Organize the parameters tab (#5767)

docs/03 - Parameters Tab.md

@@ -61,10 +61,6 @@ For more information about the parameters, the [transformers documentation](http
 * **Seed**: Set the Pytorch seed to this number. Note that some loaders do not use Pytorch (notably llama.cpp), and others are not deterministic (ExLlamaV2). For these loaders, the seed has no effect.
 * **encoder_repetition_penalty**: Also known as the "Hallucinations filter". Used to penalize tokens that are *not* in the prior text. Higher value = more likely to stay in context, lower value = more likely to diverge.
 * **no_repeat_ngram_size**: If not set to 0, specifies the length of token sets that are completely blocked from repeating at all. Higher values = blocks larger phrases, lower values = blocks words or letters from repeating. Only 0 or high values are a good idea in most cases.
-* **min_length**: Minimum generation length in tokens. This is a built-in parameter in the transformers library that has never been very useful. Typically you want to check "Ban the eos_token" instead.
-* **num_beams**: Number of beams for beam search. 1 means no beam search.
-* **length_penalty**: Used by beam search only. `length_penalty > 0.0` promotes longer sequences, while `length_penalty < 0.0` encourages shorter sequences.
-* **early_stopping**: Used by beam search only. When checked, the generation stops as soon as there are "num_beams" complete candidates; otherwise, a heuristic is applied and the generation stops when is it very unlikely to find better candidates (I just copied this from the transformers documentation and have never gotten beam search to generate good results).
 
 To the right (or below if you are on mobile), the following parameters are present:
 

extensions/openai/typing.py

@@ -33,10 +33,6 @@ class GenerationOptions(BaseModel):
     seed: int = -1
     encoder_repetition_penalty: float = 1
     no_repeat_ngram_size: int = 0
-    min_length: int = 0
-    num_beams: int = 1
-    length_penalty: float = 1
-    early_stopping: bool = False
     truncation_length: int = 0
     max_tokens_second: int = 0
     prompt_lookup_num_tokens: int = 0

modules/loaders.py

@@ -185,13 +185,9 @@ def transformers_samplers():
         'repetition_penalty_range',
         'encoder_repetition_penalty',
         'no_repeat_ngram_size',
-        'min_length',
         'seed',
         'do_sample',
         'penalty_alpha',
-        'num_beams',
-        'length_penalty',
-        'early_stopping',
         'mirostat_mode',
         'mirostat_tau',
         'mirostat_eta',
@@ -262,7 +258,6 @@ loaders_samplers = {
         'repetition_penalty_range',
         'encoder_repetition_penalty',
         'no_repeat_ngram_size',
-        'min_length',
         'seed',
         'do_sample',
         'mirostat_mode',
@@ -321,7 +316,6 @@ loaders_samplers = {
         'repetition_penalty_range',
         'encoder_repetition_penalty',
         'no_repeat_ngram_size',
-        'min_length',
         'seed',
         'do_sample',
         'mirostat_mode',

modules/presets.py

@@ -40,10 +40,6 @@ def default_preset():
         'do_sample': True,
         'encoder_repetition_penalty': 1,
         'no_repeat_ngram_size': 0,
-        'min_length': 0,
-        'num_beams': 1,
-        'length_penalty': 1,
-        'early_stopping': False,
         'sampler_priority': 'temperature\ndynamic_temperature\nquadratic_sampling\ntop_k\ntop_p\ntypical_p\nepsilon_cutoff\neta_cutoff\ntfs\ntop_a\nmin_p\nmirostat'
     }
 
@@ -52,7 +48,7 @@ def presets_params():
     return [k for k in default_preset()]
 
 
-def load_preset(name):
+def load_preset(name, verbose=False):
     generate_params = default_preset()
     if name not in ['None', None, '']:
         path = Path(f'presets/{name}.yaml')
@@ -65,6 +61,10 @@ def load_preset(name):
         else:
             logger.error(f"The preset \"{name}\" does not exist under \"{path}\". Using the default parameters.")
 
