Spaces:
Running
Running
update app.py
Browse files- QuALITY.v1.0.1.htmlstripped.dev +0 -0
- app.py +441 -47
- example.py +63 -0
- requirements.txt +3 -1
QuALITY.v1.0.1.htmlstripped.dev
ADDED
The diff for this file is too large to render.
See raw diff
|
|
app.py
CHANGED
@@ -1,63 +1,457 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
"""
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
"""
|
7 |
-
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
8 |
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
)
|
18 |
-
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
-
|
21 |
-
if val[0]:
|
22 |
-
messages.append({"role": "user", "content": val[0]})
|
23 |
-
if val[1]:
|
24 |
-
messages.append({"role": "assistant", "content": val[1]})
|
25 |
|
26 |
-
messages.append({"role": "user", "content": message})
|
27 |
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
max_tokens=max_tokens,
|
33 |
-
stream=True,
|
34 |
-
temperature=temperature,
|
35 |
-
top_p=top_p,
|
36 |
-
):
|
37 |
-
token = message.choices[0].delta.content
|
38 |
|
39 |
-
|
40 |
-
|
|
|
41 |
|
|
|
42 |
"""
|
43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
"""
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
|
62 |
-
if __name__ ==
|
63 |
-
demo.launch()
|
|
|
1 |
+
import copy
|
2 |
+
import datetime
|
3 |
+
import json
|
4 |
+
import os
|
5 |
+
import re
|
6 |
+
import string
|
7 |
+
import time
|
8 |
+
|
9 |
import gradio as gr
|
10 |
+
import openai
|
11 |
+
import google.generativeai as genai
|
12 |
+
|
13 |
+
|
14 |
+
openai_key = os.environ.get('OPEN_AI_KEY')
|
15 |
+
gpt_client = openai.OpenAI(api_key=openai_key)
|
16 |
+
|
17 |
+
gemini_key = os.environ.get('GEMINI_API_KEY')
|
18 |
+
genai.configure(api_key=gemini_key)
|
19 |
+
|
20 |
+
|
21 |
+
def query_gpt_model(
|
22 |
+
prompt: str,
|
23 |
+
llm: str = 'gpt-3.5-turbo-1106',
|
24 |
+
temperature: float = 0.0,
|
25 |
+
max_decode_steps: int = 512,
|
26 |
+
seconds_to_reset_tokens: float = 30.0,
|
27 |
+
) -> str:
|
28 |
+
|
29 |
+
while True:
|
30 |
+
try:
|
31 |
+
raw_response = gpt_client.chat.completions.with_raw_response.create(
|
32 |
+
model=llm,
|
33 |
+
max_tokens=max_decode_steps,
|
34 |
+
temperature=temperature,
|
35 |
+
messages=[
|
36 |
+
{'role': 'user', 'content': prompt},
|
37 |
+
]
|
38 |
+
)
|
39 |
+
completion = raw_response.parse()
|
40 |
+
return completion.choices[0].message.content
|
41 |
+
except openai.RateLimitError as e:
|
42 |
+
print(f'{datetime.datetime.now()}: query_gpt_model: RateLimitError {e.message}: {e}')
|
43 |
+
time.sleep(seconds_to_reset_tokens)
|
44 |
+
except openai.APIError as e:
|
45 |
+
print(f'{datetime.datetime.now()}: query_gpt_model: APIError {e.message}: {e}')
|
46 |
+
print(f'{datetime.datetime.now()}: query_gpt_model: Retrying after 5 seconds...')
