Update app.py
Browse files
app.py
CHANGED
|
@@ -1,32 +1,46 @@
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from parrot import Parrot
|
| 3 |
import nltk
|
| 4 |
from nltk.tokenize import sent_tokenize, word_tokenize
|
| 5 |
import re
|
| 6 |
import time
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
# -----------------------------
|
| 9 |
# Setup
|
| 10 |
# -----------------------------
|
| 11 |
-
|
|
|
|
| 12 |
parrot = None # Lazy-loaded global model
|
| 13 |
|
|
|
|
|
|
|
|
|
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
def get_parrot():
|
| 16 |
-
"""Load the Parrot model lazily"""
|
| 17 |
global parrot
|
| 18 |
if parrot is None:
|
| 19 |
-
print("β³ Loading Parrot model
|
| 20 |
parrot = Parrot(model_tag="prithivida/parrot_paraphraser_on_T5", use_gpu=False)
|
| 21 |
-
print("β
Parrot model loaded
|
| 22 |
return parrot
|
| 23 |
|
| 24 |
|
| 25 |
-
MAX_TOKENS = 150 # limit per chunk for stability
|
| 26 |
-
|
| 27 |
-
|
| 28 |
# -----------------------------
|
| 29 |
-
# Helper
|
| 30 |
# -----------------------------
|
| 31 |
def clean_sentence(sent):
|
| 32 |
sent = sent.strip()
|
|
@@ -48,11 +62,7 @@ def split_long_sentence(sentence, max_tokens=MAX_TOKENS):
|
|
| 48 |
return [" ".join(words[i:i + max_tokens]) for i in range(0, len(words), max_tokens)]
|
| 49 |
|
| 50 |
|
| 51 |
-
# -----------------------------
|
| 52 |
-
# πΉ Retry Wrapper
|
| 53 |
-
# -----------------------------
|
| 54 |
def with_retry(func, *args, retries=1, delay=3, **kwargs):
|
| 55 |
-
"""Try running a function twice before giving up"""
|
| 56 |
for attempt in range(retries + 1):
|
| 57 |
try:
|
| 58 |
return func(*args, **kwargs)
|
|
@@ -65,7 +75,7 @@ def with_retry(func, *args, retries=1, delay=3, **kwargs):
|
|
| 65 |
|
| 66 |
|
| 67 |
# -----------------------------
|
| 68 |
-
# πΉ
|
| 69 |
# -----------------------------
|
| 70 |
def rephrase(text):
|
| 71 |
model = get_parrot()
|
|
@@ -93,7 +103,7 @@ def rephrase(text):
|
|
| 93 |
|
| 94 |
|
| 95 |
# -----------------------------
|
| 96 |
-
# πΉ
|
| 97 |
# -----------------------------
|
| 98 |
def generate_unique_paraphrases(sentence, N_OPTIONS=3):
|
| 99 |
model = get_parrot()
|
|
@@ -104,10 +114,8 @@ def generate_unique_paraphrases(sentence, N_OPTIONS=3):
|
|
| 104 |
adequacy_threshold=0.85,
|
| 105 |
fluency_threshold=0.9,
|
| 106 |
)
|
| 107 |
-
|
| 108 |
if not paraphrases:
|
| 109 |
return [sentence]
|
| 110 |
-
|
| 111 |
texts = [p[0] for p in paraphrases]
|
| 112 |
unique = []
|
| 113 |
for t in texts:
|
|
@@ -132,12 +140,73 @@ def rephrase_sentencewise_unique(text, N_OPTIONS=3):
|
|
| 132 |
|
| 133 |
|
| 134 |
# -----------------------------
|
| 135 |
-
# πΉ
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
# -----------------------------
|
| 137 |
def warmup():
|
| 138 |
-
"""Ping the model once to prevent timeout at first request"""
|
| 139 |
-
print("π₯ Warming up Parrot model...")
|
| 140 |
try:
|
|
|
|
| 141 |
model = get_parrot()
|
| 142 |
_ = model.augment(input_phrase="hello world", do_diverse=False)
|
| 143 |
print("β
Warmup complete.")