+    if verbose:
+        logger.info(f"\"{name}\" preset:")
+        pprint.PrettyPrinter(indent=4, width=1, sort_dicts=False).pprint(remove_defaults(generate_params))
+
     return generate_params
 
 
@@ -74,7 +74,7 @@ def load_preset_memoized(name):
 
 
 def load_preset_for_ui(name, state):
-    generate_params = load_preset(name)
+    generate_params = load_preset(name, verbose=True)
     state.update(generate_params)
     return state, *[generate_params[k] for k in presets_params()]
 

modules/text_generation.py

@@ -279,7 +279,7 @@ def get_reply_from_output_ids(output_ids, state=None, starting_from=0):
 
 def generate_reply_HF(question, original_question, seed, state, stopping_strings=None, is_chat=False):
     generate_params = {}
-    for k in ['max_new_tokens', 'temperature', 'temperature_last', 'dynamic_temperature', 'dynatemp_low', 'dynatemp_high', 'dynatemp_exponent', 'smoothing_factor', 'smoothing_curve', 'top_p', 'min_p', 'top_k', 'repetition_penalty', 'presence_penalty', 'frequency_penalty', 'repetition_penalty_range', 'typical_p', 'tfs', 'top_a', 'guidance_scale', 'penalty_alpha', 'mirostat_mode', 'mirostat_tau', 'mirostat_eta', 'do_sample', 'encoder_repetition_penalty', 'no_repeat_ngram_size', 'min_length', 'num_beams', 'length_penalty', 'early_stopping']:
+    for k in ['max_new_tokens', 'temperature', 'temperature_last', 'dynamic_temperature', 'dynatemp_low', 'dynatemp_high', 'dynatemp_exponent', 'smoothing_factor', 'smoothing_curve', 'top_p', 'min_p', 'top_k', 'repetition_penalty', 'presence_penalty', 'frequency_penalty', 'repetition_penalty_range', 'typical_p', 'tfs', 'top_a', 'guidance_scale', 'penalty_alpha', 'mirostat_mode', 'mirostat_tau', 'mirostat_eta', 'do_sample', 'encoder_repetition_penalty', 'no_repeat_ngram_size']:
         if k in state:
             generate_params[k] = state[k]
 

modules/ui.py

@@ -139,12 +139,8 @@ def list_interface_input_elements():
         'repetition_penalty_range',
         'encoder_repetition_penalty',
         'no_repeat_ngram_size',
-        'min_length',
         'do_sample',
         'penalty_alpha',
-        'num_beams',
-        'length_penalty',
-        'early_stopping',
         'mirostat_mode',
         'mirostat_tau',
         'mirostat_eta',