|
47 |
+
time.sleep(5)
|
48 |
+
|
49 |
+
safety_settings=[
|
50 |
+
{"category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "BLOCK_ONLY_HIGH"},
|
51 |
+
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "BLOCK_ONLY_HIGH"},
|
52 |
+
{"category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "BLOCK_ONLY_HIGH"},
|
53 |
+
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "BLOCK_ONLY_HIGH"}
|
54 |
+
]
|
55 |
+
|
56 |
+
def query_gemini_model(
|
57 |
+
prompt: str,
|
58 |
+
llm: str = 'gemini-pro',
|
59 |
+
retries: int = 10,
|
60 |
+
) -> str:
|
61 |
+
model = genai.GenerativeModel(llm)
|
62 |
+
while True and retries > 0:
|
63 |
+
try:
|
64 |
+
response = model.generate_content(prompt, safety_settings=safety_settings)
|
65 |
+
text_response = response.text.replace("**", "")
|
66 |
+
return text_response
|
67 |
+
except Exception as e:
|
68 |
+
print(f'{datetime.datetime.now()}: query_gemini_model: Error: {e}')
|
69 |
+
print(f'{datetime.datetime.now()}: query_gemini_model: Retrying after 5 seconds...')
|
70 |
+
retries -= 1
|
71 |
+
time.sleep(5)
|
72 |
+
|
73 |
+
|
74 |
+
def query_model(
|
75 |
+
prompt: str,
|
76 |
+
model_name: str = 'gemini-pro',
|
77 |
+
) -> str:
|
78 |
+
model_type = model_name.split('-')[0]
|
79 |
+
if model_type == "gpt":
|
80 |
+
return query_gpt_model(prompt, llm=model_name)
|
81 |
+
elif model_type == "gemini":
|
82 |
+
return query_gemini_model(prompt, llm=model_name)
|
83 |
+
else:
|
84 |
+
raise ValueError('Unexpected model_name: ', model_name)
|
85 |
+
|
86 |
+
# Load QuALITY dataset
|
87 |
+
|
88 |
+
_ONE2ONE_FIELDS = (
|
89 |
+
'article',
|
90 |
+
'article_id',
|
91 |
+
'set_unique_id',
|
92 |
+
'writer_id',
|
93 |
+
'source',
|
94 |
+
'title',
|
95 |
+
'topic',
|
96 |
+
'url',
|
97 |
+
'writer_id',
|
98 |
+
'author',
|
99 |
+
)
|
100 |
+
|
101 |
+
quality_dev = []
|
102 |
+
with open('QuALITY.v1.0.1.htmlstripped.dev', 'r') as f:
|
103 |
+
for line in f.readlines():
|
104 |
+
j = json.loads(line)
|
105 |
+
fields = {k: j[k] for k in _ONE2ONE_FIELDS}
|
106 |
+
fields.update({
|
107 |
+
'questions': [q['question'] for q in j['questions']],
|
108 |
+
'question_ids': [q['question_unique_id'] for q in j['questions']],
|
109 |
+
'difficults': [q['difficult'] for q in j['questions']],
|
110 |
+
'options': [q['options'] for q in j['questions']],
|
111 |
+
})
|
112 |
+
|
113 |
+
fields.update({
|
114 |
+
'gold_labels': [q['gold_label'] for q in j['questions']],
|
115 |
+
'writer_labels': [q['writer_label'] for q in j['questions']],
|
116 |
+
})
|
117 |
+
|
118 |
+
quality_dev.append(fields)
|
119 |
+
|
120 |
+
|
121 |
+
|
122 |
+
# Helper functions
|
123 |
+
all_lowercase_letters = string.ascii_lowercase # "abcd...xyz"
|
124 |
+
bracketed_lowercase_letters_set = set(
|
125 |
+
[f"({l})" for l in all_lowercase_letters]
|
126 |
+
) # {"(a)", ...}
|
127 |
+
bracketed_uppercase_letters_set = set(
|
128 |
+
[f"({l.upper()})" for l in all_lowercase_letters]
|
129 |
+
) # {"(a)", ...}
|
130 |
+
|
131 |
+
choices = ['(A)', '(B)', '(C)', '(D)']
|
132 |
+
|
133 |
+
def get_index_from_symbol(answer):
|
134 |
+
"""Get the index from the letter symbols A, B, C, D, to extract answer texts.
|
135 |
+
|
136 |
+
Args:
|
137 |
+
answer (str): the string of answer like "(B)".
