NoaiGPT commited on
Commit
9ddb2ad
1 Parent(s): 580ef8a
Files changed (1) hide show
  1. app.py +20 -8
app.py CHANGED
@@ -577,23 +577,33 @@ def generate_paraphrases(text, setting, output_format):
577
 
578
  if setting == 1:
579
  num_return_sequences = 3
580
- num_beams = 5
 
 
581
  max_length = 128
582
  elif setting == 2:
583
  num_return_sequences = 3
584
- num_beams = 7
 
 
585
  max_length = 192
586
  elif setting == 3:
587
  num_return_sequences = 3
588
- num_beams = 9
 
 
589
  max_length = 256
590
  elif setting == 4:
591
  num_return_sequences = 3
592
- num_beams = 11
 
 
593
  max_length = 320
594
  else:
595
  num_return_sequences = 3
596
- num_beams = 15
 
 
597
  max_length = 384
598
 
599
  formatted_output = "Original text:\n" + text + "\n\n"
@@ -616,9 +626,11 @@ def generate_paraphrases(text, setting, output_format):
616
  input_ids=input_ids,
617
  attention_mask=attention_mask,
618
  max_length=max_length,
619
- early_stopping=True,
620
- num_beams=num_beams,
621
- num_return_sequences=num_return_sequences
 
 
622
  )
623
 
624
  paraphrases_texts = [paraphraser_tokenizer.decode(beam_output, skip_special_tokens=True, clean_up_tokenization_spaces=True) for beam_output in beam_outputs]
 
577
 
578
  if setting == 1:
579
  num_return_sequences = 3
580
+ temperature = 0.7
581
+ top_k = 50
582
+ top_p = 0.9
583
  max_length = 128
584
  elif setting == 2:
585
  num_return_sequences = 3
586
+ temperature = 0.8
587
+ top_k = 50
588
+ top_p = 0.9
589
  max_length = 192
590
  elif setting == 3:
591
  num_return_sequences = 3
592
+ temperature = 0.9
593
+ top_k = 50
594
+ top_p = 0.9
595
  max_length = 256
596
  elif setting == 4:
597
  num_return_sequences = 3
598
+ temperature = 1.0
599
+ top_k = 50
600
+ top_p = 0.9
601
  max_length = 320
602
  else:
603
  num_return_sequences = 3
604
+ temperature = 1.1
605
+ top_k = 50
606
+ top_p = 0.9
607
  max_length = 384
608
 
609
  formatted_output = "Original text:\n" + text + "\n\n"
 
626
  input_ids=input_ids,
627
  attention_mask=attention_mask,
628
  max_length=max_length,
629
+ num_return_sequences=num_return_sequences,
630
+ do_sample=True,
631
+ top_k=top_k,
632
+ top_p=top_p,
633
+ temperature=temperature
634
  )
635
 
636
  paraphrases_texts = [paraphraser_tokenizer.decode(beam_output, skip_special_tokens=True, clean_up_tokenization_spaces=True) for beam_output in beam_outputs]