modules/ui_parameters.py

@@ -30,64 +30,35 @@ def create_ui(default_preset):
                             shared.gradio['max_new_tokens'] = gr.Slider(minimum=shared.settings['max_new_tokens_min'], maximum=shared.settings['max_new_tokens_max'], step=1, label='max_new_tokens', value=shared.settings['max_new_tokens'])
                             shared.gradio['temperature'] = gr.Slider(0.01, 5, value=generate_params['temperature'], step=0.01, label='temperature')
                             shared.gradio['top_p'] = gr.Slider(0.0, 1.0, value=generate_params['top_p'], step=0.01, label='top_p')
-                            shared.gradio['min_p'] = gr.Slider(0.0, 1.0, value=generate_params['min_p'], step=0.01, label='min_p')
                             shared.gradio['top_k'] = gr.Slider(0, 200, value=generate_params['top_k'], step=1, label='top_k')
+                            shared.gradio['typical_p'] = gr.Slider(0.0, 1.0, value=generate_params['typical_p'], step=0.01, label='typical_p')
                             shared.gradio['repetition_penalty'] = gr.Slider(1.0, 1.5, value=generate_params['repetition_penalty'], step=0.01, label='repetition_penalty')
                             shared.gradio['presence_penalty'] = gr.Slider(0, 2, value=generate_params['presence_penalty'], step=0.05, label='presence_penalty')
                             shared.gradio['frequency_penalty'] = gr.Slider(0, 2, value=generate_params['frequency_penalty'], step=0.05, label='frequency_penalty')
                             shared.gradio['repetition_penalty_range'] = gr.Slider(0, 4096, step=64, value=generate_params['repetition_penalty_range'], label='repetition_penalty_range')
-                            shared.gradio['typical_p'] = gr.Slider(0.0, 1.0, value=generate_params['typical_p'], step=0.01, label='typical_p')
-                            shared.gradio['tfs'] = gr.Slider(0.0, 1.0, value=generate_params['tfs'], step=0.01, label='tfs')
-                            shared.gradio['top_a'] = gr.Slider(0.0, 1.0, value=generate_params['top_a'], step=0.01, label='top_a')
-                            shared.gradio['epsilon_cutoff'] = gr.Slider(0, 9, value=generate_params['epsilon_cutoff'], step=0.01, label='epsilon_cutoff')
-                            shared.gradio['eta_cutoff'] = gr.Slider(0, 20, value=generate_params['eta_cutoff'], step=0.01, label='eta_cutoff')
-
-                        with gr.Column():
-                            shared.gradio['guidance_scale'] = gr.Slider(-0.5, 2.5, step=0.05, value=generate_params['guidance_scale'], label='guidance_scale', info='For CFG. 1.5 is a good value.')
-                            shared.gradio['negative_prompt'] = gr.Textbox(value=shared.settings['negative_prompt'], label='Negative prompt', lines=3, elem_classes=['add_scrollbar'])
-                            shared.gradio['penalty_alpha'] = gr.Slider(0, 5, value=generate_params['penalty_alpha'], label='penalty_alpha', info='For Contrastive Search. do_sample must be unchecked.')
-                            shared.gradio['mirostat_mode'] = gr.Slider(0, 2, step=1, value=generate_params['mirostat_mode'], label='mirostat_mode', info='mode=1 is for llama.cpp only.')
-                            shared.gradio['mirostat_tau'] = gr.Slider(0, 10, step=0.01, value=generate_params['mirostat_tau'], label='mirostat_tau')
-                            shared.gradio['mirostat_eta'] = gr.Slider(0, 1, step=0.01, value=generate_params['mirostat_eta'], label='mirostat_eta')
-                            shared.gradio['smoothing_factor'] = gr.Slider(0.0, 10.0, value=generate_params['smoothing_factor'], step=0.01, label='smoothing_factor', info='Activates Quadratic Sampling.')
-                            shared.gradio['smoothing_curve'] = gr.Slider(1.0, 10.0, value=generate_params['smoothing_curve'], step=0.01, label='smoothing_curve', info='Adjusts the dropoff curve of Quadratic Sampling.')
-                            shared.gradio['dynamic_temperature'] = gr.Checkbox(value=generate_params['dynamic_temperature'], label='dynamic_temperature')
-                            shared.gradio['dynatemp_low'] = gr.Slider(0.01, 5, value=generate_params['dynatemp_low'], step=0.01, label='dynatemp_low', visible=generate_params['dynamic_temperature'])
-                            shared.gradio['dynatemp_high'] = gr.Slider(0.01, 5, value=generate_params['dynatemp_high'], step=0.01, label='dynatemp_high', visible=generate_params['dynamic_temperature'])
-                            shared.gradio['dynatemp_exponent'] = gr.Slider(0.01, 5, value=generate_params['dynatemp_exponent'], step=0.01, label='dynatemp_exponent', visible=generate_params['dynamic_temperature'])
-                            shared.gradio['temperature_last'] = gr.Checkbox(value=generate_params['temperature_last'], label='temperature_last', info='Moves temperature/dynamic temperature/quadratic sampling to the end of the sampler stack, ignoring their positions in "Sampler priority".')
                             shared.gradio['do_sample'] = gr.Checkbox(value=generate_params['do_sample'], label='do_sample')
-                            shared.gradio['seed'] = gr.Number(value=shared.settings['seed'], label='Seed (-1 for random)')
-                            with gr.Accordion('Other parameters', open=False):
-                                shared.gradio['encoder_repetition_penalty'] = gr.Slider(0.8, 1.5, value=generate_params['encoder_repetition_penalty'], step=0.01, label='encoder_repetition_penalty')
-                                shared.gradio['no_repeat_ngram_size'] = gr.Slider(0, 20, step=1, value=generate_params['no_repeat_ngram_size'], label='no_repeat_ngram_size')
-                                shared.gradio['min_length'] = gr.Slider(0, 2000, step=1, value=generate_params['min_length'], label='min_length')
-                                shared.gradio['num_beams'] = gr.Slider(1, 20, step=1, value=generate_params['num_beams'], label='num_beams', info='For Beam Search, along with length_penalty and early_stopping.')
-                                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')
-
                             gr.Markdown("[Learn more](https://github.com/oobabooga/text-generation-webui/wiki/03-%E2%80%90-Parameters-Tab)")
 