|
138 |
+
|
139 |
+
Returns:
|
140 |
+
index (int): how far the given choice is from "a", like 1 for answer "(B)".
|
141 |
+
"""
|
142 |
+
answer = str(answer).lower()
|
143 |
+
# extract the choice letter from within bracket
|
144 |
+
if answer in bracketed_lowercase_letters_set:
|
145 |
+
answer = re.findall(r".*?", answer)[0][1]
|
146 |
+
index = ord(answer) - ord("a")
|
147 |
+
return index
|
148 |
+
|
149 |
+
def count_words(text):
|
150 |
+
"""Simple word counting."""
|
151 |
+
return len(text.split())
|
152 |
+
|
153 |
+
def quality_gutenberg_parser(raw_article):
|
154 |
+
"""Parse Gutenberg articles in the QuALITY dataset."""
|
155 |
+
lines = []
|
156 |
+
previous_line = None
|
157 |
+
for i, line in enumerate(raw_article.split('\n')):
|
158 |
+
line = line.strip()
|
159 |
+
original_line = line
|
160 |
+
if line == '':
|
161 |
+
if previous_line == '':
|
162 |
+
line = '\n'
|
163 |
+
else:
|
164 |
+
previous_line = original_line
|
165 |
+
continue
|
166 |
+
previous_line = original_line
|
167 |
+
lines.append(line)
|
168 |
+
return ' '.join(lines)
|
169 |
+
|
170 |
+
|
171 |
+
|
172 |
+
# ReadAgent (1) Episode Pagination
|
173 |
+
prompt_pagination_template = """
|
174 |
+
You are given a passage that is taken from a larger text (article, book, ...) and some numbered labels between the paragraphs in the passage.
|
175 |
+
Numbered label are in angeled brackets. For example, if the label number is 19, it shows as <19> in text.
|
176 |
+
Please choose one label that it is natural to break reading.
|
177 |
+
Such point can be scene transition, end of a dialogue, end of an argument, narrative transition, etc.
|
178 |
+
Please answer the break point label and explain.
|
179 |
+
For example, if <57> is a good point to break, answer with \"Break point: <57>\n Because ...\"
|
180 |
+
|
181 |
+
Passage:
|
182 |
+
|
183 |
+
{0}
|
184 |
+
{1}
|
185 |
+
{2}
|
186 |
|
187 |
"""
|
188 |
+
|
189 |
+
def parse_pause_point(text):
|
190 |
+
text = text.strip("Break point: ")
|
191 |
+
if text[0] != '<':
|
192 |
+
return None
|
193 |
+
for i, c in enumerate(text):
|
194 |
+
if c == '>':
|
195 |
+
if text[1:i].isnumeric():
|
196 |
+
return int(text[1:i])
|
197 |
+
else:
|
198 |
+
return None
|
199 |
+
return None
|
200 |
+
|
201 |
+
|
202 |
+
def quality_pagination(example,
|
203 |
+
model_name='gemini-pro',
|
204 |
+
word_limit=600,
|
205 |
+
start_threshold=280,
|
206 |
+
max_retires=10,
|
207 |
+
verbose=True,
|
208 |
+
allow_fallback_to_last=True):
|
209 |
+
article = example['article']
|
210 |
+
title = example['title']
|
211 |
+
text_output = f"[Pagination][Article {title}]" + '\n\n'
|
212 |
+
paragraphs = quality_gutenberg_parser(article).split('\n')
|
213 |
+
|
214 |
+
i = 0
|
215 |
+
pages = []
|
216 |
+
while i < len(paragraphs):
|
217 |
+
preceding = "" if i == 0 else "...\n" + '\n'.join(pages[-1])
|
218 |
+
passage = [paragraphs[i]]
|
219 |
+
wcount = count_words(paragraphs[i])
|
220 |
+
j = i + 1
|
221 |
+
while wcount < word_limit and j < len(paragraphs):
|
222 |
+
wcount += count_words(paragraphs[j])
|
223 |
+
if wcount >= start_threshold:
|
224 |
+
passage.append(f"<{j}>")
|
225 |
+
passage.append(paragraphs[j])
|
226 |
+
j += 1
|
227 |
+
passage.append(f"<{j}>")
|
228 |
+
end_tag = "" if j == len(paragraphs) else paragraphs[j] + "\n..."
