 |
-
except requests.exceptions.SSLError:
|
218 |
-
status_text = standard_error_msg + ssl_error_prompt + error_retrieve_prompt
|
219 |
-
return chatbot, history, status_text, all_token_counts
|
220 |
-
response = json.loads(response.text)
|
221 |
-
content = response["choices"][0]["message"]["content"]
|
222 |
-
history[-1] = construct_assistant(content)
|
223 |
-
chatbot[-1] = (parse_text(inputs), parse_text(content))
|
224 |
-
total_token_count = response["usage"]["total_tokens"]
|
225 |
-
all_token_counts[-1] = total_token_count - sum(all_token_counts)
|
226 |
-
status_text = construct_token_message(total_token_count)
|
227 |
-
return chatbot, history, status_text, all_token_counts
|
228 |
-
|
229 |
-
|
230 |
-
def predict(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, stream=False, selected_model = MODELS[0], use_websearch_checkbox = False, should_check_token_count = True): # repetition_penalty, top_k
|
231 |
-
logging.info("输入为:" +colorama.Fore.BLUE + f"{inputs}" + colorama.Style.RESET_ALL)
|
232 |
-
if use_websearch_checkbox:
|
233 |
-
results = ddg(inputs, max_results=3)
|
234 |
-
web_results = []
|
235 |
-
for idx, result in enumerate(results):
|
236 |
-
logging.info(f"搜索结果{idx + 1}:{result}")
|
237 |
-
web_results.append(f'[{idx+1}]"{result["body"]}"\nURL: {result["href"]}')
|
238 |
-
web_results = "\n\n".join(web_results)
|
239 |
-
today = datetime.datetime.today().strftime("%Y-%m-%d")
|
240 |
-
inputs = websearch_prompt.replace("{current_date}", today).replace("{query}", inputs).replace("{web_results}", web_results)
|
241 |
-
if len(openai_api_key) != 51:
|
242 |
-
status_text = standard_error_msg + no_apikey_msg
|
243 |
-
logging.info(status_text)
|
244 |
-
chatbot.append((parse_text(inputs), ""))
|
245 |
-
if len(history) == 0:
|
246 |
-
history.append(construct_user(inputs))
|
247 |
-
history.append("")
|
248 |
-
all_token_counts.append(0)
|
249 |
-
else:
|
250 |
-
history[-2] = construct_user(inputs)
|
251 |
-
yield chatbot, history, status_text, all_token_counts
|
252 |
-
return
|
253 |
-
if stream:
|
254 |
-
yield chatbot, history, "开始生成回答……", all_token_counts
|
255 |
-
if stream:
|
256 |
-
logging.info("使用流式传输")
|
257 |
-
iter = stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model)
|
258 |
-
for chatbot, history, status_text, all_token_counts in iter:
|
259 |
-
yield chatbot, history, status_text, all_token_counts
|
260 |
-
else:
|
261 |
-
logging.info("不使用流式传输")
|
262 |
-
chatbot, history, status_text, all_token_counts = predict_all(openai_api_key, system_prompt, history, inputs, chatbot, all_token_counts, top_p, temperature, selected_model)
|
263 |
-
yield chatbot, history, status_text, all_token_counts
|
264 |
-
logging.info(f"传输完毕。当前token计数为{all_token_counts}")
|
265 |
-
if len(history) > 1 and history[-1]['content'] != inputs:
|
266 |
-
logging.info("回答为:" +colorama.Fore.BLUE + f"{history[-1]['content']}" + colorama.Style.RESET_ALL)
|
267 |
-
if stream:
|
268 |
-
max_token = max_token_streaming
|
269 |
-
else:
|
270 |
-
max_token = max_token_all
|
271 |
-
if sum(all_token_counts) > max_token and should_check_token_count:
|
272 |
-
status_text = f"精简token中{all_token_counts}/{max_token}"
|
273 |
-
logging.info(status_text)
|
274 |
-
yield chatbot, history, status_text, all_token_counts
|
275 |
-
iter = reduce_token_size(openai_api_key, system_prompt, history, chatbot, all_token_counts, top_p, temperature, stream=False, selected_model=selected_model, hidden=True)
|
276 |
-
for chatbot, history, status_text, all_token_counts in iter:
|
277 |
-
status_text = f"Token 达到上限,已自动降低Token计数至 {status_text}"
|
278 |
-
yield chatbot, history, status_text, all_token_counts
|
279 |
-
|
280 |
-
|
281 |
-