                         with gr.Column():
-                    with gr.Row():
+                            with gr.Group():
-                        with gr.Column():
-                            shared.gradio['truncation_length'] = gr.Slider(value=get_truncation_length(), minimum=shared.settings['truncation_length_min'], maximum=shared.settings['truncation_length_max'], step=256, 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['max_tokens_second'] = gr.Slider(value=shared.settings['max_tokens_second'], minimum=0, maximum=20, step=1, label='Maximum tokens/second', info='To make text readable in real time.')
-                            shared.gradio['prompt_lookup_num_tokens'] = gr.Slider(value=shared.settings['prompt_lookup_num_tokens'], minimum=0, maximum=10, step=1, label='prompt_lookup_num_tokens', info='Activates Prompt Lookup Decoding.')
-
-                            shared.gradio['custom_stopping_strings'] = gr.Textbox(lines=2, value=shared.settings["custom_stopping_strings"] or None, label='Custom stopping strings', info='In addition to the defaults. Written between "" and separated by commas.', placeholder='"\\n", "\\nYou:"')
-                            shared.gradio['custom_token_bans'] = gr.Textbox(value=shared.settings['custom_token_bans'] or None, label='Custom token bans', info='Specific token IDs to ban from generating, comma-separated. The IDs can be found in the Default or Notebook tab.')
-
-                        with gr.Column():
                                 shared.gradio['auto_max_new_tokens'] = gr.Checkbox(value=shared.settings['auto_max_new_tokens'], label='auto_max_new_tokens', info='Expand max_new_tokens to the available context length.')
                                 shared.gradio['ban_eos_token'] = gr.Checkbox(value=shared.settings['ban_eos_token'], label='Ban the eos_token', info='Forces the model to never end the generation prematurely.')
                                 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['skip_special_tokens'] = gr.Checkbox(value=shared.settings['skip_special_tokens'], label='Skip special tokens', info='Some specific models need this unset.')
+                                shared.gradio['custom_stopping_strings'] = gr.Textbox(lines=2, value=shared.settings["custom_stopping_strings"] or None, label='Custom stopping strings', info='In addition to the defaults. Written between "" and separated by commas.', placeholder='"\\n", "\\nYou:"')
-                            shared.gradio['stream'] = gr.Checkbox(value=shared.settings['stream'], label='Activate text streaming')
+                                shared.gradio['custom_token_bans'] = gr.Textbox(value=shared.settings['custom_token_bans'] or None, label='Token bans', info='Specific token IDs to ban from generating, comma-separated. The IDs can be found in the Default or Notebook tab.')
 
-                            with gr.Blocks():
+                            shared.gradio['penalty_alpha'] = gr.Slider(0, 5, value=generate_params['penalty_alpha'], label='penalty_alpha', info='For Contrastive Search. do_sample must be unchecked.')
-                                shared.gradio['sampler_priority'] = gr.Textbox(value=generate_params['sampler_priority'], lines=12, label='Sampler priority', info='Parameter names separated by new lines or commas.')
+                            shared.gradio['guidance_scale'] = gr.Slider(-0.5, 2.5, step=0.05, value=generate_params['guidance_scale'], label='guidance_scale', info='For CFG. 1.5 is a good value.')
+                            shared.gradio['negative_prompt'] = gr.Textbox(value=shared.settings['negative_prompt'], label='Negative prompt', lines=3, elem_classes=['add_scrollbar'])
+                            shared.gradio['mirostat_mode'] = gr.Slider(0, 2, step=1, value=generate_params['mirostat_mode'], label='mirostat_mode', info='mode=1 is for llama.cpp only.')
+                            shared.gradio['mirostat_tau'] = gr.Slider(0, 10, step=0.01, value=generate_params['mirostat_tau'], label='mirostat_tau')
+                            shared.gradio['mirostat_eta'] = gr.Slider(0, 1, step=0.01, value=generate_params['mirostat_eta'], label='mirostat_eta')
+                            shared.gradio['epsilon_cutoff'] = gr.Slider(0, 9, value=generate_params['epsilon_cutoff'], step=0.01, label='epsilon_cutoff')
+                            shared.gradio['eta_cutoff'] = gr.Slider(0, 20, value=generate_params['eta_cutoff'], step=0.01, label='eta_cutoff')
+                            shared.gradio['encoder_repetition_penalty'] = gr.Slider(0.8, 1.5, value=generate_params['encoder_repetition_penalty'], step=0.01, label='encoder_repetition_penalty')
+                            shared.gradio['no_repeat_ngram_size'] = gr.Slider(0, 20, step=1, value=generate_params['no_repeat_ngram_size'], label='no_repeat_ngram_size')
 
