with_chat_gpt_now
Browse files- app.py +2 -60
- search_errors_logic.py +105 -17
app.py
CHANGED
|
@@ -2,7 +2,7 @@ import language_tool_python
|
|
| 2 |
import openai
|
| 3 |
import streamlit as st
|
| 4 |
import subprocess
|
| 5 |
-
from search_errors_logic import check_text
|
| 6 |
import html
|
| 7 |
import docx
|
| 8 |
from io import BytesIO
|
|
@@ -73,64 +73,6 @@ def load_assets():
|
|
| 73 |
return tool
|
| 74 |
|
| 75 |
|
| 76 |
-
# def generate_gpt_comment_openai(inp):
|
| 77 |
-
# response = openai.ChatCompletion.create(
|
| 78 |
-
# model="gpt-4o-mini",
|
| 79 |
-
# messages=[
|
| 80 |
-
# {"role": "user", "content": inp}
|
| 81 |
-
# ]
|
| 82 |
-
# )
|
| 83 |
-
# return response.choices[0].message['content']
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
# def generate_gpt_comment_vsegpt(inp):
|
| 87 |
-
# client = openai.OpenAI(
|
| 88 |
-
# api_key='sk-or-vv-44ca318747a45e810149bf769e9dfbe5f42046695875efd7c9121cca590d6906',
|
| 89 |
-
# base_url='https://api.vsegpt.ru/v1',
|
| 90 |
-
# )
|
| 91 |
-
# response = client.chat.completions.create(
|
| 92 |
-
# model="gpt-4o-mini",
|
| 93 |
-
# messages=[{"role": "user", "content": inp}]
|
| 94 |
-
# ).choices[0].message.content
|
| 95 |
-
|
| 96 |
-
# return response
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
# def generate_gpt_comment(inp, client_name):
|
| 100 |
-
# if client_name == "openai":
|
| 101 |
-
# return generate_gpt_comment_openai(inp)
|
| 102 |
-
# elif client_name == "vsegpt":
|
| 103 |
-
# return generate_gpt_comment_vsegpt(inp)
|
| 104 |
-
# else:
|
| 105 |
-
# raise ValueError(f"Unsupported client: {client_name}")
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
# def check_text(text, tool):
|
| 109 |
-
# matches = tool.check(text)
|
| 110 |
-
# errors = []
|
| 111 |
-
# for match in matches:
|
| 112 |
-
# inp = prompt.format(text[match.offset:match.offset + match.errorLength],
|
| 113 |
-
# match.message, match.context)
|
| 114 |
-
# error_info = {
|
| 115 |
-
# 'start': match.offset,
|
| 116 |
-
# 'end': match.offset + match.errorLength,
|
| 117 |
-
# 'message': generate_gpt_comment(inp, client_name="vsegpt"),
|
| 118 |
-
# }
|
| 119 |
-
# errors.append(error_info)
|
| 120 |
-
|
| 121 |
-
# errors = sorted(errors, key=lambda x: x['end'])
|
| 122 |
-
|
| 123 |
-
# shift = 0
|
| 124 |
-
# for i, error in enumerate(errors, 1):
|
| 125 |
-
# error['start'] += shift
|
| 126 |
-
# error['end'] += shift
|
| 127 |
-
# inp = f"({i})"
|
| 128 |
-
# text = text[:error['end']] + inp + text[error['end']:]
|
| 129 |
-
# shift += len(inp)
|
| 130 |
-
|
| 131 |
-
# return text.replace("\n", " "), errors
|
| 132 |
-
|
| 133 |
-
|
| 134 |
def extract_text_from_docx(file):
|
| 135 |
doc = docx.Document(BytesIO(file.getvalue()))
|
| 136 |
full_text = []
|
|
@@ -194,7 +136,7 @@ def main():
|
|
| 194 |
if not text.strip():
|
| 195 |
st.warning("Введите текст для проверки")
|
| 196 |
else:
|
| 197 |
-
text, errors =
|
| 198 |
if not errors:
|
| 199 |
st.success("Ошибок не найдено! 👍")
|
| 200 |
else:
|
|
|
|
| 2 |
import openai
|
| 3 |
import streamlit as st
|
| 4 |
import subprocess
|
| 5 |
+
from checked.search_errors_logic import check_text, check_text_chat_gpt
|
| 6 |
import html
|
| 7 |
import docx
|
| 8 |
from io import BytesIO
|
|
|
|
| 73 |
return tool
|
| 74 |
|
| 75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
def extract_text_from_docx(file):
|
| 77 |
doc = docx.Document(BytesIO(file.getvalue()))
|
| 78 |
full_text = []
|
|
|
|
| 136 |
if not text.strip():
|
| 137 |
st.warning("Введите текст для проверки")
|
| 138 |
else:
|
| 139 |
+
text, errors = check_text_chat_gpt(text, tool)
|
| 140 |
if not errors:
|
| 141 |
st.success("Ошибок не найдено! 👍")
|
| 142 |
else:
|
search_errors_logic.py
CHANGED
|
@@ -1,6 +1,44 @@
|
|
| 1 |
import openai
|
| 2 |
from openai import OpenAI
|
|
|
|
| 3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
prompt = """
|
| 6 |
Ты должен писать комментарии об ошибках в тексте.