|
229 |
+
|
230 |
+
pause_point = None
|
231 |
+
if wcount < 350:
|
232 |
+
pause_point = len(paragraphs)
|
233 |
+
else:
|
234 |
+
prompt = prompt_pagination_template.format(preceding, '\n'.join(passage), end_tag)
|
235 |
+
response = query_model(prompt=prompt, model_name=model_name).strip()
|
236 |
+
pause_point = parse_pause_point(response)
|
237 |
+
if pause_point and (pause_point <= i or pause_point > j):
|
238 |
+
# process += f"prompt:\n{prompt},\nresponse:\n{response}\n"
|
239 |
+
# process += f"i:{i} j:{j} pause_point:{pause_point}" + '\n'
|
240 |
+
pause_point = None
|
241 |
+
if pause_point is None:
|
242 |
+
if allow_fallback_to_last:
|
243 |
+
pause_point = j
|
244 |
+
else:
|
245 |
+
raise ValueError(f"prompt:\n{prompt},\nresponse:\n{response}\n")
|
246 |
+
|
247 |
+
page = paragraphs[i:pause_point]
|
248 |
+
pages.append(page)
|
249 |
+
text_output += f"Paragraph {i}-{pause_point-1}: {page}\n\n"
|
250 |
+
i = pause_point
|
251 |
+
text_output += f"\n\n[Pagination] Done with {len(pages)} pages"
|
252 |
+
return pages, text_output
|
253 |
+
|
254 |
+
# pages = quality_pagination(example)
|
255 |
+
|
256 |
+
|
257 |
+
# ReadAgent (2) Memory Gisting
|
258 |
+
prompt_shorten_template = """
|
259 |
+
Please shorten the following passage.
|
260 |
+
Just give me a shortened version. DO NOT explain your reason.
|
261 |
+
|
262 |
+
Passage:
|
263 |
+
{}
|
264 |
+
|
265 |
"""
|
|
|
266 |
|
267 |
+
def quality_gisting(example, pages, model_name, word_limit=600, start_threshold=280, verbose=True):
|
268 |
+
article = example['article']
|
269 |
+
title = example['title']
|
270 |
+
word_count = count_words(article)
|
271 |
+
text_output = f"[Gisting][Article {title}], {word_count} words\n\n"
|
272 |
|
273 |
+
shortened_pages = []
|
274 |
+
for i, page in enumerate(pages):
|
275 |
+
prompt = prompt_shorten_template.format('\n'.join(page))
|
276 |
+
response = query_model(prompt, model_name)
|
277 |
+
shortened_text = response.strip()
|
278 |
+
shortened_pages.append(shortened_text)
|
279 |
+
text_output += "[gist] page {}: {}\n\n".format(i, shortened_text)
|
280 |
+
shortened_article = '\n'.join(shortened_pages)
|
281 |
+
gist_word_count = count_words(shortened_article)
|
282 |
+
text_output += '\n\n' + f"Shortened article:\n{shortened_article}\n\n"
|
283 |
+
output = copy.deepcopy(example)
|
284 |
+
output.update({'title': title, 'word_count': word_count, 'gist_word_count': gist_word_count, 'shortened_pages': shortened_pages, 'pages': pages})
|
285 |
+
text_output += f"\n\ncompression rate {round(100.0 - gist_word_count/word_count*100, 2)}% ({gist_word_count}/{word_count})"
|
286 |
+
return output, text_output
|
287 |
|
288 |
+
# example_with_gists = quality_gisting(example, pages)
|
|
|
|
|
|
|
|
|
289 |
|
|
|
290 |
|
291 |
+
# ReadAgent (3) Look-Up
|
292 |
+
prompt_lookup_template = """
|
293 |
+
The following text is what you remembered from reading an article and a multiple choice question related to it.