+                with gr.Column():
                     with gr.Row() as shared.gradio['grammar_file_row']:
                         shared.gradio['grammar_file'] = gr.Dropdown(value='None', choices=utils.get_available_grammars(), label='Load grammar from file (.gbnf)', elem_classes='slim-dropdown')
                         ui.create_refresh_button(shared.gradio['grammar_file'], lambda: None, lambda: {'choices': utils.get_available_grammars()}, 'refresh-button', interactive=not mu)
@@ -96,6 +67,28 @@ def create_ui(default_preset):
 
                     shared.gradio['grammar_string'] = gr.Textbox(value='', label='Grammar', lines=16, elem_classes=['add_scrollbar', 'monospace'])
 
+                    with gr.Row():
+                        with gr.Column():
+                            shared.gradio['min_p'] = gr.Slider(0.0, 1.0, value=generate_params['min_p'], step=0.01, label='min_p')
+                            shared.gradio['tfs'] = gr.Slider(0.0, 1.0, value=generate_params['tfs'], step=0.01, label='tfs')
+                            shared.gradio['top_a'] = gr.Slider(0.0, 1.0, value=generate_params['top_a'], step=0.01, label='top_a')
+                            shared.gradio['smoothing_factor'] = gr.Slider(0.0, 10.0, value=generate_params['smoothing_factor'], step=0.01, label='smoothing_factor', info='Activates Quadratic Sampling.')
+                            shared.gradio['smoothing_curve'] = gr.Slider(1.0, 10.0, value=generate_params['smoothing_curve'], step=0.01, label='smoothing_curve', info='Adjusts the dropoff curve of Quadratic Sampling.')
+                            shared.gradio['dynamic_temperature'] = gr.Checkbox(value=generate_params['dynamic_temperature'], label='dynamic_temperature')
+                            shared.gradio['dynatemp_low'] = gr.Slider(0.01, 5, value=generate_params['dynatemp_low'], step=0.01, label='dynatemp_low', visible=generate_params['dynamic_temperature'])
+                            shared.gradio['dynatemp_high'] = gr.Slider(0.01, 5, value=generate_params['dynatemp_high'], step=0.01, label='dynatemp_high', visible=generate_params['dynamic_temperature'])
+                            shared.gradio['dynatemp_exponent'] = gr.Slider(0.01, 5, value=generate_params['dynatemp_exponent'], step=0.01, label='dynatemp_exponent', visible=generate_params['dynamic_temperature'])
+                            shared.gradio['temperature_last'] = gr.Checkbox(value=generate_params['temperature_last'], label='temperature_last', info='Moves temperature/dynamic temperature/quadratic sampling to the end of the sampler stack, ignoring their positions in "Sampler priority".')
+                            shared.gradio['sampler_priority'] = gr.Textbox(value=generate_params['sampler_priority'], lines=12, label='Sampler priority', info='Parameter names separated by new lines or commas.')
+
+                        with gr.Column():
+                            shared.gradio['truncation_length'] = gr.Slider(value=get_truncation_length(), minimum=shared.settings['truncation_length_min'], maximum=shared.settings['truncation_length_max'], step=256, 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['prompt_lookup_num_tokens'] = gr.Slider(value=shared.settings['prompt_lookup_num_tokens'], minimum=0, maximum=10, step=1, label='prompt_lookup_num_tokens', info='Activates Prompt Lookup Decoding.')
+                            shared.gradio['max_tokens_second'] = gr.Slider(value=shared.settings['max_tokens_second'], minimum=0, maximum=20, step=1, label='Maximum tokens/second', info='To make text readable in real time.')
+                            shared.gradio['seed'] = gr.Number(value=shared.settings['seed'], label='Seed (-1 for random)')
+                            shared.gradio['skip_special_tokens'] = gr.Checkbox(value=shared.settings['skip_special_tokens'], label='Skip special tokens', info='Some specific models need this unset.')
+                            shared.gradio['stream'] = gr.Checkbox(value=shared.settings['stream'], label='Activate text streaming')
+
         ui_chat.create_chat_settings_ui()