Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,265 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
#import urllib.request
|
3 |
+
import requests
|
4 |
+
#import bs4
|
5 |
+
#import lxml
|
6 |
+
import os
|
7 |
+
#import subprocess
|
8 |
+
from huggingface_hub import InferenceClient,HfApi
|
9 |
+
import random
|
10 |
+
import json
|
11 |
+
import datetime
|
12 |
+
#from pypdf import PdfReader
|
13 |
+
import uuid
|
14 |
+
#from query import tasks
|
15 |
+
from agent import (
|
16 |
+
PREFIX,
|
17 |
+
SAVE_MEMORY,
|
18 |
+
COMPRESS_DATA_PROMPT,
|
19 |
+
COMPRESS_DATA_PROMPT_SMALL,
|
20 |
+
LOG_PROMPT,
|
21 |
+
LOG_RESPONSE,
|
22 |
+
)
|
23 |
+
client = InferenceClient(
|
24 |
+
"mistralai/Mixtral-8x7B-Instruct-v0.1"
|
25 |
+
)
|
26 |
+
reponame="Omnibus/tmp"
|
27 |
+
save_data=f'https://huggingface.co/datasets/{reponame}/raw/main/'
|
28 |
+
#token_self = os.environ['HF_TOKEN']
|
29 |
+
#api=HfApi(token=token_self)
|
30 |
+
|
31 |
+
|
32 |
+
|
33 |
+
VERBOSE = True
|
34 |
+
MAX_HISTORY = 100
|
35 |
+
MAX_DATA = 20000
|
36 |
+
|
37 |
+
def format_prompt(message, history):
|
38 |
+
prompt = "<s>"
|
39 |
+
for user_prompt, bot_response in history:
|
40 |
+
prompt += f"[INST] {user_prompt} [/INST]"
|
41 |
+
prompt += f" {bot_response}</s> "
|
42 |
+
prompt += f"[INST] {message} [/INST]"
|
43 |
+
return prompt
|
44 |
+
|
45 |
+
|
46 |
+
|
47 |
+
def run_gpt(
|
48 |
+
prompt_template,
|
49 |
+
stop_tokens,
|
50 |
+
max_tokens,
|
51 |
+
seed,
|
52 |
+
**prompt_kwargs,
|
53 |
+
):
|
54 |
+
print(seed)
|
55 |
+
timestamp=datetime.datetime.now()
|
56 |
+
|
57 |
+
generate_kwargs = dict(
|
58 |
+
temperature=0.9,
|
59 |
+
max_new_tokens=max_tokens,
|
60 |
+
top_p=0.95,
|
61 |
+
repetition_penalty=1.0,
|
62 |
+
do_sample=True,
|
63 |
+
seed=seed,
|
64 |
+
)
|
65 |
+
|
66 |
+
content = PREFIX.format(
|
67 |
+
timestamp=timestamp,
|
68 |
+
purpose="Compile the provided data and complete the users task"
|
69 |
+
) + prompt_template.format(**prompt_kwargs)
|
70 |
+
if VERBOSE:
|
71 |
+
print(LOG_PROMPT.format(content))
|
72 |
+
|
73 |
+
|
74 |
+
#formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history)
|
75 |
+
#formatted_prompt = format_prompt(f'{content}', history)
|
76 |
+
|
77 |
+
stream = client.text_generation(content, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
78 |
+
resp = ""
|
79 |
+
for response in stream:
|
80 |
+
resp += response.token.text
|
81 |
+
#yield resp
|
82 |
+
|
83 |
+
if VERBOSE:
|
84 |
+
print(LOG_RESPONSE.format(resp))
|
85 |
+
return resp
|
86 |
+
|
87 |
+
|
88 |
+
def compress_data(c, instruct, history):
|
89 |
+
seed=random.randint(1,1000000000)
|
90 |
+
|
91 |
+
print (c)
|
92 |
+
#tot=len(purpose)
|
93 |
+
#print(tot)
|
94 |
+
divr=int(c)/MAX_DATA
|
95 |
+
divi=int(divr)+1 if divr != int(divr) else int(divr)
