Fix silero_tts

Useful for starting a new reply.

README.md

@@ -13,7 +13,7 @@ Its goal is to become the [AUTOMATIC1111/stable-diffusion-webui](https://github.
 * Dropdown menu for switching between models
 * Notebook mode that resembles OpenAI's playground
 * Chat mode for conversation and role playing
-* Instruct mode compatible with Alpaca and Open Assistant formats **\*NEW!\***
+* Instruct mode compatible with Alpaca, Vicuna, and Open Assistant formats **\*NEW!\***
 * Nice HTML output for GPT-4chan
 * Markdown output for [GALACTICA](https://github.com/paperswithcode/galai), including LaTeX rendering
 * [Custom chat characters](https://github.com/oobabooga/text-generation-webui/wiki/Custom-chat-characters)
@@ -289,7 +289,9 @@ Check the [wiki](https://github.com/oobabooga/text-generation-webui/wiki/System-
 
 ## Contributing
 
-Pull requests, suggestions, and issue reports are welcome.
+Pull requests, suggestions, and issue reports are welcome. 
+
+You are also welcome to review open pull requests.
 
 Before reporting a bug, make sure that you have:
 

api-example-stream.py

@@ -12,6 +12,11 @@ import string
 
 import websockets
 
+# Note, Gradio may pick a different fn value as the definition of the Gradio app changes.
+# You can always launch the web UI and inspect the websocket stream using your browser's dev tools
+# to determine what value Gradio expects here.
+GRADIO_FN = 29
+
 
 def random_hash():
     letters = string.ascii_lowercase + string.digits
@@ -36,6 +41,10 @@ async def run(context):
         'length_penalty': 1,
         'early_stopping': False,
         'seed': -1,
+        'add_bos_token': True,
+        'truncation_length': 2048,
+        'custom_stopping_strings': [],
+        'ban_eos_token': False
     }
     payload = json.dumps([context, params])
     session = random_hash()
@@ -47,14 +56,14 @@ async def run(context):
                 case "send_hash":
                     await websocket.send(json.dumps({
                         "session_hash": session,
-                        "fn_index": 12
+                        "fn_index": GRADIO_FN
                     }))
                 case "estimation":
                     pass
                 case "send_data":
                     await websocket.send(json.dumps({
                         "session_hash": session,
-                        "fn_index": 12,
+                        "fn_index": GRADIO_FN,
                         "data": [
                             payload
                         ]

api-example.py

@@ -35,6 +35,10 @@ params = {
     'length_penalty': 1,
     'early_stopping': False,
     'seed': -1,
+    'add_bos_token': True,
+    'custom_stopping_strings': [],
+    'truncation_length': 2048,
+    'ban_eos_token': False,
 }
 
 # Input prompt

css/html_cai_style.css

@@ -7,11 +7,13 @@
     padding-right: 20px;
     display: flex;
     flex-direction: column-reverse;
+    word-break: break-word;
+    overflow-wrap: anywhere;
 }
 
 .message {
     display: grid;
-    grid-template-columns: 60px 1fr;
+    grid-template-columns: 60px minmax(0, 1fr);
     padding-bottom: 25px;
     font-size: 15px;
     font-family: Helvetica, Arial, sans-serif;
@@ -73,6 +75,13 @@
     display: inline !important;
 }
 
+.message-body code {
+    overflow-x: auto;
+}
+.message-body :not(pre) > code {
+    white-space: normal !important;
+}
+
 .dark .message-body p em {
     color: rgb(138, 138, 138) !important;
 }

css/html_instruct_style.css

@@ -7,6 +7,8 @@
     padding-right: 20px;
     display: flex;
     flex-direction: column-reverse;
+    word-break: break-word;
+    overflow-wrap: anywhere;
 }
 
