Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
|
@@ -297,8 +297,8 @@ class HumanLikeVariations:
|
|
| 297 |
openers = ["Look,", "See,", "Thing is,", "Honestly,", "Actually,"]
|
| 298 |
sentence = random.choice(openers) + " " + sentence[0].lower() + sentence[1:]
|
| 299 |
|
| 300 |
-
# Add subtle errors that humans make (
|
| 301 |
-
if random.random() < 0.
|
| 302 |
words = sentence.split()
|
| 303 |
if len(words) > 5:
|
| 304 |
# Common comma omissions
|
|
@@ -560,8 +560,8 @@ class EnhancedDipperHumanizer:
|
|
| 560 |
# Natural contractions throughout
|
| 561 |
sentence = self.apply_contractions(sentence)
|
| 562 |
|
| 563 |
-
# Add natural speech patterns (15% chance)
|
| 564 |
-
if random.random() < 0.
|
| 565 |
# Natural interruptions that humans actually use
|
| 566 |
if random.random() < 0.5:
|
| 567 |
# Add "you know" or "I mean" naturally
|
|
@@ -878,13 +878,13 @@ class EnhancedDipperHumanizer:
|
|
| 878 |
continue
|
| 879 |
|
| 880 |
try:
|
| 881 |
-
#
|
| 882 |
if len(sentence.split()) < 10:
|
| 883 |
-
lex_diversity =
|
| 884 |
-
order_diversity =
|
| 885 |
else:
|
| 886 |
-
lex_diversity =
|
| 887 |
-
order_diversity =
|
| 888 |
|
| 889 |
lex_code = int(100 - lex_diversity)
|
| 890 |
order_code = int(100 - order_diversity)
|
|
@@ -916,8 +916,8 @@ class EnhancedDipperHumanizer:
|
|
| 916 |
max_new_length = int(original_length * 1.4)
|
| 917 |
|
| 918 |
# High variation parameters
|
| 919 |
-
temp = 0.
|
| 920 |
-
top_p_val = 0.
|
| 921 |
|
| 922 |
with torch.no_grad():
|
| 923 |
outputs = self.model.generate(
|
|
@@ -1373,8 +1373,8 @@ class EnhancedDipperHumanizer:
|
|
| 1373 |
if not sentence.strip():
|
| 1374 |
continue
|
| 1375 |
|
| 1376 |
-
# Add stream-of-consciousness elements (
|
| 1377 |
-
if random.random() < 0.
|
| 1378 |
stream_elements = [
|
| 1379 |
" - wait, let me back up - ",
|
| 1380 |
" - actually, scratch that - ",
|
|
@@ -1388,8 +1388,8 @@ class EnhancedDipperHumanizer:
|
|
| 1388 |
words.insert(pos, random.choice(stream_elements))
|
| 1389 |
sentence = ' '.join(words)
|
| 1390 |
|
| 1391 |
-
# Add human-like self-corrections (
|
| 1392 |
-
if random.random() < 0.
|
| 1393 |
corrections = [
|
| 1394 |
" - or rather, ",
|
| 1395 |
" - well, actually, ",
|
|
@@ -1407,8 +1407,8 @@ class EnhancedDipperHumanizer:
|
|
| 1407 |
words.insert(pos, correction)
|
| 1408 |
sentence = ' '.join(words)
|
| 1409 |
|
| 1410 |
-
# Add thinking-out-loud patterns (
|
| 1411 |
-
if random.random() < 0.
|
| 1412 |
thinking_patterns = [
|
| 1413 |
"Come to think of it, ",
|
| 1414 |
"Actually, you know what? ",
|
|
@@ -1483,10 +1483,10 @@ class EnhancedDipperHumanizer:
|
|
| 1483 |
order_diversity=20
|
| 1484 |
)
|
| 1485 |
|
| 1486 |
-
# Second pass with BART for longer texts (
|
| 1487 |
if self.use_bart and len(paraphrased_text.split()) > 8:
|
| 1488 |
-
#
|
| 1489 |
-
if random.random() < 0.