def retry(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False, selected_model = MODELS[0]):
|
282 |
-
logging.info("重试中……")
|
283 |
-
if len(history) == 0:
|
284 |
-
yield chatbot, history, f"{standard_error_msg}上下文是空的", token_count
|
285 |
-
return
|
286 |
-
history.pop()
|
287 |
-
inputs = history.pop()["content"]
|
288 |
-
token_count.pop()
|
289 |
-
iter = predict(openai_api_key, system_prompt, history, inputs, chatbot, token_count, top_p, temperature, stream=stream, selected_model=selected_model)
|
290 |
-
logging.info("重试完毕")
|
291 |
-
for x in iter:
|
292 |
-
yield x
|
293 |
-
|
294 |
-
|
295 |
-
def reduce_token_size(openai_api_key, system_prompt, history, chatbot, token_count, top_p, temperature, stream=False, selected_model = MODELS[0], hidden=False):
|
296 |
-
logging.info("开始减少token数量……")
|
297 |
-
iter = predict(openai_api_key, system_prompt, history, summarize_prompt, chatbot, token_count, top_p, temperature, stream=stream, selected_model = selected_model, should_check_token_count=False)
|
298 |
-
logging.info(f"chatbot: {chatbot}")
|
299 |
-
for chatbot, history, status_text, previous_token_count in iter:
|
300 |
-
history = history[-2:]
|
301 |
-
token_count = previous_token_count[-1:]
|
302 |
-
if hidden:
|
303 |
-
chatbot.pop()
|
304 |
-
yield chatbot, history, construct_token_message(sum(token_count), stream=stream), token_count
|
305 |
-
logging.info("减少token数量完毕")
|
306 |
-
|
307 |
|
308 |
def delete_last_conversation(chatbot, history, previous_token_count):
|
309 |
if len(chatbot) > 0 and standard_error_msg in chatbot[-1][1]:
|
@@ -320,7 +89,12 @@ def delete_last_conversation(chatbot, history, previous_token_count):
|
|
320 |
if len(previous_token_count) > 0:
|
321 |
logging.info("删除了一组对话的token计数记录")
|
322 |
previous_token_count.pop()
|
323 |
-
return
|
|
|
|
|
|
|
|
|
|
|
324 |
|
325 |
|
326 |
def save_file(filename, system, history, chatbot):
|
@@ -340,6 +114,7 @@ def save_file(filename, system, history, chatbot):
|
|
340 |
logging.info("保存对话历史完毕")
|
341 |
return os.path.join(HISTORY_DIR, filename)
|
342 |
|
|
|
343 |
def save_chat_history(filename, system, history, chatbot):
|
344 |
if filename == "":
|
345 |
return
|
@@ -347,6 +122,7 @@ def save_chat_history(filename, system, history, chatbot):
|
|
347 |
filename += ".json"
|
348 |
return save_file(filename, system, history, chatbot)
|
349 |
|
|
|
350 |
def export_markdown(filename, system, history, chatbot):
|
351 |
if filename == "":
|
352 |
return
|
@@ -382,9 +158,11 @@ def load_chat_history(filename, system, history, chatbot):
|
|
382 |
logging.info("没有找到对话历史文件,不执行任何操作")
|
383 |
return filename, system, history, chatbot
|
384 |
|
|
|
385 |
def sorted_by_pinyin(list):
|
386 |
return sorted(list, key=lambda char: lazy_pinyin(char)[0][0])
|
387 |
|
|
|
388 |
def get_file_names(dir, plain=False, filetypes=[".json"]):
|
389 |
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}")
|
390 |
files = []
|
@@ -401,10 +179,12 @@ def get_file_names(dir, plain=False, filetypes=[".json"]):
|
|
401 |
else:
|
402 |
return gr.Dropdown.update(choices=files)
|
403 |
|
|
|
404 |
def get_history_names(plain=False):
|
405 |
logging.info("获取历史记录文件名列表")
|
406 |
return get_file_names(HISTORY_DIR, plain)
|
407 |
|
|
|
408 |
def load_template(filename, mode=0):
|
409 |
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)")
|
410 |
lines = []
|
@@ -414,22 +194,28 @@ def load_template(filename, mode=0):
|
|
414 |
lines = json.load(f)
|
415 |
lines = [[i["act"], i["prompt"]] for i in lines]
|
416 |
else:
|
417 |
-
with open(
|
|
|
|
|
418 |
reader = csv.reader(csvfile)
|
419 |
lines = list(reader)
|
420 |
lines = lines[1:]
|
421 |
if mode == 1:
|
422 |