 .message {
@@ -37,6 +39,13 @@
     display: inline !important;
 }
 
+.message-body code {
+    overflow-x: auto;
+}
+.message-body :not(pre) > code {
+    white-space: normal !important;
+}
+
 .dark .message-body p em {
     color: rgb(138, 138, 138) !important;
 }

extensions/api/script.py

@@ -58,12 +58,14 @@ class Handler(BaseHTTPRequestHandler):
                 'early_stopping': bool(body.get('early_stopping', False)),
                 'seed': int(body.get('seed', -1)),
                 'add_bos_token': int(body.get('add_bos_token', True)),
+                'custom_stopping_strings': body.get('custom_stopping_strings', []),
+                'truncation_length': int(body.get('truncation_length', 2048)),
+                'ban_eos_token': bool(body.get('ban_eos_token', False)),
             }
 
             generator = generate_reply(
                 prompt,
                 generate_params,
-                stopping_strings=body.get('stopping_strings', []),
             )
 
             answer = ''
@@ -79,6 +81,19 @@ class Handler(BaseHTTPRequestHandler):
                 }]
             })
             self.wfile.write(response.encode('utf-8'))
+        elif self.path == '/api/v1/token-count':
+            # Not compatible with KoboldAI api
+            self.send_response(200)
+            self.send_header('Content-Type', 'application/json')
+            self.end_headers()
+
+            tokens = encode(body['prompt'])[0]
+            response = json.dumps({
+                'results': [{
+                    'tokens': len(tokens)
+                }]
+            })
+            self.wfile.write(response.encode('utf-8'))
         else:
             self.send_error(404)
 

extensions/silero_tts/script.py

@@ -165,13 +165,13 @@ def ui():
     convert_arr = [convert_confirm, convert, convert_cancel]
     convert.click(lambda: [gr.update(visible=True), gr.update(visible=False), gr.update(visible=True)], None, convert_arr)
     convert_confirm.click(lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, convert_arr)
-    convert_confirm.click(remove_tts_from_history, [shared.gradio[k] for k in ['name1', 'name2', 'Chat mode']], shared.gradio['display'])
+    convert_confirm.click(remove_tts_from_history, [shared.gradio[k] for k in ['name1', 'name2', 'mode']], shared.gradio['display'])
     convert_confirm.click(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
     convert_cancel.click(lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, convert_arr)
 
     # Toggle message text in history
     show_text.change(lambda x: params.update({"show_text": x}), show_text, None)
-    show_text.change(toggle_text_in_history, [shared.gradio[k] for k in ['name1', 'name2', 'Chat mode']], shared.gradio['display'])
+    show_text.change(toggle_text_in_history, [shared.gradio[k] for k in ['name1', 'name2', 'mode']], shared.gradio['display'])
     show_text.change(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
 
     # Event functions to update the parameters in the backend

modules/chat.py

@@ -18,35 +18,35 @@ from modules.text_generation import (encode, generate_reply,
                                      get_max_prompt_length)
 
 
-def generate_chat_prompt(user_input, max_new_tokens, name1, name2, context, chat_prompt_size, **kwargs):
+def generate_chat_prompt(user_input, state, **kwargs):
-    is_instruct = kwargs['is_instruct'] if 'is_instruct' in kwargs else False
-    end_of_turn = kwargs['end_of_turn'] if 'end_of_turn' in kwargs else ''
     impersonate = kwargs['impersonate'] if 'impersonate' in kwargs else False
     _continue = kwargs['_continue'] if '_continue' in kwargs else False
     also_return_rows = kwargs['also_return_rows'] if 'also_return_rows' in kwargs else False
-    rows = [f"{context.strip()}\n"]
+    is_instruct = state['mode'] == 'instruct'
+    rows = [f"{state['context'].strip()}\n"]
 
     # Finding the maximum prompt size
+    chat_prompt_size = state['chat_prompt_size']
     if shared.soft_prompt:
         chat_prompt_size -= shared.soft_prompt_tensor.shape[1]
-    max_length = min(get_max_prompt_length(max_new_tokens), chat_prompt_size)
+    max_length = min(get_max_prompt_length(state), chat_prompt_size)
 
     if is_instruct:
-        prefix1 = f"{name1}\n"
+        prefix1 = f"{state['name1']}\n"
-        prefix2 = f"{name2}\n"
+        prefix2 = f"{state['name2']}\n"
     else:
-        prefix1 = f"{name1}: "
+        prefix1 = f"{state['name1']}: "
-        prefix2 = f"{name2}: "
+        prefix2 = f"{state['name2']}: "
 
