Merge branch 'main' into main

app.py

@@ -198,4 +198,5 @@ if __name__ == '__main__':
         )
 
     concurrency = int(os.environ.get('CONCURRENCY', os.cpu_count()))
-    demo.queue().launch(max_threads=concurrency)
+    favicon_path = os.path.join(os.path.dirname(__file__), 'llmriddles', 'assets', 'avatar.png')
+    demo.queue().launch(max_threads=concurrency, favicon_path=favicon_path, share=True)

llmriddles/questions/__init__.py

@@ -1,11 +1,13 @@
 from .executor import QuestionExecutor
 from .level1 import __file__ as _level1_file_
+from .level2 import __file__ as _level2_file_
 from .level3 import __file__ as _level3_file_
 from .level4 import __file__ as _level4_file_
 from .level5 import __file__ as _level5_file_
 from .question import Question, register_question, list_ordered_questions
 
 _ = _level1_file_
+_ = _level2_file_
 _ = _level3_file_
 _ = _level4_file_
 _ = _level5_file_

llmriddles/questions/level2.py

@@ -0,0 +1,299 @@
+import re
+from typing import Optional, Tuple
+
+import sympy
+
+from .question import register_question
+
+CN_TEXT_1 = """
+你需要提出一个字数是质数的问题，使回答的长度刚好是它的下一个质数。
+"""
+EN_TEXT_1 = """
+You need to come up with a question that has a prime number of words, so the answer's length is exactly the next prime number.
+"""
+
+
+def _is_prime(v):
+    return sympy.isprime(v)
+
+
+def _next_prime(v):
+    while v:
+        v += 1
+        if _is_prime(v):
+            return v
+
+
+def _cn_checker_1(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    qs_length = len(user_text.strip())
+    if not _is_prime(qs_length):
+        return False, f'问题长度为{qs_length}，非质数'
+
+    answer_value = len(answer_text)
+    next_prime = _next_prime(qs_length)
+    if answer_value != next_prime:
+        return False, f'下一个质数为{next_prime}，但回答长度为{answer_value}'
+
+    return True, None
+
+
+def _en_words(text: str):
+    return len(re.findall(r'\w+', text))
+
+
+def _en_checker_1(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    qs_length = _en_words(user_text.strip())
+    if not _is_prime(qs_length):
+        return False, f'The question has a length of {qs_length}, which is not a prime number'
+
+    answer_value = _en_words(answer_text)
+    next_prime = _next_prime(qs_length)
+    if answer_value != next_prime:
+        return False, f'The next prime number is {next_prime}, but the answer\'s length is {answer_value}'
+
+    return True, None
+
+
+register_question({
+    'cn': CN_TEXT_1,
+    'en': EN_TEXT_1,
+}, checkers={
+    'cn': _cn_checker_1,
+    'en': _en_checker_1,
+})
+
+CN_TEXT_2 = """
+请输入一个大于一的正整数作为问题，使回答里包含至少比它大一千的数。
+"""
+EN_TEXT_2 = """
+Please enter a positive integer greater than one as a question, so that the answer contains a number that is at least one thousand more than it.
+"""
+
+
+def _cn_checker_2(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    try:
+        value = int(user_text.strip())
+        if value <= 1:
+            raise ValueError
+    except (TypeError, ValueError):
+        return False, f'输入内容{user_text!r}，并非一个大于1的正整数'
+
+    for value_item in re.findall('[-+]?\d+', answer_text):
+        value_item = int(value_item)
+        if value_item >= value + 1000:
+            return True, f'检测到输出中数字{value_item}，满足要求'
+
+    return False, f'未在输出中检测到不少于{value + 1000}的数字'
+
+
+def _en_checker_2(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    try:
+        value = int(user_text.strip())
+        if value <= 1:
+            raise ValueError
+    except (TypeError, ValueError):
+        return False, f'You entered {user_text!r}, which is not a positive integer greater than 1'
+
+    for value_item in re.findall('[-+]?\d+', answer_text):
+        value_item = int(value_item)