|
| 1490 |
paraphrased_text = self.paraphrase_with_bart(paraphrased_text)
|
| 1491 |
|
| 1492 |
# Apply sentence variation
|
|
|
|
| 297 |
openers = ["Look,", "See,", "Thing is,", "Honestly,", "Actually,"]
|
| 298 |
sentence = random.choice(openers) + " " + sentence[0].lower() + sentence[1:]
|
| 299 |
|
| 300 |
+
# Add subtle errors that humans make (10% chance - reduced)
|
| 301 |
+
if random.random() < 0.10:
|
| 302 |
words = sentence.split()
|
| 303 |
if len(words) > 5:
|
| 304 |
# Common comma omissions
|
|
|
|
| 560 |
# Natural contractions throughout
|
| 561 |
sentence = self.apply_contractions(sentence)
|
| 562 |
|
| 563 |
+
# Add natural speech patterns (15% chance - balanced)
|
| 564 |
+
if random.random() < 0.15 and len(sentence.split()) > 10:
|
| 565 |
# Natural interruptions that humans actually use
|
| 566 |
if random.random() < 0.5:
|
| 567 |
# Add "you know" or "I mean" naturally
|
|
|
|
| 878 |
continue
|
| 879 |
|
| 880 |
try:
|
| 881 |
+
# BALANCED diversity for Originality AI (100% human with better quality)
|
| 882 |
if len(sentence.split()) < 10:
|
| 883 |
+
lex_diversity = 70 # High but not extreme
|
| 884 |
+
order_diversity = 25
|
| 885 |
else:
|
| 886 |
+
lex_diversity = 82 # Balanced diversity
|
| 887 |
+
order_diversity = 30 # Moderate order diversity
|
| 888 |
|
| 889 |
lex_code = int(100 - lex_diversity)
|
| 890 |
order_code = int(100 - order_diversity)
|
|
|
|
| 916 |
max_new_length = int(original_length * 1.4)
|
| 917 |
|
| 918 |
# High variation parameters
|
| 919 |
+
temp = 0.85 # Slightly reduced from 0.9
|
| 920 |
+
top_p_val = 0.92 # Slightly reduced from 0.95
|
| 921 |
|
| 922 |
with torch.no_grad():
|
| 923 |
outputs = self.model.generate(
|
|
|
|
| 1373 |
if not sentence.strip():
|
| 1374 |
continue
|
| 1375 |
|
| 1376 |
+
# Add stream-of-consciousness elements (8% chance - reduced)
|
| 1377 |
+
if random.random() < 0.08 and len(sentence.split()) > 10:
|
| 1378 |
stream_elements = [
|
| 1379 |
" - wait, let me back up - ",
|
| 1380 |
" - actually, scratch that - ",
|
|
|
|
| 1388 |
words.insert(pos, random.choice(stream_elements))
|
| 1389 |
sentence = ' '.join(words)
|
| 1390 |
|
| 1391 |
+
# Add human-like self-corrections (7% chance - reduced)
|
| 1392 |
+
if random.random() < 0.07:
|
| 1393 |
corrections = [
|
| 1394 |
" - or rather, ",
|
| 1395 |
" - well, actually, ",
|
|
|
|
| 1407 |
words.insert(pos, correction)
|
| 1408 |
sentence = ' '.join(words)
|
| 1409 |
|
| 1410 |
+
# Add thinking-out-loud patterns (10% chance - reduced)
|
| 1411 |
+
if random.random() < 0.10 and i > 0:
|
| 1412 |
thinking_patterns = [
|
| 1413 |
"Come to think of it, ",
|
| 1414 |
"Actually, you know what? ",
|
|
|
|
| 1483 |
order_diversity=20
|
| 1484 |
)
|
| 1485 |
|
| 1486 |
+
# Second pass with BART for longer texts (balanced probability)
|
| 1487 |
if self.use_bart and len(paraphrased_text.split()) > 8:
|
| 1488 |
+
# 30% chance to use BART for more variation (balanced)
|
| 1489 |
+
if random.random() < 0.3:
|
| 1490 |
paraphrased_text = self.paraphrase_with_bart(paraphrased_text)
|
| 1491 |
|
| 1492 |
# Apply sentence variation
|