     i = len(shared.history['internal']) - 1
-    while i >= 0 and len(encode(''.join(rows), max_new_tokens)[0]) < max_length:
+    while i >= 0 and len(encode(''.join(rows))[0]) < max_length:
         if _continue and i == len(shared.history['internal']) - 1:
             rows.insert(1, f"{prefix2}{shared.history['internal'][i][1]}")
         else:
-            rows.insert(1, f"{prefix2}{shared.history['internal'][i][1].strip()}{end_of_turn}\n")
+            rows.insert(1, f"{prefix2}{shared.history['internal'][i][1].strip()}{state['end_of_turn']}\n")
         string = shared.history['internal'][i][0]
         if string not in ['', '<|BEGIN-VISIBLE-CHAT|>']:
-            rows.insert(1, f"{prefix1}{string.strip()}{end_of_turn}\n")
+            rows.insert(1, f"{prefix1}{string.strip()}{state['end_of_turn']}\n")
         i -= 1
 
     if impersonate:
@@ -58,13 +58,13 @@ def generate_chat_prompt(user_input, max_new_tokens, name1, name2, context, chat
         # Adding the user message
         user_input = fix_newlines(user_input)
         if len(user_input) > 0:
-            rows.append(f"{prefix1}{user_input}{end_of_turn}\n")
+            rows.append(f"{prefix1}{user_input}{state['end_of_turn']}\n")
 
         # Adding the Character prefix
         rows.append(apply_extensions(f"{prefix2.strip() if not is_instruct else prefix2}", "bot_prefix"))
         limit = 3
 
-    while len(rows) > limit and len(encode(''.join(rows), max_new_tokens)[0]) >= max_length:
+    while len(rows) > limit and len(encode(''.join(rows))[0]) >= max_length:
         rows.pop(1)
     prompt = ''.join(rows)
 
@@ -74,8 +74,18 @@ def generate_chat_prompt(user_input, max_new_tokens, name1, name2, context, chat
         return prompt
 
 
+def get_stopping_strings(state):
+    if state['mode'] == 'instruct':
+        stopping_strings = [f"\n{state['name1']}", f"\n{state['name2']}"]
+    else:
+        stopping_strings = [f"\n{state['name1']}:", f"\n{state['name2']}:"]
+    stopping_strings += state['custom_stopping_strings']
+    return stopping_strings
+
+
 def extract_message_from_reply(reply, state):
     next_character_found = False
+    stopping_strings = get_stopping_strings(state)
 
     if state['stop_at_newline']:
         lines = reply.split('\n')
@@ -83,7 +93,7 @@ def extract_message_from_reply(reply, state):
         if len(lines) > 1:
             next_character_found = True
     else:
-        for string in [f"\n{state['name1']}:", f"\n{state['name2']}:"]:
+        for string in stopping_strings:
             idx = reply.find(string)
             if idx != -1:
                 reply = reply[:idx]
@@ -92,7 +102,7 @@ def extract_message_from_reply(reply, state):
         # If something like "\nYo" is generated just before "\nYou:"
         # is completed, trim it
         if not next_character_found:
-            for string in [f"\n{state['name1']}:", f"\n{state['name2']}:"]:
+            for string in stopping_strings:
                 for j in range(len(string) - 1, 0, -1):
                     if reply[-j:] == string[:j]:
                         reply = reply[:-j]
@@ -106,10 +116,6 @@ def extract_message_from_reply(reply, state):
 
 
 def chatbot_wrapper(text, state, regenerate=False, _continue=False):
-    if state['mode'] == 'instruct':
-        stopping_strings = [f"\n{state['name1']}", f"\n{state['name2']}"]
-    else:
-        stopping_strings = [f"\n{state['name1']}:", f"\n{state['name2']}:"]
 
     # Defining some variables
     cumulative_reply = ''
@@ -117,6 +123,7 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
     just_started = True
     visible_text = custom_generate_chat_prompt = None
     eos_token = '\n' if state['stop_at_newline'] else None
+    stopping_strings = get_stopping_strings(state)
 
     # Check if any extension wants to hijack this function call
     for extension, _ in extensions_module.iterator():
@@ -132,15 +139,11 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
         text = apply_extensions(text, "input")
 
     # Generating the prompt
-    kwargs = {
+    kwargs = {'_continue': _continue}
-        'end_of_turn': state['end_of_turn'],
-        'is_instruct': state['mode'] == 'instruct',
-        '_continue': _continue
-    }
     if custom_generate_chat_prompt is None:
-        prompt = generate_chat_prompt(text, state['max_new_tokens'], state['name1'], state['name2'], state['context'], state['chat_prompt_size'], **kwargs)
+        prompt = generate_chat_prompt(text, state, **kwargs)
     else:
-        prompt = custom_generate_chat_prompt(text, state['max_new_tokens'], state['name1'], state['name2'], state['context'], state['chat_prompt_size'], **kwargs)
+        prompt = custom_generate_chat_prompt(text, state, **kwargs)
 