return sorted_by_pinyin([row[0] for row in lines])
|
423 |
elif mode == 2:
|
424 |
-
return {row[0]:row[1] for row in lines}
|
425 |
else:
|
426 |
choices = sorted_by_pinyin([row[0] for row in lines])
|
427 |
-
return {row[0]:row[1] for row in lines}, gr.Dropdown.update(
|
|
|
|
|
|
|
428 |
|
429 |
def get_template_names(plain=False):
|
430 |
logging.info("获取模板文件名列表")
|
431 |
return get_file_names(TEMPLATES_DIR, plain, filetypes=[".csv", "json"])
|
432 |
|
|
|
433 |
def get_template_content(templates, selection, original_system_prompt):
|
434 |
logging.info(f"应用模板中,选择为{selection},原始系统提示为{original_system_prompt}")
|
435 |
try:
|
@@ -437,9 +223,62 @@ def get_template_content(templates, selection, original_system_prompt):
|
|
437 |
except:
|
438 |
return original_system_prompt
|
439 |
|
|
|
440 |
def reset_state():
|
441 |
logging.info("重置状态")
|
442 |
return [], [], [], construct_token_message(0)
|
443 |
|
|
|
444 |
def reset_textbox():
|
445 |
-
return gr.update(value=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Tuple, Type
|
4 |
import logging
|
5 |
import json
|
|
|
|
|
6 |
import os
|
7 |
+
import datetime
|
8 |
+
import hashlib
|
|
|
9 |
import csv
|
10 |
+
|
11 |
+
import gradio as gr
|
12 |
from pypinyin import lazy_pinyin
|
|
|
13 |
import tiktoken
|
14 |
+
|
15 |
+
from presets import *
|
|
|
|
|
16 |
|
17 |
# logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s")
|
18 |
|
|
|
23 |
headers: List[str]
|
24 |
data: List[List[str | int | bool]]
|
25 |
|
26 |
+
|
27 |
initial_prompt = "You are a helpful assistant."
|
28 |
API_URL = "https://api.openai.com/v1/chat/completions"
|
29 |
HISTORY_DIR = "history"
|
30 |
TEMPLATES_DIR = "templates"
|
31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
def count_token(message):
|
34 |
encoding = tiktoken.get_encoding("cl100k_base")
|
|
|
36 |
length = len(encoding.encode(input_str))
|
37 |
return length
|
38 |
|
39 |
+
|
40 |
def parse_text(text):
|
41 |
+
in_code_block = False
|
42 |
+
new_lines = []
|
43 |
+
for line in text.split("\n"):
|
44 |
+
if line.strip().startswith("```"):
|
45 |
+
in_code_block = not in_code_block
|
46 |
+
if in_code_block:
|
47 |
+
if line.strip() != "":
|
48 |
+
new_lines.append(line)
|
|
|
|
|
|
|
49 |
else:
|
50 |
+
new_lines.append(line)
|
51 |
+
if in_code_block:
|
52 |
+
new_lines.append("```")
|
53 |
+
text = "\n".join(new_lines)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
return text
|
55 |
|
56 |
+
|
57 |
def construct_text(role, text):
|
58 |
return {"role": role, "content": text}
|
59 |
|
60 |
+
|
61 |
def construct_user(text):
|
62 |
return construct_text("user", text)
|
63 |
|
64 |
+
|
65 |
def construct_system(text):
|
66 |
return construct_text("system", text)
|
67 |
|
68 |
+
|
69 |
def construct_assistant(text):
|
70 |
return construct_text("assistant", text)
|
71 |
|
72 |
+
|
73 |
def construct_token_message(token, stream=False):
|
74 |
return f"Token 计数: {token}"
|
75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
def delete_last_conversation(chatbot, history, previous_token_count):
|
78 |
if len(chatbot) > 0 and standard_error_msg in chatbot[-1][1]:
|
|
|
89 |
if len(previous_token_count) > 0:
|
90 |
logging.info("删除了一组对话的token计数记录")
|
91 |
previous_token_count.pop()
|
92 |
+
return (
|
93 |
+
chatbot,
|
94 |
+
history,
|
95 |
+
previous_token_count,
|
96 |
+
construct_token_message(sum(previous_token_count)),
|
97 |
+
)
|
98 |
|
99 |
|
100 |
def save_file(filename, system, history, chatbot):
|
|
|
114 |
logging.info("保存对话历史完毕")
|
115 |
return os.path.join(HISTORY_DIR, filename)
|
116 |
|
117 |
+
|
118 |
def save_chat_history(filename, system, history, chatbot):
|
119 |
if filename == "":
|
120 |
return
|
|
|
122 |
filename += ".json"