|
294 |
+
You may read 1 to 6 page(s) of the article again to refresh your memory to prepare yourselve for the question.
|
295 |
+
Please respond with which page(s) you would like to read.
|
296 |
+
For example, if your only need to read Page 8, respond with \"I want to look up Page [8] to ...\";
|
297 |
+
if your would like to read Page 7 and 12, respond with \"I want to look up Page [7, 12] to ...\";
|
298 |
+
if your would like to read Page 2, 3, 7, 15 and 18, respond with \"I want to look up Page [2, 3, 7, 15, 18] to ...\".
|
299 |
+
if your would like to read Page 3, 4, 5, 12, 13 and 16, respond with \"I want to look up Page [3, 3, 4, 12, 13, 16] to ...\".
|
300 |
+
DO NOT select more pages if you don't need to.
|
301 |
+
DO NOT answer the question yet.
|
302 |
|
303 |
+
Text:
|
304 |
+
{}
|
|
|
|
|
|
|
|
|
|
|
|
|
305 |
|
306 |
+
Question:
|
307 |
+
{}
|
308 |
+
{}
|
309 |
|
310 |
+
Take a deep breath and tell me: Which page(s) would you like to read again?
|
311 |
"""
|
312 |
+
|
313 |
+
prompt_answer_template = """
|
314 |
+
Read the following article and answer a multiple choice question.
|
315 |
+
For example, if (C) is correct, answer with \"Answer: (C) ...\"
|
316 |
+
|
317 |
+
Article:
|
318 |
+
{}
|
319 |
+
|
320 |
+
Question:
|
321 |
+
{}
|
322 |
+
{}
|
323 |
+
|
324 |
"""
|
325 |
+
|
326 |
+
def quality_parallel_lookup(example, verbose=True):
|
327 |
+
preprocessed_pages = example['pages']
|
328 |
+
article = example['article']
|
329 |
+
title = example['title']
|
330 |
+
word_count = example['word_count']
|
331 |
+
gist_word_count = example['gist_word_count']
|
332 |
+
pages = example['pages']
|
333 |
+
shortened_pages = example['shortened_pages']
|
334 |
+
questions = example['questions']
|
335 |
+
options = example['options']
|
336 |
+
gold_labels = example['gold_labels'] # numerical [1, 2, 3, 4]
|
337 |
+
|
338 |
+
text_outputs = [f"[Look-Up][Article {title}] {word_count} words"]
|
339 |
+
|
340 |
+
model_choices = []
|
341 |
+
lookup_page_ids = []
|
342 |
+
|
343 |
+
shortened_pages_pidx = []
|
344 |
+
for i, shortened_text in enumerate(shortened_pages):
|
345 |
+
shortened_pages_pidx.append("\n".format(i) + shortened_text)
|
346 |
+
shortened_article = '\n'.join(shortened_pages_pidx)
|
347 |
+
|
348 |
+
expanded_gist_word_counts = []
|
349 |
+
|
350 |
+
for i, label in enumerate(gold_labels):
|
351 |
+
# only test the first question for demo
|
352 |
+
if i != 1:
|
353 |
+
continue
|
354 |
+
q = questions[i]
|
355 |
+
text_output = f"question {i}: {q}" + '\n\n'
|
356 |
+
options_i = [f"{ol} {o}" for ol, o in zip(choices, options[i])]
|
357 |
+
text_output += "options: " + "\n".join(options_i)
|
358 |
+
text_output += '\n\n'
|
359 |
+
prompt_lookup = prompt_lookup_template.format(shortened_article, q, '\n'.join(options_i))