|
|
@@ -149,30 +218,35 @@ warmup()
|
|
| 149 |
|
| 150 |
|
| 151 |
# -----------------------------
|
| 152 |
-
# πΉ Gradio
|
| 153 |
# -----------------------------
|
| 154 |
rephrase_iface = gr.Interface(
|
| 155 |
fn=rephrase,
|
| 156 |
inputs=gr.Textbox(lines=10, placeholder="Paste your text here..."),
|
| 157 |
outputs="text",
|
| 158 |
-
title="Parrot Rephraser (Long Text)",
|
| 159 |
-
description="
|
| 160 |
)
|
| 161 |
|
| 162 |
sentencewise_iface = gr.Interface(
|
| 163 |
fn=rephrase_sentencewise_unique,
|
| 164 |
inputs=gr.Textbox(lines=10, placeholder="Paste text here..."),
|
| 165 |
outputs="text",
|
| 166 |
-
title="
|
| 167 |
-
description="Generates top 3
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 168 |
)
|
| 169 |
|
| 170 |
-
# -----------------------------
|
| 171 |
-
# πΉ Combine both interfaces into Tabs
|
| 172 |
-
# -----------------------------
|
| 173 |
demo = gr.TabbedInterface(
|
| 174 |
-
[rephrase_iface, sentencewise_iface],
|
| 175 |
-
["Full Text Rephraser", "Sentence-wise Paraphrases"],
|
| 176 |
)
|
| 177 |
|
| 178 |
demo.launch(server_port=7860, server_name="0.0.0.0", show_error=True)
|
|
|
|
| 1 |
+
# main.py
|
| 2 |
import gradio as gr
|
| 3 |
from parrot import Parrot
|
| 4 |
import nltk
|
| 5 |
from nltk.tokenize import sent_tokenize, word_tokenize
|
| 6 |
import re
|
| 7 |
import time
|
| 8 |
+
import os
|
| 9 |
+
from ddgs import DDGS
|
| 10 |
+
from googleapiclient.discovery import build
|
| 11 |
+
from dotenv import load_dotenv
|
| 12 |
|
| 13 |
# -----------------------------
|
| 14 |
# Setup
|
| 15 |
# -----------------------------
|
| 16 |
+
load_dotenv()
|
| 17 |
+
nltk.data.path.append("./nltk_data") # Local punkt
|
| 18 |
parrot = None # Lazy-loaded global model
|
| 19 |
|
| 20 |
+
# Load Google credentials
|
| 21 |
+
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
|
| 22 |
+
GOOGLE_CX = os.getenv("GOOGLE_CX")
|
| 23 |
|
| 24 |
+
if not GOOGLE_API_KEY or not GOOGLE_CX:
|
| 25 |
+
print("β οΈ Warning: GOOGLE_API_KEY or GOOGLE_CX not set. Google fallback may fail.")
|
| 26 |
+
|
| 27 |
+
MAX_TOKENS = 150
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
# -----------------------------
|
| 31 |
+
# Parrot Model Loader
|
| 32 |
+
# -----------------------------
|
| 33 |
def get_parrot():
|
|
|
|
| 34 |
global parrot
|
| 35 |
if parrot is None:
|
| 36 |
+
print("β³ Loading Parrot model...")
|
| 37 |
parrot = Parrot(model_tag="prithivida/parrot_paraphraser_on_T5", use_gpu=False)
|
| 38 |
+
print("β
Parrot model loaded!")
|
| 39 |
return parrot
|
| 40 |
|
| 41 |
|
|
|
|
|
|
|
|
|
|
| 42 |
# -----------------------------
|
| 43 |
+
# Helper Functions
|
| 44 |
# -----------------------------
|
| 45 |
def clean_sentence(sent):
|
| 46 |
sent = sent.strip()
|
|
|
|
| 62 |
return [" ".join(words[i:i + max_tokens]) for i in range(0, len(words), max_tokens)]
|
| 63 |
|
| 64 |
|
|
|
|
|
|
|
|
|
|
| 65 |
def with_retry(func, *args, retries=1, delay=3, **kwargs):
|
|
|
|
| 66 |
for attempt in range(retries + 1):
|
| 67 |
try:
|
| 68 |
return func(*args, **kwargs)
|
|
|
|
| 75 |
|
| 76 |
|
| 77 |
# -----------------------------
|
| 78 |
+
# πΉ Full Paragraph Rephraser
|
| 79 |
# -----------------------------
|
| 80 |
def rephrase(text):
|
| 81 |
model = get_parrot()
|
|
|
|
| 103 |
|
| 104 |
|
| 105 |
# -----------------------------
|
| 106 |
+
# πΉ Sentence-wise Paraphrases
|
| 107 |
# -----------------------------
|
| 108 |
def generate_unique_paraphrases(sentence, N_OPTIONS=3):
|
| 109 |
model = get_parrot()
|
|
|
|
| 114 |
adequacy_threshold=0.85,
|
| 115 |
fluency_threshold=0.9,
|
| 116 |
)
|
|
|
|
| 117 |
if not paraphrases:
|
| 118 |
return [sentence]
|
|
|
|
| 119 |
texts = [p[0] for p in paraphrases]
|
| 120 |
unique = []
|
| 121 |
for t in texts:
|
|
|
|
| 140 |
|
| 141 |
|
| 142 |
# -----------------------------
|
| 143 |
+
# πΉ Hybrid Plagiarism Detection (DuckDuckGo + Google)
|
| 144 |
+
# -----------------------------
|
| 145 |
+