+        if value_item >= value + 1000:
+            return True, f'Detected the number {value_item} in the output, which meets the requirement'
+
+    return False, f'Did not detect a number of at least {value + 1000} in the output'
+
+
+register_question({
+    'cn': CN_TEXT_2,
+    'en': EN_TEXT_2,
+}, checkers={
+    'cn': _cn_checker_2,
+    'en': _en_checker_2,
+})
+
+CN_TEXT_3 = """
+请输入一个大于一的正整数作为问题，使回答里包含至少10个不同的至少比它小一千的自然数。
+"""
+EN_TEXT_3 = """
+Please enter a positive integer greater than one as a question, so that the answer contains at least 10 different natural numbers, each of them at least 1000 less than the question.
+"""
+
+
+def _cn_checker_3(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    try:
+        value = int(user_text.strip())
+        if value <= 1:
+            raise ValueError
+    except (TypeError, ValueError):
+        return False, f'输入内容{user_text!r}，并非一个大于1的正整数'
+
+    collected_values = []
+    for value_item in re.findall('[-+]?\d+', answer_text):
+        value_item = int(value_item)
+        if value_item <= value - 1000:
+            collected_values.append(value_item)
+
+    collected_values = sorted(set(collected_values))
+    if len(collected_values) >= 10:
+        return True, f'检测到{len(collected_values)}个不大于{value - 1000}的数字：{collected_values!r}'
+    else:
+        return False, f'检测到{len(collected_values)}个不大于{value - 1000}的数字,未达到10个：{collected_values!r}'
+
+
+def _en_checker_3(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    try:
+        value = int(user_text.strip())
+        if value <= 1:
+            raise ValueError
+    except (TypeError, ValueError):
+        return False, f'You entered {user_text!r}, which is not a positive integer greater than 1'
+
+    collected_values = []
+    for value_item in re.findall('[-+]?\d+', answer_text):
+        value_item = int(value_item)
+        if value_item <= value - 1000:
+            collected_values.append(value_item)
+
+    collected_values = sorted(set(collected_values))
+    if len(collected_values) >= 10:
+        return True, f'Detected {len(collected_values)} numbers not greater than {value - 1000}: {collected_values!r}'
+    else:
+        return False, f'Detected {len(collected_values)} numbers not greater than {value - 1000}, ' \
+                      f'but it doesn\'t reach 10: {collected_values!r}'
+
+
+register_question({
+    'cn': CN_TEXT_3,
+    'en': EN_TEXT_3,
+}, checkers={
+    'cn': _cn_checker_3,
+    'en': _en_checker_3,
+})
+
+CN_TEXT_4 = """
+请在不提及1,4,5这三个字符的情况下让模型回答114514，不要有多余的字符。
+"""
+EN_TEXT_4 = """
+Please make the model answer "114514" without mentioning the characters 1, 4, and 5, and without any extra characters.
+"""
+
+
+def _cn_checker_4(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    if "1" in user_text or "4" in user_text or "5" in user_text:
+        return False, '请不要在提问中出现1、4、5三个数字'
+
+    return answer_text.strip() == '114514', None
+
+
+def _en_checker_4(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    if "1" in user_text or "4" in user_text or "5" in user_text:
+        return False, 'Please don\'t include the numbers 1, 4, and 5 in the question'
+
+    return answer_text.strip() == '114514', None
+
+
+register_question({
+    'cn': CN_TEXT_4,
+    'en': EN_TEXT_4,
+}, checkers={
+    'cn': _cn_checker_4,
+    'en': _en_checker_4,
+})
+
+CN_TEXT_5 = """
+请输入一个大于一的平方数作为问题，使回答里包含它的平方根对应的数。
+"""
+EN_TEXT_5 = """
+Please provide a square number greater than one as the question, and the answer should include the number corresponding to its square root.
+"""
+
+
+def _cn_checker_5(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    try:
+        value = int(user_text.strip())
+        if value <= 1:
+            raise ValueError
+        if int(value ** 0.5) ** 2 != value:
+            raise ValueError
+    except (TypeError, ValueError):
+        return False, f'输入内容{user_text!r}，并非一个大于1的平方数'
+
+    sq = int(value ** 0.5)
+    for value_item in re.findall('[-+]?\d+', answer_text):
+        value_item = int(value_item)
+        if value_item == sq:
+            return True, f'检测到平方根数{sq}'
+    else:
+        return False, f'未检测到平方根数{sq}'
+
+
+def _en_checker_5(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    try:
+        value = int(user_text.strip())
+        if value <= 1:
+            raise ValueError
+        if int(value ** 0.5) ** 2 != value:
+            raise ValueError
+    except (TypeError, ValueError):
+        return False, f'The input, {user_text!r}, is not a square number greater than 1'
+
+    sq = int(value ** 0.5)
+    for value_item in re.findall('[-+]?\d+', answer_text):
+        value_item = int(value_item)
+        if value_item == sq:
+            return True, f'Detected square root: {sq}'
+    else:
+        return False, f'Square root {sq} not detected'
+
+
+register_question({
+    'cn': CN_TEXT_5,
+    'en': EN_TEXT_5,
+}, checkers={
+    'cn': _cn_checker_5,
+    'en': _en_checker_5,
+})
+
+CN_TEXT_6 = """
+请提一个不包含“狗”这个字的10个字以内的问题，使回答中“狗”这个字出现的次数至少是问题字数的2倍。
+"""
+EN_TEXT_6 = """
+Please ask a question in under 10 words without the word "dog" and ensure that the word "dog" appears at least twice in the answer for each word in the question.
+"""
+
+
+def _cn_checker_6(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    if len(user_text) > 10:
+        return False, '问题不得超过10个字'
+    if '狗' in user_text:
+        return False, '问题不得包含“狗”字'
+
+    dog_count = len(re.findall('狗', answer_text))
+    if dog_count >= len(user_text) * 2:
+        return True, f'“狗”字的出现次数为{dog_count}次'
+    else:
+        return False, f'“狗”字的出现次数为{dog_count}次，未达到{len(user_text) * 2}次'
+
+
+def _en_checker_6(question_text: str, user_text: str, answer_text: str) -> Tuple[bool, Optional[str]]:
+    q_words = re.findall(r'\w+', user_text.lower())
+    if len(q_words) > 10:
+        return False, 'The question must not exceed 10 words'
+    if any(word in {'dog', 'dogs'} for word in q_words):
+        return False, 'The question must not contain the word "dog" or "dogs"'
+
+    a_words = re.findall(r'\w+', answer_text.lower())
+    a_dog_count = sum(1 if word in {'dog', 'dogs'} else 0 for word in a_words)
+    if a_dog_count >= len(q_words) * 2:
+        return True, f'The word "dog" (or "dogs") appears {a_dog_count} times.'
+    else:
+        return False, f'The word "dog" (or "dogs") appears {a_dog_count} times, ' \
+                      f'which is less than {len(q_words) * 2} times.'
+
+
+register_question({
+    'cn': CN_TEXT_6,
+    'en': EN_TEXT_6,
+}, checkers={
+    'cn': _cn_checker_6,
+    'en': _en_checker_6,
+})

llmriddles/questions/question.py

@@ -23,8 +23,8 @@ def register_question(text: Union[Mapping[str, str], str],
     if isinstance(checkers, collections.abc.Mapping):
         _origin_checkers = checkers
 
-        def _integrated_checker(question_text: str, answer_text: str, lang: str):
-            return _origin_checkers[lang](question_text, answer_text)
+        def _integrated_checker(question_text: str, user_text: str, answer_text: str, lang: str):
+            return _origin_checkers[lang](question_text, user_text, answer_text)
 
         checker: MultiLangCheckerTyping = _integrated_checker
     else:

requirements.txt

@@ -3,5 +3,6 @@ tqdm
 requests>=2.20
 gradio==4.1.1
 openai>=1
+sympy
 flask
-transformers
+transformers