|
360 |
+
|
361 |
+
page_ids = []
|
362 |
+
|
363 |
+
response = query_model(prompt=prompt_lookup).strip()
|
364 |
+
|
365 |
+
try: start = response.index('[')
|
366 |
+
except ValueError: start = len(response)
|
367 |
+
try: end = response.index(']')
|
368 |
+
except ValueError: end = 0
|
369 |
+
if start < end:
|
370 |
+
page_ids_str = response[start+1:end].split(',')
|
371 |
+
page_ids = []
|
372 |
+
for p in page_ids_str:
|
373 |
+
if p.strip().isnumeric():
|
374 |
+
page_id = int(p)
|
375 |
+
if page_id < 0 or page_id >= len(pages):
|
376 |
+
text_output += f"Skip invalid page number: {page_id}\n\n"
|
377 |
+
else:
|
378 |
+
page_ids.append(page_id)
|
379 |
+
|
380 |
+
text_output += "Model chose to look up page {}\n\n".format(page_ids)
|
381 |
+
|
382 |
+
# Memory expansion after look-up, replacing the target shortened page with the original page
|
383 |
+
expanded_shortened_pages = shortened_pages[:]
|
384 |
+
if len(page_ids) > 0:
|
385 |
+
for page_id in page_ids:
|
386 |
+
expanded_shortened_pages[page_id] = '\n'.join(pages[page_id])
|
387 |
+
|
388 |
+
expanded_shortened_article = '\n'.join(expanded_shortened_pages)
|
389 |
+
expanded_gist_word_count = count_words(expanded_shortened_article)
|
390 |
+
text_output += "Expanded shortened article:\n" + expanded_shortened_article + '\n\n'
|
391 |
+
prompt_answer = prompt_answer_template.format(expanded_shortened_article, q, '\n'.join(options_i))
|
392 |
+
|
393 |
+
model_choice = None
|
394 |
+
response = query_model(prompt=prompt_answer)
|
395 |
+
response = response.strip()
|
396 |
+
for j, choice in enumerate(choices):
|
397 |
+
if response.startswith(f"Answer: {choice}") or response.startswith(f"Answer: {choice[1]}"):
|
398 |
+
model_choice = j+1
|
399 |
+
break
|
400 |
+
is_correct = 1 if model_choice == label else 0
|
401 |
+
text_output += f"reference answer: {choices[label]}, model prediction: {choices[model_choice]}, is_correct: {is_correct}" + '\n\n'
|
402 |
+
text_output += f"compression rate {round(100.0 - gist_word_count/word_count*100, 2)}% ({gist_word_count}/{word_count})" + '\n\n'
|
403 |
+
text_output += f"compression rate after look-up {round(100.0 - expanded_gist_word_count/word_count*100, 2)}% ({expanded_gist_word_count}/{word_count})" + '\n\n'
|
404 |
+
text_output += '\n\n'
|
405 |
+
text_outputs.append(text_output)
|
406 |
+
return text_outputs
|
407 |
+
|
408 |
+
|
409 |
+
def query_model_with_quality(
|
410 |
+
index: int,
|
411 |
+
model_name: str = 'gemini-pro'
|
412 |
+
):
|
413 |
+
example = quality_dev[index]
|
414 |
+
pages, pagination = quality_pagination(example, model_name)
|
415 |
+
print('Finish Pagination.')
|
416 |
+
example_with_gists, gisting = quality_gisting(example, pages, model_name)
|
417 |
+
print('Finish Gisting.')