     # Yield *Is typing...*
     if not any((regenerate, _continue)):
@@ -168,7 +171,7 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
                     shared.history['visible'].append(['', ''])
 
             if _continue:
-                sep = list(map(lambda x: ' ' if x[-1] != ' ' else '', last_reply))
+                sep = list(map(lambda x: ' ' if len(x) > 0 and x[-1] != ' ' else '', last_reply))
                 shared.history['internal'][-1] = [text, f'{last_reply[0]}{sep[0]}{reply}']
                 shared.history['visible'][-1] = [visible_text, f'{last_reply[1]}{sep[1]}{visible_reply}']
             else:
@@ -186,15 +189,12 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
 
 
 def impersonate_wrapper(text, state):
-    if state['mode'] == 'instruct':
-        stopping_strings = [f"\n{state['name1']}", f"\n{state['name2']}"]
-    else:
-        stopping_strings = [f"\n{state['name1']}:", f"\n{state['name2']}:"]
 
     # Defining some variables
     cumulative_reply = ''
     eos_token = '\n' if state['stop_at_newline'] else None
-    prompt = generate_chat_prompt(text, state['max_new_tokens'], state['name1'], state['name2'], state['context'], state['chat_prompt_size'], end_of_turn=state['end_of_turn'], impersonate=True)
+    prompt = generate_chat_prompt(text, state, impersonate=True)
+    stopping_strings = get_stopping_strings(state)
 
     # Yield *Is typing...*
     yield shared.processing_message
@@ -267,6 +267,21 @@ def replace_last_reply(text, name1, name2, mode):
     return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
 
 
+def send_dummy_message(text, name1, name2, mode):
+    shared.history['visible'].append([text, ''])
+    shared.history['internal'].append([apply_extensions(text, "input"), ''])
+    return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
+
+
+def send_dummy_reply(text, name1, name2, mode):
+    if len(shared.history['visible']) > 0 and not shared.history['visible'][-1][1] == '':
+        shared.history['visible'].append(['', ''])
+        shared.history['internal'].append(['', ''])
+    shared.history['visible'][-1][1] = text
+    shared.history['internal'][-1][1] = apply_extensions(text, "input")
+    return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
+
+
 def clear_html():
     return chat_html_wrapper([], "", "")
 
@@ -498,4 +513,4 @@ def upload_your_profile_picture(img, name1, name2, mode):
         img.save(Path('cache/pfp_me.png'))
         print('Profile picture saved to "cache/pfp_me.png"')
 
-    return chat_html_wrapper(shared.history['visible'], name1, name2, mode, reset_cache=True)
+    return chat_html_wrapper(shared.history['visible'], name1, name2, mode, reset_cache=True)

modules/models.py

@@ -189,7 +189,6 @@ def load_model(model_name):
             pass
     else:
         tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/"))
-    tokenizer.truncation_side = 'left'
 
     print(f"Loaded the model in {(time.time()-t0):.2f} seconds.")
     return model, tokenizer

modules/shared.py

@@ -34,8 +34,13 @@ settings = {
     'context': 'This is a conversation with your Assistant. The Assistant is very helpful and is eager to chat with you and answer your questions.',
     'greeting': 'Hello there!',
     'end_of_turn': '',
+    'custom_stopping_strings': '',
     'stop_at_newline': False,
     'add_bos_token': True,
+    'ban_eos_token': False,
+    'truncation_length': 2048,
+    'truncation_length_min': 0,
+    'truncation_length_max': 4096,
     'chat_prompt_size': 2048,
     'chat_prompt_size_min': 0,
     'chat_prompt_size_max': 2048,

modules/text_generation.py

@@ -15,20 +15,20 @@ from modules.html_generator import generate_4chan_html, generate_basic_html
 from modules.models import clear_torch_cache, local_rank
 
 
-def get_max_prompt_length(tokens):
+def get_max_prompt_length(state):
-    max_length = 2048 - tokens
+    max_length = state['truncation_length'] - state['max_new_tokens']
     if shared.soft_prompt:
         max_length -= shared.soft_prompt_tensor.shape[1]
     return max_length
 