|
123 |
return save_file(filename, system, history, chatbot)
|
124 |
|
125 |
+
|
126 |
def export_markdown(filename, system, history, chatbot):
|
127 |
if filename == "":
|
128 |
return
|
|
|
158 |
logging.info("没有找到对话历史文件,不执行任何操作")
|
159 |
return filename, system, history, chatbot
|
160 |
|
161 |
+
|
162 |
def sorted_by_pinyin(list):
|
163 |
return sorted(list, key=lambda char: lazy_pinyin(char)[0][0])
|
164 |
|
165 |
+
|
166 |
def get_file_names(dir, plain=False, filetypes=[".json"]):
|
167 |
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}")
|
168 |
files = []
|
|
|
179 |
else:
|
180 |
return gr.Dropdown.update(choices=files)
|
181 |
|
182 |
+
|
183 |
def get_history_names(plain=False):
|
184 |
logging.info("获取历史记录文件名列表")
|
185 |
return get_file_names(HISTORY_DIR, plain)
|
186 |
|
187 |
+
|
188 |
def load_template(filename, mode=0):
|
189 |
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)")
|
190 |
lines = []
|
|
|
194 |
lines = json.load(f)
|
195 |
lines = [[i["act"], i["prompt"]] for i in lines]
|
196 |
else:
|
197 |
+
with open(
|
198 |
+
os.path.join(TEMPLATES_DIR, filename), "r", encoding="utf8"
|
199 |
+
) as csvfile:
|
200 |
reader = csv.reader(csvfile)
|
201 |
lines = list(reader)
|
202 |
lines = lines[1:]
|
203 |
if mode == 1:
|
204 |
return sorted_by_pinyin([row[0] for row in lines])
|
205 |
elif mode == 2:
|
206 |
+
return {row[0]: row[1] for row in lines}
|
207 |
else:
|
208 |
choices = sorted_by_pinyin([row[0] for row in lines])
|
209 |
+
return {row[0]: row[1] for row in lines}, gr.Dropdown.update(
|
210 |
+
choices=choices, value=choices[0]
|
211 |
+
)
|
212 |
+
|
213 |
|
214 |
def get_template_names(plain=False):
|
215 |
logging.info("获取模板文件名列表")
|
216 |
return get_file_names(TEMPLATES_DIR, plain, filetypes=[".csv", "json"])
|
217 |
|
218 |
+
|
219 |
def get_template_content(templates, selection, original_system_prompt):
|
220 |
logging.info(f"应用模板中,选择为{selection},原始系统提示为{original_system_prompt}")
|
221 |
try:
|
|
|
223 |
except:
|
224 |
return original_system_prompt
|
225 |
|
226 |
+
|
227 |
def reset_state():
|
228 |
logging.info("重置状态")
|
229 |
return [], [], [], construct_token_message(0)
|
230 |
|
231 |
+
|
232 |
def reset_textbox():
|
233 |
+
return gr.update(value="")
|
234 |
+
|
235 |
+
|
236 |
+
def reset_default():
|
237 |
+
global API_URL
|
238 |
+
API_URL = "https://api.openai.com/v1/chat/completions"
|
239 |
+
os.environ.pop("HTTPS_PROXY", None)
|
240 |
+
os.environ.pop("https_proxy", None)
|
241 |
+
return gr.update(value=API_URL), gr.update(value=""), "API URL 和代理已重置"
|
242 |
+
|
243 |
+
|
244 |
+
def change_api_url(url):
|
245 |
+
global API_URL
|
246 |
+
API_URL = url
|
247 |
+
msg = f"API地址更改为了{url}"
|
248 |
+
logging.info(msg)
|
249 |
+
return msg
|
250 |
+
|
251 |
+
|
252 |
+
def change_proxy(proxy):
|
253 |
+
os.environ["HTTPS_PROXY"] = proxy
|
254 |
+
msg = f"代理更改为了{proxy}"
|
255 |
+
logging.info(msg)
|
256 |
+
return msg
|
257 |
+
|
258 |
+
|
259 |
+
def hide_middle_chars(s):
|
260 |
+
if len(s) <= 8:
|
261 |
+
return s
|
262 |
+
else:
|
263 |
+
head = s[:4]
|
264 |
+
tail = s[-4:]
|
265 |
+
hidden = "*" * (len(s) - 8)
|
266 |
+
return head + hidden + tail
|
267 |
+
|
268 |
+
|
269 |
+
def submit_key(key):
|
270 |
+
key = key.strip()
|
271 |
+
msg = f"API密钥更改为了{hide_middle_chars(key)}"
|
272 |
+
logging.info(msg)
|
273 |
+
return key, msg
|
274 |
+
|
275 |
+
|
276 |
+
def sha1sum(filename):
|
277 |
+
sha1 = hashlib.sha1()
|
278 |
+
sha1.update(filename.encode("utf-8"))
|
279 |
+
return sha1.hexdigest()
|
280 |
+
|
281 |
+
|
282 |
+
def replace_today(prompt):
|
283 |
+
today = datetime.datetime.today().strftime("%Y-%m-%d")
|
284 |
+
return prompt.replace("{current_date}", today)
|