|
96 |
+
chunk = int(int(c)/divr)
|
97 |
+
print(f'chunk:: {chunk}')
|
98 |
+
print(f'divr:: {divr}')
|
99 |
+
print (f'divi:: {divi}')
|
100 |
+
out = []
|
101 |
+
#out=""
|
102 |
+
s=0
|
103 |
+
e=chunk
|
104 |
+
print(f'e:: {e}')
|
105 |
+
new_history=""
|
106 |
+
#task = f'Compile this data to fulfill the task: {task}, and complete the purpose: {purpose}\n'
|
107 |
+
for z in range(divi):
|
108 |
+
print(f's:e :: {s}:{e}')
|
109 |
+
|
110 |
+
hist = history[s:e]
|
111 |
+
|
112 |
+
resp = run_gpt(
|
113 |
+
COMPRESS_DATA_PROMPT_SMALL,
|
114 |
+
stop_tokens=["observation:", "task:", "action:", "thought:"],
|
115 |
+
max_tokens=8192,
|
116 |
+
seed=seed,
|
117 |
+
direction=instruct,
|
118 |
+
knowledge="",
|
119 |
+
history=hist,
|
120 |
+
)
|
121 |
+
out.append(resp)
|
122 |
+
#new_history = resp
|
123 |
+
print (resp)
|
124 |
+
#out+=resp
|
125 |
+
e=e+chunk
|
126 |
+
s=s+chunk
|
127 |
+
return out
|
128 |
+
|
129 |
+
|
130 |
+
def compress_data_og(c, instruct, history):
|
131 |
+
seed=random.randint(1,1000000000)
|
132 |
+
|
133 |
+
print (c)
|
134 |
+
#tot=len(purpose)
|
135 |
+
#print(tot)
|
136 |
+
divr=int(c)/MAX_DATA
|
137 |
+
divi=int(divr)+1 if divr != int(divr) else int(divr)
|
138 |
+
chunk = int(int(c)/divr)
|
139 |
+
print(f'chunk:: {chunk}')
|
140 |
+
print(f'divr:: {divr}')
|
141 |
+
print (f'divi:: {divi}')
|
142 |
+
out = []
|
143 |
+
#out=""
|
144 |
+
s=0
|
145 |
+
e=chunk
|
146 |
+
print(f'e:: {e}')
|
147 |
+
new_history=""
|
148 |
+
#task = f'Compile this data to fulfill the task: {task}, and complete the purpose: {purpose}\n'
|
149 |
+
for z in range(divi):
|
150 |
+
print(f's:e :: {s}:{e}')
|
151 |
+
|
152 |
+
hist = history[s:e]
|
153 |
+
|
154 |
+
resp = run_gpt(
|
155 |
+
COMPRESS_DATA_PROMPT,
|
156 |
+
stop_tokens=["observation:", "task:", "action:", "thought:"],
|
157 |
+
max_tokens=8192,
|
158 |
+
seed=seed,
|
159 |
+
direction=instruct,
|
160 |
+
knowledge=new_history,
|
161 |
+
history=hist,
|
162 |
+
)
|
163 |
+
|
164 |
+
new_history = resp
|
165 |
+
print (resp)
|
166 |
+
out+=resp
|
167 |
+
e=e+chunk
|
168 |
+
s=s+chunk
|
169 |
+
'''
|
170 |
+
resp = run_gpt(
|
171 |
+
COMPRESS_DATA_PROMPT,
|
172 |
+
stop_tokens=["observation:", "task:", "action:", "thought:"],
|
173 |
+
max_tokens=8192,
|
174 |
+
seed=seed,
|
175 |
+
direction=instruct,
|
176 |
+
knowledge=new_history,
|
177 |
+
history="All data has been recieved.",
|
178 |
+
)'''
|
179 |
+
print ("final" + resp)
|
180 |
+
#history = "observation: {}\n".format(resp)
|
181 |
+
return resp
|
182 |
+
|
183 |
+
|
184 |
+
|
185 |
+
def summarize(inp,history,report_check,sum_mem_check,data=None):
|
186 |
+
json_box=[]
|
187 |
+
if inp == "":
|
188 |
+
inp = "Process this data"
|
189 |
+
history.clear()
|
190 |
+
history = [(inp,"Working on it...")]
