Spaces:
Running
on
T4
Running
on
T4
Update app.py
Browse files
app.py
CHANGED
|
@@ -371,9 +371,10 @@ def get_beam_search_html(input_text, number_steps, number_beams, length_penalty)
|
|
| 371 |
output_scores=True,
|
| 372 |
do_sample=False,
|
| 373 |
)
|
| 374 |
-
|
| 375 |
-
|
| 376 |
-
|
|
|
|
| 377 |
|
| 378 |
original_tree = generate_beams(
|
| 379 |
input_text,
|
|
@@ -382,7 +383,7 @@ def get_beam_search_html(input_text, number_steps, number_beams, length_penalty)
|
|
| 382 |
length_penalty,
|
| 383 |
)
|
| 384 |
html = generate_html(input_text, original_tree)
|
| 385 |
-
return html
|
| 386 |
|
| 387 |
|
| 388 |
with gr.Blocks(
|
|
@@ -391,13 +392,24 @@ with gr.Blocks(
|
|
| 391 |
),
|
| 392 |
css=STYLE,
|
| 393 |
) as demo:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 394 |
text = gr.Textbox(label="Sentence to decode from", value="Today is")
|
| 395 |
with gr.Row():
|
| 396 |
steps = gr.Slider(label="Number of steps", minimum=1, maximum=8, step=1, value=4)
|
| 397 |
beams = gr.Slider(label="Number of beams", minimum=2, maximum=4, step=1, value=3)
|
| 398 |
length_penalty = gr.Slider(label="Length penalty", minimum=-5, maximum=5, step=0.5, value=1)
|
| 399 |
button = gr.Button()
|
| 400 |
-
|
| 401 |
-
|
|
|
|
| 402 |
|
| 403 |
demo.launch()
|
|
|
|
| 371 |
output_scores=True,
|
| 372 |
do_sample=False,
|
| 373 |
)
|
| 374 |
+
markdown = "Sequences:"
|
| 375 |
+
decoded_sequences = tokenizer.batch_decode(outputs.sequences)
|
| 376 |
+
for i, sequence in enumerate(decoded_sequences):
|
| 377 |
+
markdown += f"\n- {sequence} ( score {outputs.sequences_scores[i]:.2f})"
|
| 378 |
|
| 379 |
original_tree = generate_beams(
|
| 380 |
input_text,
|
|
|
|
| 383 |
length_penalty,
|
| 384 |
)
|
| 385 |
html = generate_html(input_text, original_tree)
|
| 386 |
+
return html, markdown
|
| 387 |
|
| 388 |
|
| 389 |
with gr.Blocks(
|
|
|
|
| 392 |
),
|
| 393 |
css=STYLE,
|
| 394 |
) as demo:
|
| 395 |
+
gr.Markdown("""# Beam search visualizer
|
| 396 |
+
|
| 397 |
+
Play with the parameters below to understand how beam search decoding works!
|
| 398 |
+
|
| 399 |
+
#### Parameters:
|
| 400 |
+
- **Sentence to decode from**: the input sequence to your decoder.
|
| 401 |
+
- **Number of steps**: the number of tokens to generate
|
| 402 |
+
- **Number of beams**: the number of beams to use
|
| 403 |
+
- **Length penalty**: the length penalty to apply to outputs. `length_penalty` > 0.0 promotes longer sequences, while `length_penalty` < 0.0 encourages shorter sequences.
|
| 404 |
+
""")
|
| 405 |
text = gr.Textbox(label="Sentence to decode from", value="Today is")
|
| 406 |
with gr.Row():
|
| 407 |
steps = gr.Slider(label="Number of steps", minimum=1, maximum=8, step=1, value=4)
|
| 408 |
beams = gr.Slider(label="Number of beams", minimum=2, maximum=4, step=1, value=3)
|
| 409 |
length_penalty = gr.Slider(label="Length penalty", minimum=-5, maximum=5, step=0.5, value=1)
|
| 410 |
button = gr.Button()
|
| 411 |
+
out_html = gr.Markdown()
|
| 412 |
+
out_markdown = gr.Markdown()
|
| 413 |
+
button.click(get_beam_search_html, inputs=[text, steps, beams, length_penalty], outputs=[out_html, out_markdown])
|
| 414 |
|
| 415 |
demo.launch()
|