 
-def encode(prompt, tokens_to_generate=0, add_special_tokens=True, add_bos_token=True):
+def encode(prompt, add_special_tokens=True, add_bos_token=True, truncation_length=None):
     if any((shared.is_RWKV, shared.is_llamacpp)):
         input_ids = shared.tokenizer.encode(str(prompt))
         input_ids = np.array(input_ids).reshape(1, len(input_ids))
         return input_ids
     else:
-        input_ids = shared.tokenizer.encode(str(prompt), return_tensors='pt', truncation=True, max_length=get_max_prompt_length(tokens_to_generate), add_special_tokens=add_special_tokens)
+        input_ids = shared.tokenizer.encode(str(prompt), return_tensors='pt', add_special_tokens=add_special_tokens)
 
         # This is a hack for making replies more creative.
         if not add_bos_token and input_ids[0][0] == shared.tokenizer.bos_token_id:
@@ -39,17 +39,21 @@ def encode(prompt, tokens_to_generate=0, add_special_tokens=True, add_bos_token=
         if type(shared.tokenizer) is transformers.LlamaTokenizer and input_ids[0][0] == 29871:
             input_ids = input_ids[:, 1:]
 
-        if shared.args.cpu:
+    # Handling truncation
-            return input_ids
+    if truncation_length is not None:
-        elif shared.args.flexgen:
+        input_ids = input_ids[:, -truncation_length:]
-            return input_ids.numpy()
+
-        elif shared.args.deepspeed:
+    if any((shared.is_RWKV, shared.is_llamacpp, shared.args.cpu)):
-            return input_ids.to(device=local_rank)
+        return input_ids
-        elif torch.has_mps:
+    elif shared.args.flexgen:
-            device = torch.device('mps')
+        return input_ids.numpy()
-            return input_ids.to(device)
+    elif shared.args.deepspeed:
-        else:
+        return input_ids.to(device=local_rank)
-            return input_ids.cuda()
+    elif torch.has_mps:
+        device = torch.device('mps')
+        return input_ids.to(device)
+    else:
+        return input_ids.cuda()
 
 
 def decode(output_ids):
@@ -129,12 +133,14 @@ def generate_reply(question, state, eos_token=None, stopping_strings=[]):
     original_question = question
     if not shared.is_chat():
         question = apply_extensions(question, 'input')
-    if shared.args.verbose:
-        print(f'\n\n{question}\n--------------------\n')
 
     # These models are not part of Hugging Face, so we handle them
     # separately and terminate the function call earlier
     if any((shared.is_RWKV, shared.is_llamacpp)):
+
+        if shared.args.verbose:
+            print(f'\n\n{question}\n--------------------\n')
+
         for k in ['temperature', 'top_p', 'top_k', 'repetition_penalty']:
             generate_params[k] = state[k]
         generate_params['token_count'] = state['max_new_tokens']
@@ -166,24 +172,33 @@ def generate_reply(question, state, eos_token=None, stopping_strings=[]):
             print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens}, seed {seed})')
             return
 
-    input_ids = encode(question, state['max_new_tokens'], add_bos_token=state['add_bos_token'])
+    input_ids = encode(question, add_bos_token=state['add_bos_token'], truncation_length=get_max_prompt_length(state))
     original_input_ids = input_ids
     output = input_ids[0]
 
+    if shared.args.verbose:
+        print(f'\n\n{decode(input_ids[0])}\n--------------------\n')
+
     cuda = not any((shared.args.cpu, shared.args.deepspeed, shared.args.flexgen))
     eos_token_ids = [shared.tokenizer.eos_token_id] if shared.tokenizer.eos_token_id is not None else []
     if eos_token is not None:
         eos_token_ids.append(int(encode(eos_token)[0][-1]))
+
+    # Handling the stopping strings
     stopping_criteria_list = transformers.StoppingCriteriaList()
-    if type(stopping_strings) is list and len(stopping_strings) > 0:
+    for st in [stopping_strings, state['custom_stopping_strings']]:
-        t = [encode(string, 0, add_special_tokens=False) for string in stopping_strings]
+        if type(st) is list and len(st) > 0:
-        stopping_criteria_list.append(_SentinelTokenStoppingCriteria(sentinel_token_ids=t, starting_idx=len(input_ids[0])))
+            sentinel_token_ids = [encode(string, add_special_tokens=False) for string in st]
+            stopping_criteria_list.append(_SentinelTokenStoppingCriteria(sentinel_token_ids=sentinel_token_ids, starting_idx=len(input_ids[0])))
+            break
 