def search_duckduckgo(query):
|
| 146 |
+
try:
|
| 147 |
+
with DDGS() as ddgs:
|
| 148 |
+
results = list(ddgs.text(f'"{query}"', max_results=3))
|
| 149 |
+
return results, len(results)
|
| 150 |
+
except Exception as e:
|
| 151 |
+
print(f"β οΈ DDG error: {e}")
|
| 152 |
+
return [], 0
|
| 153 |
+
|
| 154 |
+
|
| 155 |
+
def detect_duckduckgo(text):
|
| 156 |
+
sentences = sent_tokenize(text)
|
| 157 |
+
matches = []
|
| 158 |
+
for sent in sentences:
|
| 159 |
+
results, total = search_duckduckgo(sent)
|
| 160 |
+
if total > 0:
|
| 161 |
+
url = results[0].get("href", "Unknown")
|
| 162 |
+
matches.append((sent, total, url))
|
| 163 |
+
time.sleep(1)
|
| 164 |
+
return matches
|
| 165 |
+
|
| 166 |
+
|
| 167 |
+
def search_google(query):
|
| 168 |
+
if not GOOGLE_API_KEY or not GOOGLE_CX:
|
| 169 |
+
return [], 0
|
| 170 |
+
try:
|
| 171 |
+
service = build("customsearch", "v1", developerKey=GOOGLE_API_KEY)
|
| 172 |
+
res = service.cse().list(q=f'"{query}"', cx=GOOGLE_CX, num=3).execute()
|
| 173 |
+
items = res.get("items", [])
|
| 174 |
+
total = int(res.get("searchInformation", {}).get("totalResults", 0))
|
| 175 |
+
return items, total
|
| 176 |
+
except Exception as e:
|
| 177 |
+
print(f"β οΈ Google search error: {e}")
|
| 178 |
+
return [], 0
|
| 179 |
+
|
| 180 |
+
|
| 181 |
+
def detect_google(text):
|
| 182 |
+
sentences = sent_tokenize(text)
|
| 183 |
+
matches = []
|
| 184 |
+
for sent in sentences:
|
| 185 |
+
results, total = search_google(sent)
|
| 186 |
+
if total > 0:
|
| 187 |
+
url = results[0].get("link", "Unknown")
|
| 188 |
+
matches.append((sent, total, url))
|
| 189 |
+
time.sleep(0.3)
|
| 190 |
+
return matches
|
| 191 |
+
|
| 192 |
+
|
| 193 |
+
def hybrid_detect(text):
|
| 194 |
+
ddg_matches = detect_duckduckgo(text)
|
| 195 |
+
matches = ddg_matches if ddg_matches else detect_google(text)
|
| 196 |
+
highlighted = text
|
| 197 |
+
urls = []
|
| 198 |
+
for sent, _, url in matches:
|
| 199 |
+
highlighted = highlighted.replace(sent, f"**{sent}**")
|
| 200 |
+
urls.append(url)
|
| 201 |
+
return {"highlighted_text": highlighted, "sources": urls}
|
| 202 |
+
|
| 203 |
+
|
| 204 |
+
# -----------------------------
|
| 205 |
+
# Warm-up Parrot Model
|
| 206 |
# -----------------------------
|
| 207 |
def warmup():
|
|
|
|
|
|
|
| 208 |
try:
|
| 209 |
+
print("π₯ Warming up Parrot model...")
|
| 210 |
model = get_parrot()
|
| 211 |
_ = model.augment(input_phrase="hello world", do_diverse=False)
|
| 212 |
print("β
Warmup complete.")
|
|
|
|
| 218 |
|
| 219 |
|
| 220 |
# -----------------------------
|
| 221 |
+
# πΉ Gradio UI
|
| 222 |
# -----------------------------
|
| 223 |
rephrase_iface = gr.Interface(
|
| 224 |
fn=rephrase,
|
| 225 |
inputs=gr.Textbox(lines=10, placeholder="Paste your text here..."),
|
| 226 |
outputs="text",
|
| 227 |
+
title="π¦ Parrot Rephraser (Long Text)",
|
| 228 |
+
description="Rephrase paragraphs while maintaining meaning.",
|
| 229 |
)
|
| 230 |
|
| 231 |
sentencewise_iface = gr.Interface(
|
| 232 |
fn=rephrase_sentencewise_unique,
|
| 233 |
inputs=gr.Textbox(lines=10, placeholder="Paste text here..."),
|
| 234 |
outputs="text",
|
| 235 |
+
title="π§© Sentence-wise Paraphraser",
|
| 236 |
+
description="Generates top 3 diverse rephrases per sentence.",
|
| 237 |
+
)
|
| 238 |
+
|
| 239 |
+
plagiarism_iface = gr.Interface(
|
| 240 |
+
fn=hybrid_detect,
|
| 241 |
+
inputs=gr.Textbox(lines=10, placeholder="Paste text to check plagiarism..."),
|
| 242 |
+
outputs=gr.JSON(),
|
| 243 |
+
title="π Hybrid Plagiarism Detector",
|
| 244 |
+
description="Detects copied sentences using DuckDuckGo & Google Custom Search.",
|
| 245 |
)
|
| 246 |
|
|
|
|
|
|
|
|
|
|
| 247 |
demo = gr.TabbedInterface(
|
| 248 |
+
[rephrase_iface, sentencewise_iface, plagiarism_iface],
|
| 249 |
+
["Full Text Rephraser", "Sentence-wise Paraphrases", "Plagiarism Checker"],
|
| 250 |
)
|
| 251 |
|
| 252 |
demo.launch(server_port=7860, server_name="0.0.0.0", show_error=True)
|