mmarch/text-generation-webui
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Two new options: truncation length and ban eos token

Browse source
This commit is contained in:
oobabooga 2023-04-11 18:46:06 -03:00 • committed by GitHub
parent 749c08a4ff
commit cacbcda208
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
6 changed files with 62 additions and 48 deletions
Download patch file Download diff file Expand all files Collapse all files

13
server.py
View file

@ -263,7 +263,7 @@ def create_settings_menus(default_preset):
with gr.Box():
gr.Markdown('Contrastive search')
shared.gradio['penalty_alpha'] = gr.Slider(0, 5, value=generate_params['penalty_alpha'], label='penalty_alpha')
with gr.Box():
gr.Markdown('Beam search (uses a lot of VRAM)')
with gr.Row():
with gr.Column():
@ -272,10 +272,11 @@ def create_settings_menus(default_preset):
shared.gradio['length_penalty'] = gr.Slider(-5, 5, value=generate_params['length_penalty'], label='length_penalty')
shared.gradio['early_stopping'] = gr.Checkbox(value=generate_params['early_stopping'], label='early_stopping')
with gr.Row():
shared.gradio['add_bos_token'] = gr.Checkbox(value=shared.settings['add_bos_token'], label='Add the bos_token to the beginning of prompts', info='Disabling this can make the replies more creative.')
with gr.Row():
with gr.Group():
with gr.Row():
shared.gradio['add_bos_token'] = gr.Checkbox(value=shared.settings['add_bos_token'], label='Add the bos_token to the beginning of prompts', info='Disabling this can make the replies more creative.')
shared.gradio['ban_eos_token'] = gr.Checkbox(value=shared.settings['ban_eos_token'], label='Ban the eos token', info='This forces the model to never end the generation prematurely.')
shared.gradio['truncation_length'] = gr.Slider(value=shared.settings['truncation_length'], minimum=shared.settings['truncation_length_min'], maximum=shared.settings['truncation_length_max'], step=1, label='Truncate the prompt up to this length', info='The leftmost tokens are removed if the prompt exceeds this length. Most models require this to be at most 2048.')
shared.gradio['custom_stopping_strings'] = gr.Textbox(lines=1, value=shared.settings["custom_stopping_strings"] or None, label='Custom stopping strings', info='In addition to the defaults. Written between "" and separated by commas. For instance: "\\nYour Assistant:", "\\nThe assistant:"')
with gr.Accordion('Soft prompt', open=False):
@ -361,7 +362,7 @@ title = 'Text generation web UI'
def list_interface_input_elements(chat=False):
elements = ['max_new_tokens', 'seed', 'temperature', 'top_p', 'top_k', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'no_repeat_ngram_size', 'min_length', 'do_sample', 'penalty_alpha', 'num_beams', 'length_penalty', 'early_stopping', 'add_bos_token', 'custom_stopping_strings']
elements = ['max_new_tokens', 'seed', 'temperature', 'top_p', 'top_k', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'no_repeat_ngram_size', 'min_length', 'do_sample', 'penalty_alpha', 'num_beams', 'length_penalty', 'early_stopping', 'add_bos_token', 'ban_eos_token', 'truncation_length', 'custom_stopping_strings']
if chat:
elements += ['name1', 'name2', 'greeting', 'context', 'end_of_turn', 'chat_prompt_size', 'chat_generation_attempts', 'stop_at_newline', 'mode']
return elements