     if not shared.args.flexgen:
         for k in ['max_new_tokens', 'do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping']:
             generate_params[k] = state[k]
         generate_params['eos_token_id'] = eos_token_ids
         generate_params['stopping_criteria'] = stopping_criteria_list
+        if state['ban_eos_token']:
+            generate_params['suppress_tokens'] = [shared.tokenizer.eos_token_id]
     else:
         for k in ['max_new_tokens', 'do_sample', 'temperature']:
             generate_params[k] = state[k]

presets/Default.txt

@@ -1,7 +1,6 @@
 do_sample=True
-top_p=0.5
+top_p=0.95
-top_k=40
+top_k=50
-temperature=0.7
+temperature=1
 repetition_penalty=1.2
 typical_p=1.0
-early_stopping=False

server.py

@@ -232,10 +232,8 @@ def create_model_menus():
 
 
 def create_settings_menus(default_preset):
+
     generate_params = load_preset_values(default_preset if not shared.args.flexgen else 'Naive', {}, return_dict=True)
-    for k in ['max_new_tokens', 'seed', 'stop_at_newline', 'chat_prompt_size', 'chat_generation_attempts', 'add_bos_token']:
-        generate_params[k] = shared.settings[k]
-    shared.gradio['generate_state'] = gr.State(generate_params)
 
     with gr.Row():
         with gr.Column():
@@ -265,7 +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():
@@ -273,7 +271,13 @@ def create_settings_menus(default_preset):
                     with gr.Column():
                         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')
-            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.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):
         with gr.Row():
@@ -284,7 +288,7 @@ def create_settings_menus(default_preset):
         with gr.Row():
             shared.gradio['upload_softprompt'] = gr.File(type='binary', file_types=['.zip'])
 
-    shared.gradio['preset_menu'].change(load_preset_values, [shared.gradio[k] for k in ['preset_menu', 'generate_state']], [shared.gradio[k] for k in ['generate_state', 'do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping']])
+    shared.gradio['preset_menu'].change(load_preset_values, [shared.gradio[k] for k in ['preset_menu', 'interface_state']], [shared.gradio[k] for k in ['interface_state', 'do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping']])
     shared.gradio['softprompts_menu'].change(load_soft_prompt, shared.gradio['softprompts_menu'], shared.gradio['softprompts_menu'], show_progress=True)
     shared.gradio['upload_softprompt'].upload(upload_soft_prompt, shared.gradio['upload_softprompt'], shared.gradio['softprompts_menu'])
 
@@ -358,7 +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']
+    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
@@ -368,6 +372,7 @@ def gather_interface_values(*args):
     output = {}
     for i, element in enumerate(shared.input_elements):
         output[element] = args[i]
+    output['custom_stopping_strings'] = eval(f"[{output['custom_stopping_strings']}]")
     return output
 
 
@@ -394,13 +399,15 @@ def create_interface():
                     shared.gradio['Continue'] = gr.Button('Continue')
                     shared.gradio['Impersonate'] = gr.Button('Impersonate')
                 with gr.Row():
-                    shared.gradio['Copy last reply'] = gr.Button('Copy last reply')
+                    shared.gradio['Send dummy message'] = gr.Button('Send dummy message')
+                    shared.gradio['Send dummy reply'] = gr.Button('Send dummy reply')
                     shared.gradio['Replace last reply'] = gr.Button('Replace last reply')
-                    shared.gradio['Remove last'] = gr.Button('Remove last')
+                    shared.gradio['Copy last reply'] = gr.Button('Copy last reply')
-
+                with gr.Row():
                     shared.gradio['Clear history'] = gr.Button('Clear history')
                     shared.gradio['Clear history-confirm'] = gr.Button('Confirm', variant="stop", visible=False)
                     shared.gradio['Clear history-cancel'] = gr.Button('Cancel', visible=False)
+                    shared.gradio['Remove last'] = gr.Button('Remove last')
 
