Spaces:
Running
Running
Commit
β’
504ac73
1
Parent(s):
a558847
Bring up advanced settings
Browse files- app_dialogue.py +69 -71
app_dialogue.py
CHANGED
@@ -376,80 +376,78 @@ with gr.Blocks(title="AI Dad Jokes", theme=gr.themes.Base()) as demo:
|
|
376 |
clear_btn = gr.ClearButton([textbox, imagebox, chatbot], value="π§Ή Clear")
|
377 |
regenerate_btn = gr.Button(value="π Regenerate", visible=True)
|
378 |
upload_btn = gr.UploadButton("π Upload image", file_types=["image"],visible=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
379 |
with gr.Column():
|
380 |
chatbot.render()
|
381 |
|
382 |
-
with gr.Row():
|
383 |
-
with gr.Accordion("Advanced settings", open=False, visible=True) as parameter_row:
|
384 |
-
system_prompt = gr.Textbox(
|
385 |
-
value=SYSTEM_PROMPT,
|
386 |
-
visible=False,
|
387 |
-
lines=20,
|
388 |
-
max_lines=50,
|
389 |
-
interactive=True,
|
390 |
-
)
|
391 |
-
max_new_tokens = gr.Slider(
|
392 |
-
minimum=8,
|
393 |
-
maximum=256,
|
394 |
-
value=64,
|
395 |
-
step=1,
|
396 |
-
interactive=True,
|
397 |
-
label="Maximum number of new tokens to generate",
|
398 |
-
)
|
399 |
-
repetition_penalty = gr.Slider(
|
400 |
-
minimum=0.0,
|
401 |
-
maximum=5.0,
|
402 |
-
value=1.2,
|
403 |
-
step=0.01,
|
404 |
-
interactive=True,
|
405 |
-
label="Repetition penalty",
|
406 |
-
info="1.0 is equivalent to no penalty",
|
407 |
-
)
|
408 |
-
decoding_strategy = gr.Radio(
|
409 |
-
[
|
410 |
-
"Greedy",
|
411 |
-
"Top P Sampling",
|
412 |
-
],
|
413 |
-
value="Top P Sampling",
|
414 |
-
label="Decoding strategy",
|
415 |
-
interactive=True,
|
416 |
-
info="Higher values is equivalent to sampling more low-probability tokens.",
|
417 |
-
)
|
418 |
-
temperature = gr.Slider(
|
419 |
-
minimum=0.0,
|
420 |
-
maximum=5.0,
|
421 |
-
value=0.6,
|
422 |
-
step=0.1,
|
423 |
-
interactive=True,
|
424 |
-
visible=True,
|
425 |
-
label="Sampling temperature",
|
426 |
-
info="Higher values will produce more diverse outputs.",
|
427 |
-
)
|
428 |
-
decoding_strategy.change(
|
429 |
-
fn=lambda selection: gr.Slider.update(
|
430 |
-
visible=(
|
431 |
-
selection in ["contrastive_sampling", "beam_sampling", "Top P Sampling", "sampling_top_k"]
|
432 |
-
)
|
433 |
-
),
|
434 |
-
inputs=decoding_strategy,
|
435 |
-
outputs=temperature,
|
436 |
-
)
|
437 |
-
top_p = gr.Slider(
|
438 |
-
minimum=0.01,
|
439 |
-
maximum=0.99,
|
440 |
-
value=0.8,
|
441 |
-
step=0.01,
|
442 |
-
interactive=True,
|
443 |
-
visible=True,
|
444 |
-
label="Top P",
|
445 |
-
info="Higher values is equivalent to sampling more low-probability tokens.",
|
446 |
-
)
|
447 |
-
decoding_strategy.change(
|
448 |
-
fn=lambda selection: gr.Slider.update(visible=(selection in ["Top P Sampling"])),
|
449 |
-
inputs=decoding_strategy,
|
450 |
-
outputs=top_p,
|
451 |
-
)
|
452 |
-
|
453 |
def model_inference(
|
454 |
model_selector,
|
455 |
system_prompt,
|
|
|
376 |
clear_btn = gr.ClearButton([textbox, imagebox, chatbot], value="π§Ή Clear")
|
377 |
regenerate_btn = gr.Button(value="π Regenerate", visible=True)
|
378 |
upload_btn = gr.UploadButton("π Upload image", file_types=["image"],visible=False)
|
379 |
+
with gr.Accordion("Advanced settings", open=False, visible=True) as parameter_row:
|
380 |
+
system_prompt = gr.Textbox(
|
381 |
+
value=SYSTEM_PROMPT,
|
382 |
+
visible=False,
|
383 |
+
lines=20,
|
384 |
+
max_lines=50,
|
385 |
+
interactive=True,
|
386 |
+
)
|
387 |
+
max_new_tokens = gr.Slider(
|
388 |
+
minimum=8,
|
389 |
+
maximum=256,
|
390 |
+
value=64,
|
391 |
+
step=1,
|
392 |
+
interactive=True,
|
393 |
+
label="Maximum number of new tokens to generate",
|
394 |
+
)
|
395 |
+
repetition_penalty = gr.Slider(
|
396 |
+
minimum=0.0,
|
397 |
+
maximum=5.0,
|
398 |
+
value=1.2,
|
399 |
+
step=0.01,
|
400 |
+
interactive=True,
|
401 |
+
label="Repetition penalty",
|
402 |
+
info="1.0 is equivalent to no penalty",
|
403 |
+
)
|
404 |
+
decoding_strategy = gr.Radio(
|
405 |
+
[
|
406 |
+
"Greedy",
|
407 |
+
"Top P Sampling",
|
408 |
+
],
|
409 |
+
value="Top P Sampling",
|
410 |
+
label="Decoding strategy",
|
411 |
+
interactive=True,
|
412 |
+
info="Higher values is equivalent to sampling more low-probability tokens.",
|
413 |
+
)
|
414 |
+
temperature = gr.Slider(
|
415 |
+
minimum=0.0,
|
416 |
+
maximum=5.0,
|
417 |
+
value=0.6,
|
418 |
+
step=0.1,
|
419 |
+
interactive=True,
|
420 |
+
visible=True,
|
421 |
+
label="Sampling temperature",
|
422 |
+
info="Higher values will produce more diverse outputs.",
|
423 |
+
)
|
424 |
+
decoding_strategy.change(
|
425 |
+
fn=lambda selection: gr.Slider.update(
|
426 |
+
visible=(
|
427 |
+
selection in ["contrastive_sampling", "beam_sampling", "Top P Sampling", "sampling_top_k"]
|
428 |
+
)
|
429 |
+
),
|
430 |
+
inputs=decoding_strategy,
|
431 |
+
outputs=temperature,
|
432 |
+
)
|
433 |
+
top_p = gr.Slider(
|
434 |
+
minimum=0.01,
|
435 |
+
maximum=0.99,
|
436 |
+
value=0.8,
|
437 |
+
step=0.01,
|
438 |
+
interactive=True,
|
439 |
+
visible=True,
|
440 |
+
label="Top P",
|
441 |
+
info="Higher values is equivalent to sampling more low-probability tokens.",
|
442 |
+
)
|
443 |
+
decoding_strategy.change(
|
444 |
+
fn=lambda selection: gr.Slider.update(visible=(selection in ["Top P Sampling"])),
|
445 |
+
inputs=decoding_strategy,
|
446 |
+
outputs=top_p,
|
447 |
+
)
|
448 |
with gr.Column():
|
449 |
chatbot.render()
|
450 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
451 |
def model_inference(
|
452 |
model_selector,
|
453 |
system_prompt,
|