m-ric HF staff commited on
Commit
a45a1f9
β€’
1 Parent(s): 8e8c050

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +8 -8
app.py CHANGED
@@ -392,14 +392,14 @@ def generate_beams(start_sentence, scores, length_penalty, decoded_sequences):
392
  return original_tree
393
 
394
  @spaces.GPU
395
- def get_beam_search_html(input_text, number_steps, number_beams, length_penalty, number_sequences):
396
  inputs = tokenizer([input_text], return_tensors="pt")
397
 
398
  outputs = model.generate(
399
  **inputs,
400
  max_new_tokens=number_steps,
401
  num_beams=number_beams,
402
- num_return_sequences=number_sequences,
403
  return_dict_in_generate=True,
404
  length_penalty=length_penalty,
405
  output_scores=True,
@@ -425,7 +425,7 @@ def get_beam_search_html(input_text, number_steps, number_beams, length_penalty,
425
  return html, markdown
426
 
427
 
428
- def change_n_sequences(n_beams):
429
  return gr.Slider(label="Number of sequences", minimum=1, maximum=n_beams, step=1, value=n_beams)
430
 
431
 
@@ -447,7 +447,7 @@ Play with the parameters below to understand how beam search decoding works!
447
  - **Number of beams** (`num_beams`): the number of beams to use
448
  - **Length penalty** (`length_penalty`): the length penalty to apply to outputs. `length_penalty` > 0.0 promotes longer sequences, while `length_penalty` < 0.0 encourages shorter sequences.
449
  This parameter will not impact the beam search paths, but only influence the choice of sequences in the end towards longer or shorter sequences.
450
- - **Number of sequences** (`num_return_sequences`): the number of sequences to be returned at the end of generation.
451
  """
452
  )
453
  text = gr.Textbox(
@@ -464,17 +464,17 @@ This parameter will not impact the beam search paths, but only influence the cho
464
  length_penalty = gr.Slider(
465
  label="Length penalty", minimum=-3, maximum=3, step=0.5, value=1
466
  )
467
- n_sequences = gr.Slider(
468
- label="Number of sequences", minimum=1, maximum=4, step=1, value=3
469
  )
470
 
471
- n_beams.change(fn=change_n_sequences, inputs=n_beams, outputs=n_sequences)
472
  button = gr.Button()
473
  out_html = gr.Markdown()
474
  out_markdown = gr.Markdown()
475
  button.click(
476
  get_beam_search_html,
477
- inputs=[text, n_steps, n_beams, length_penalty, n_sequences],
478
  outputs=[out_html, out_markdown],
479
  )
480
 
 
392
  return original_tree
393
 
394
  @spaces.GPU
395
+ def get_beam_search_html(input_text, number_steps, number_beams, length_penalty, num_return_sequences):
396
  inputs = tokenizer([input_text], return_tensors="pt")
397
 
398
  outputs = model.generate(
399
  **inputs,
400
  max_new_tokens=number_steps,
401
  num_beams=number_beams,
402
+ num_return_sequences=num_return_sequences,
403
  return_dict_in_generate=True,
404
  length_penalty=length_penalty,
405
  output_scores=True,
 
425
  return html, markdown
426
 
427
 
428
+ def change_num_return_sequences(n_beams):
429
  return gr.Slider(label="Number of sequences", minimum=1, maximum=n_beams, step=1, value=n_beams)
430
 
431
 
 
447
  - **Number of beams** (`num_beams`): the number of beams to use
448
  - **Length penalty** (`length_penalty`): the length penalty to apply to outputs. `length_penalty` > 0.0 promotes longer sequences, while `length_penalty` < 0.0 encourages shorter sequences.
449
  This parameter will not impact the beam search paths, but only influence the choice of sequences in the end towards longer or shorter sequences.
450
+ - **Number of return sequences** (`num_return_sequences`): the number of sequences to be returned at the end of generation. Should be `<= num_beams'
451
  """
452
  )
453
  text = gr.Textbox(
 
464
  length_penalty = gr.Slider(
465
  label="Length penalty", minimum=-3, maximum=3, step=0.5, value=1
466
  )
467
+ num_return_sequences = gr.Slider(
468
+ label="Number of return sequences", minimum=1, maximum=4, step=1, value=3
469
  )
470
 
471
+ n_beams.change(fn=change_num_return_sequences, inputs=n_beams, outputs=num_return_sequences)
472
  button = gr.Button()
473
  out_html = gr.Markdown()
474
  out_markdown = gr.Markdown()
475
  button.click(
476
  get_beam_search_html,
477
+ inputs=[text, n_steps, n_beams, length_penalty, num_return_sequences],
478
  outputs=[out_html, out_markdown],
479
  )
480