                 shared.gradio["mode"] = gr.Radio(choices=["cai-chat", "chat", "instruct"], value="cai-chat", label="Mode")
                 shared.gradio["Instruction templates"] = gr.Dropdown(choices=get_available_instruction_templates(), label="Instruction template", value="None", visible=False, info="Change this according to the model/LoRA that you are using.")
@@ -453,7 +460,7 @@ def create_interface():
                             shared.gradio['chat_prompt_size'] = gr.Slider(minimum=shared.settings['chat_prompt_size_min'], maximum=shared.settings['chat_prompt_size_max'], step=1, label='Maximum prompt size in tokens', value=shared.settings['chat_prompt_size'])
                         with gr.Column():
                             shared.gradio['chat_generation_attempts'] = gr.Slider(minimum=shared.settings['chat_generation_attempts_min'], maximum=shared.settings['chat_generation_attempts_max'], value=shared.settings['chat_generation_attempts'], step=1, label='Generation attempts (for longer replies)')
-                            shared.gradio['stop_at_newline'] = gr.Checkbox(value=shared.settings['stop_at_newline'], label='Stop generating at new line character?')
+                            shared.gradio['stop_at_newline'] = gr.Checkbox(value=shared.settings['stop_at_newline'], label='Stop generating at new line character')
 
                 create_settings_menus(default_preset)
 
@@ -497,6 +504,16 @@ def create_interface():
                 lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False).then(
                 chat.save_history, shared.gradio['mode'], None, show_progress=False)
 
+            shared.gradio['Send dummy message'].click(
+                chat.send_dummy_message, [shared.gradio[k] for k in ['textbox', 'name1', 'name2', 'mode']], shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+
+            shared.gradio['Send dummy reply'].click(
+                chat.send_dummy_reply, [shared.gradio[k] for k in ['textbox', 'name1', 'name2', 'mode']], shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+
             shared.gradio['Clear history-confirm'].click(
                 lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, clear_arr).then(
                 chat.clear_chat_log, [shared.gradio[k] for k in ['name1', 'name2', 'greeting', 'mode']], shared.gradio['display']).then(
@@ -563,17 +580,19 @@ def create_interface():
             with gr.Tab("Parameters", elem_id="parameters"):
                 create_settings_menus(default_preset)
 
-            shared.input_params = [shared.gradio[k] for k in ['textbox', 'generate_state']]
+            shared.input_params = [shared.gradio[k] for k in ['textbox', 'interface_state']]
             output_params = [shared.gradio[k] for k in ['textbox', 'markdown', 'html']]
 
             gen_events.append(shared.gradio['Generate'].click(
                 gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
-                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[0]; element.scrollTop = element.scrollHeight}")
             )
 
             gen_events.append(shared.gradio['textbox'].submit(
                 gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
-                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[0]; element.scrollTop = element.scrollHeight}")
             )
 
             shared.gradio['Stop'].click(stop_everything_event, None, None, queue=False, cancels=gen_events if shared.args.no_stream else None)
@@ -607,22 +626,25 @@ def create_interface():
             with gr.Tab("Parameters", elem_id="parameters"):
                 create_settings_menus(default_preset)
 
-            shared.input_params = [shared.gradio[k] for k in ['textbox', 'generate_state']]
+            shared.input_params = [shared.gradio[k] for k in ['textbox', 'interface_state']]
             output_params = [shared.gradio[k] for k in ['output_textbox', 'markdown', 'html']]
 
             gen_events.append(shared.gradio['Generate'].click(
                 gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
-                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[1]; element.scrollTop = element.scrollHeight}")
             )
 
             gen_events.append(shared.gradio['textbox'].submit(
                 gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
-                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[1]; element.scrollTop = element.scrollHeight}")
             )
 
             gen_events.append(shared.gradio['Continue'].click(
                 gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
-                generate_reply, [shared.gradio['output_textbox']] + shared.input_params[1:], output_params, show_progress=shared.args.no_stream)
+                generate_reply, [shared.gradio['output_textbox']] + shared.input_params[1:], output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[1]; element.scrollTop = element.scrollHeight}")
             )
 
             shared.gradio['Stop'].click(stop_everything_event, None, None, queue=False, cancels=gen_events if shared.args.no_stream else None)

settings-template.json

@@ -8,8 +8,13 @@
     "context": "This is a conversation with your Assistant. The Assistant is very helpful and is eager to chat with you and answer your questions.",
     "greeting": "Hello there!",
     "end_of_turn": "",
+    "custom_stopping_strings": "",
     "stop_at_newline": false,
     "add_bos_token": true,
+    "ban_eos_token": true,
+    "truncation_length": 2048,
+    "truncation_length_min": 0,
+    "truncation_length_max": 4096,
     "chat_prompt_size": 2048,
     "chat_prompt_size_min": 0,
     "chat_prompt_size_max": 2048,