|
191 |
+
yield "",history,error_box,json_box
|
192 |
+
|
193 |
+
if data != "Error" and data != "" and data != None:
|
194 |
+
print(inp)
|
195 |
+
out = str(data)
|
196 |
+
rl = len(out)
|
197 |
+
print(f'rl:: {rl}')
|
198 |
+
c=1
|
199 |
+
for i in str(out):
|
200 |
+
if i == " " or i=="," or i=="\n":
|
201 |
+
c +=1
|
202 |
+
print (f'c:: {c}')
|
203 |
+
if sum_mem_check=="Memory":
|
204 |
+
#save_memory(inp,out)
|
205 |
+
rawp = "Complete"
|
206 |
+
if sum_mem_check=="Summarize":
|
207 |
+
json_out = compress_data(c,inp,out)
|
208 |
+
|
209 |
+
out = str(json_out)
|
210 |
+
if report_check:
|
211 |
+
rl = len(out)
|
212 |
+
print(f'rl:: {rl}')
|
213 |
+
c=1
|
214 |
+
for i in str(out):
|
215 |
+
if i == " " or i=="," or i=="\n":
|
216 |
+
c +=1
|
217 |
+
print (f'c2:: {c}')
|
218 |
+
rawp = compress_data_og(c,inp,out)
|
219 |
+
else:
|
220 |
+
rawp = out
|
221 |
+
else:
|
222 |
+
rawp = "Provide a valid data source"
|
223 |
+
history.clear()
|
224 |
+
history.append((inp,rawp))
|
225 |
+
yield "", history,error_box,json_out
|
226 |
+
|
227 |
+
|
228 |
+
#################################
|
229 |
+
def clear_fn():
|
230 |
+
return "",[(None,None)]
|
231 |
+
|
232 |
+
with gr.Blocks() as app:
|
233 |
+
gr.HTML("""<center><h1>Mixtral 8x7B TLDR Summarizer + Web</h1><h3>Summarize Data of unlimited length</h3>""")
|
234 |
+
chatbot = gr.Chatbot(label="Mixtral 8x7B Chatbot",show_copy_button=True)
|
235 |
+
with gr.Row():
|
236 |
+
with gr.Column(scale=3):
|
237 |
+
prompt=gr.Textbox(label = "Instructions (optional)")
|
238 |
+
with gr.Column(scale=1):
|
239 |
+
report_check=gr.Checkbox(label="Return Report", value=True)
|
240 |
+
sum_mem_check=gr.Radio(label="Output",choices=["Summary","Memory"])
|
241 |
+
button=gr.Button()
|
242 |
+
|
243 |
+
#models_dd=gr.Dropdown(choices=[m for m in return_list],interactive=True)
|
244 |
+
with gr.Row():
|
245 |
+
stop_button=gr.Button("Stop")
|
246 |
+
clear_btn = gr.Button("Clear")
|
247 |
+
with gr.Row():
|
248 |
+
with gr.Tab("Text"):
|
249 |
+
data=gr.Textbox(label="Input Data (paste text)", lines=6)
|
250 |
+
with gr.Tab("File"):
|
251 |
+
file=gr.Files(label="Input File(s) (.pdf .txt)")
|
252 |
+
with gr.Tab("Raw HTML"):
|
253 |
+
url = gr.Textbox(label="URL")
|
254 |
+
with gr.Tab("PDF URL"):
|
255 |
+
pdf_url = gr.Textbox(label="PDF URL")
|
256 |
+
with gr.Tab("PDF Batch"):
|
257 |
+
pdf_batch = gr.Textbox(label="PDF URL Batch (comma separated)")
|
258 |
+
json_out=gr.JSON()
|
259 |
+
e_box=gr.Textbox()
|
260 |
+
#text=gr.JSON()
|
261 |
+
#inp_query.change(search_models,inp_query,models_dd)
|
262 |
+
clear_btn.click(clear_fn,None,[prompt,chatbot])
|
263 |
+
go=button.click(summarize,[prompt,chatbot,report_check,sum_mem_check,data,file,url,pdf_url,pdf_batch],[prompt,chatbot,e_box,json_out])
|
264 |
+
stop_button.click(None,None,None,cancels=[go])
|
265 |
+
app.queue(default_concurrency_limit=20).launch(show_api=False)
|