|
|
@@ -45,7 +83,7 @@ prompt = """
|
|
| 45 |
"""
|
| 46 |
|
| 47 |
|
| 48 |
-
def
|
| 49 |
response = openai.ChatCompletion.create(
|
| 50 |
model="gpt-4o-mini",
|
| 51 |
messages=[
|
|
@@ -55,7 +93,7 @@ def generate_gpt_comment_openai(inp):
|
|
| 55 |
return response.choices[0].message['content']
|
| 56 |
|
| 57 |
|
| 58 |
-
def
|
| 59 |
client = OpenAI(
|
| 60 |
api_key='sk-or-vv-44ca318747a45e810149bf769e9dfbe5f42046695875efd7c9121cca590d6906',
|
| 61 |
base_url='https://api.vsegpt.ru/v1'
|
|
@@ -68,15 +106,74 @@ def generate_gpt_comment_vsegpt(inp):
|
|
| 68 |
return response
|
| 69 |
|
| 70 |
|
| 71 |
-
def
|
| 72 |
if client_name == "openai":
|
| 73 |
-
return
|
| 74 |
elif client_name == "vsegpt":
|
| 75 |
-
return
|
| 76 |
else:
|
| 77 |
raise ValueError(f"Unsupported client: {client_name}")
|
| 78 |
|
| 79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
def check_text(text, tool):
|
| 81 |
matches = tool.check(text)
|
| 82 |
errors = []
|
|
@@ -86,18 +183,9 @@ def check_text(text, tool):
|
|
| 86 |
error_info = {
|
| 87 |
'start': match.offset,
|
| 88 |
'end': match.offset + match.errorLength,
|
| 89 |
-
'message':
|
| 90 |
}
|
| 91 |
errors.append(error_info)
|
| 92 |
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
shift = 0
|
| 96 |
-
for i, error in enumerate(errors, 1):
|
| 97 |
-
error['start'] += shift
|
| 98 |
-
error['end'] += shift
|
| 99 |
-
inp = f"({i})"
|
| 100 |
-
text = text[:error['end']] + inp + text[error['end']:]
|
| 101 |
-
shift += len(inp)
|
| 102 |
-
|
| 103 |
-
return text.replace("\n", " "), errors
|
|
|
|
| 1 |
import openai
|
| 2 |
from openai import OpenAI
|
| 3 |
+
import difflib
|
| 4 |
|
| 5 |
+
prompt_fix_text_gpt = "Исправь ошибки в данном тексте:\n{}"
|
| 6 |
+
|
| 7 |
+
prompt_compare_get_comment = """
|
| 8 |
+
Ты должен писать комментарии об ошибках в тексте.
|
| 9 |
+
Тебе дан кусок текст, в котором есть одна или несколько ошибок и этот же кусок текста с исправленными ошибками.
|
| 10 |
+
Твоя задача - кратко описать суть каждой ошибки и, если необходимо, исправить её. Отвечай на русском языке.
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
### Пример 1:
|
| 14 |
+
original text:
|
| 15 |
+
"кросивую"
|
| 16 |
+
|
| 17 |
+
corrected text:
|
| 18 |
+
"красивую"
|
| 19 |
+
|
| 20 |
+
Твой ответ:
|
| 21 |
+
Орфографическая ошибка: "кросивую" -> "красивую".
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
### Пример 2:
|
| 25 |
+
original:
|
| 26 |
+
"я решил пойти в"
|
| 27 |
+
|
| 28 |
+
corrected text:
|
| 29 |
+
"я решил пойти в"
|
| 30 |
+
|
| 31 |
+
Твой ответ:
|
| 32 |
+
Опечатка: повтор пробла.