|
418 |
+
answers = quality_parallel_lookup(example_with_gists)
|
419 |
+
return prompt_pagination_template, pagination, prompt_shorten_template, gisting, prompt_lookup_template, '\n\n'.join(answers)
|
420 |
+
|
421 |
+
|
422 |
+
llm_api_options = ['gemini-pro', 'gemini-1.5-flash', 'gpt-3.5-turbo-1106']
|
423 |
+
|
424 |
+
with gr.Blocks() as demo:
|
425 |
+
gr.Markdown(
|
426 |
+
"""
|
427 |
+
# A Human-Inspired Reading Agent with Gist Memory of Very Long Contexts
|
428 |
+
""")
|
429 |
+
with gr.Tab('ReadAgent (QuALITY)'):
|
430 |
+
llm_options = gr.Radio(llm_api_options, label="Backend LLM API", value='gemini-pro')
|
431 |
+
with gr.Row():
|
432 |
+
with gr.Column():
|
433 |
+
index = gr.Dropdown(list(range(len(quality_dev))), value=13, label="QuALITY Index",)
|
434 |
+
button = gr.Button("Execute")
|
435 |
+
prompt_pagination = gr.Textbox(label="Episode Pagination Prompt Template", lines=5)
|
436 |
+
pagination_results = gr.Textbox(label="Episode Pagination", lines=20)
|
437 |
+
prompt_gisting = gr.Textbox(label="Memory Gisting Prompt Template", lines=5)
|
438 |
+
gisting_results = gr.Textbox(label="Memory Gisting", lines=20)
|
439 |
+
prompt_lookup = gr.Textbox(label="Parallel Lookup Prompt Template", lines=5)
|
440 |
+
lookup_qa_results = gr.Textbox(label="Parallel Lookup and QA", lines=20)
|
441 |
+
|
442 |
+
button.click(
|
443 |
+
fn=query_model_with_quality,
|
444 |
+
inputs=[
|
445 |
+
index,
|
446 |
+
llm_options
|
447 |
+
],
|
448 |
+
outputs=[
|
449 |
+
prompt_pagination, pagination_results,
|
450 |
+
prompt_gisting, gisting_results,
|
451 |
+
prompt_lookup, lookup_qa_results,
|
452 |
+
]
|
453 |
+
)
|
454 |
|
455 |
|
456 |
+
if __name__ == '__main__':
|
457 |
+
demo.launch()
|
example.py
ADDED
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from huggingface_hub import InferenceClient
|
3 |
+
|
4 |
+
"""
|
5 |
+
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
|
6 |
+
"""
|
7 |
+
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
8 |
+
|
9 |
+
|
10 |
+
def respond(
|
11 |
+
message,
|
12 |
+
history: list[tuple[str, str]],
|
13 |
+
system_message,
|
14 |
+
max_tokens,
|
15 |
+
temperature,
|
16 |
+
top_p,
|
17 |
+
):
|
18 |
+
messages = [{"role": "system", "content": system_message}]
|
19 |
+
|
20 |
+
for val in history:
|
21 |
+
if val[0]:
|
22 |
+
messages.append({"role": "user", "content": val[0]})
|
23 |
+
if val[1]:
|
24 |
+
messages.append({"role": "assistant", "content": val[1]})
|
25 |
+
|
26 |
+
messages.append({"role": "user", "content": message})
|
27 |
+
|
28 |
+
response = ""
|
29 |
+
|
30 |
+
for message in client.chat_completion(
|
31 |
+
messages,
|
32 |
+
max_tokens=max_tokens,
|
33 |
+
stream=True,
|
34 |
+
temperature=temperature,
|
35 |
+
top_p=top_p,
|
36 |
+
):
|
37 |
+
token = message.choices[0].delta.content
|
38 |
+
|
39 |
+
response += token
|
40 |
+
yield response
|
41 |
+
|
42 |
+
"""
|
43 |
+
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
|
44 |
+
"""
|
45 |
+
demo = gr.ChatInterface(
|
46 |
+
respond,
|
47 |
+
additional_inputs=[
|
48 |
+
gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
|
49 |
+
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
|
50 |
+
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
|
51 |
+
gr.Slider(
|
52 |
+
minimum=0.1,
|
53 |
+
maximum=1.0,
|
54 |
+
value=0.95,
|
55 |
+
step=0.05,
|
56 |
+
label="Top-p (nucleus sampling)",
|
57 |
+
),
|
58 |
+
],
|
59 |
+
)
|
60 |
+
|
61 |
+
|
62 |
+
if __name__ == "__main__":
|
63 |
+
demo.launch()
|
requirements.txt
CHANGED
@@ -1 +1,3 @@
|
|
1 |
-
huggingface_hub==0.22.2
|
|
|
|
|
|
1 |
+
huggingface_hub==0.22.2
|
2 |
+
openai==1.37.0
|
3 |
+
google-generativeai==0.7.2
|