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
Теперь твоя очередь:
|
| 36 |
+
original:
|
| 37 |
+
"{}"
|
| 38 |
+
|
| 39 |
+
corrected text:
|
| 40 |
+
"{}"
|
| 41 |
+
"""
|
| 42 |
|
| 43 |
prompt = """
|
| 44 |
Ты должен писать комментарии об ошибках в тексте.
|
|
|
|
| 83 |
"""
|
| 84 |
|
| 85 |
|
| 86 |
+
def get_gpt_response_openai(inp):
|
| 87 |
response = openai.ChatCompletion.create(
|
| 88 |
model="gpt-4o-mini",
|
| 89 |
messages=[
|
|
|
|
| 93 |
return response.choices[0].message['content']
|
| 94 |
|
| 95 |
|
| 96 |
+
def get_gpt_response_vsegpt(inp):
|
| 97 |
client = OpenAI(
|
| 98 |
api_key='sk-or-vv-44ca318747a45e810149bf769e9dfbe5f42046695875efd7c9121cca590d6906',
|
| 99 |
base_url='https://api.vsegpt.ru/v1'
|
|
|
|
| 106 |
return response
|
| 107 |
|
| 108 |
|
| 109 |
+
def get_gpt_response(inp, client_name):
|
| 110 |
if client_name == "openai":
|
| 111 |
+
return get_gpt_response_openai(inp)
|
| 112 |
elif client_name == "vsegpt":
|
| 113 |
+
return get_gpt_response_vsegpt(inp)
|
| 114 |
else:
|
| 115 |
raise ValueError(f"Unsupported client: {client_name}")
|
| 116 |
|
| 117 |
|
| 118 |
+
def find_corrected_positions(original, corrected):
|
| 119 |
+
matcher = difflib.SequenceMatcher(None, original, corrected)
|
| 120 |
+
changes = []
|
| 121 |
+
|
| 122 |
+
for opcode in matcher.get_opcodes():
|
| 123 |
+
tag, i1, i2, j1, j2 = opcode
|
| 124 |
+
if tag != 'equal':
|
| 125 |
+
changes.append({
|
| 126 |
+
'original': (i1, i2),
|
| 127 |
+
'corrected': (j1, j2),
|
| 128 |
+
'operation': tag
|
| 129 |
+
})
|
| 130 |
+
|
| 131 |
+
return changes
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
def get_piece_of_text_bounds(s, start, end):
|
| 135 |
+
if start != 0: start -= 1
|
| 136 |
+
while start != 0 and s[start] not in [' ', "\n", '\t']:
|
| 137 |
+
start -= 1
|
| 138 |
+
start += 1
|
| 139 |
+
|
| 140 |
+
if end < len(s) - 1: end += 1
|
| 141 |
+
while end < len(s) - 1 and s[end] not in [' ', "\n", '\t']:
|
| 142 |
+
end += 1
|
| 143 |
+
return start, end
|
| 144 |
+
|
| 145 |
+
|
| 146 |
+
def add_comments_to_text(text, errors):
|
| 147 |
+
errors = sorted(errors, key=lambda x: x['end'])
|
| 148 |
+
|
| 149 |
+
shift = 0
|
| 150 |
+
for i, error in enumerate(errors, 1):
|
| 151 |
+
error['start'] += shift
|
| 152 |
+
error['end'] += shift
|
| 153 |
+
inp = f"({i})"
|
| 154 |
+
text = text[:error['end']] + inp + text[error['end']:]
|
| 155 |
+
shift += len(inp)
|
| 156 |
+
|
| 157 |
+
return text.replace("\n", " ")
|
| 158 |
+
|
| 159 |
+
|
| 160 |
+
def check_text_chat_gpt(text, *args, **kwargs):
|
| 161 |
+
fixed_text = get_gpt_response(prompt_fix_text_gpt.format(text), "vsegpt")
|
| 162 |
+
changes = find_corrected_positions(text, fixed_text)
|
| 163 |
+
errors = []
|
| 164 |
+
for change in changes:
|
| 165 |
+
start_orig, end_orid = get_piece_of_text_bounds(text, change['original'][0], change['original'][1])
|
| 166 |
+
start_corr, end_corr = get_piece_of_text_bounds(fixed_text, change['corrected'][0], change['corrected'][1])
|
| 167 |
+
inp = prompt_compare_get_comment.format(text[start_orig:end_orid], fixed_text[start_corr:end_corr])
|
| 168 |
+
errors.append({
|
| 169 |
+
'start': start_orig,
|
| 170 |
+
'end': end_orid,
|
| 171 |
+
'message': get_gpt_response(inp, client_name="vsegpt"),
|
| 172 |
+
})
|
| 173 |
+
|
| 174 |
+
text_with_comments = add_comments_to_text(text, errors)
|
| 175 |
+
return text_with_comments, errors
|
| 176 |
+
|
| 177 |
def check_text(text, tool):
|
| 178 |
matches = tool.check(text)
|
| 179 |
errors = []
|
|
|
|
| 183 |
error_info = {
|
| 184 |
'start': match.offset,
|
| 185 |
'end': match.offset + match.errorLength,
|
| 186 |
+
'message': get_gpt_response(inp, client_name="vsegpt"),
|
| 187 |
}
|
| 188 |
errors.append(error_info)
|
| 189 |
|
| 190 |
+
text_with_comments = add_comments_to_text(text, errors)
|
| 191 |
+
return text_with_comments, errors
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|