Simplify some chat functions

This commit is contained in:
oobabooga 2023-04-24 00:47:40 -03:00 committed by GitHub
parent 04b98a8485
commit 435f8cc0e7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -10,7 +10,6 @@ from pathlib import Path
import yaml import yaml
from PIL import Image from PIL import Image
import modules.extensions as extensions_module
import modules.shared as shared import modules.shared as shared
from modules.extensions import apply_extensions from modules.extensions import apply_extensions
from modules.html_generator import chat_html_wrapper, make_thumbnail from modules.html_generator import chat_html_wrapper, make_thumbnail
@ -30,8 +29,8 @@ def generate_chat_prompt(user_input, state, **kwargs):
chat_prompt_size = state['chat_prompt_size'] chat_prompt_size = state['chat_prompt_size']
if shared.soft_prompt: if shared.soft_prompt:
chat_prompt_size -= shared.soft_prompt_tensor.shape[1] chat_prompt_size -= shared.soft_prompt_tensor.shape[1]
max_length = min(get_max_prompt_length(state), chat_prompt_size)
max_length = min(get_max_prompt_length(state), chat_prompt_size)
if is_instruct: if is_instruct:
prefix1 = f"{state['name1']}\n" prefix1 = f"{state['name1']}\n"
prefix2 = f"{state['name2']}\n" prefix2 = f"{state['name2']}\n"
@ -57,7 +56,6 @@ def generate_chat_prompt(user_input, state, **kwargs):
min_rows = 2 min_rows = 2
rows.append(f"{prefix1.strip() if not is_instruct else prefix1}") rows.append(f"{prefix1.strip() if not is_instruct else prefix1}")
elif not _continue: elif not _continue:
# Adding the user message # Adding the user message
if len(user_input) > 0: if len(user_input) > 0:
this_prefix1 = prefix1.replace('<|round|>', f'{len(shared.history["internal"])}') # for ChatGLM this_prefix1 = prefix1.replace('<|round|>', f'{len(shared.history["internal"])}') # for ChatGLM
@ -68,8 +66,8 @@ def generate_chat_prompt(user_input, state, **kwargs):
while len(rows) > min_rows and len(encode(''.join(rows))[0]) >= max_length: while len(rows) > min_rows and len(encode(''.join(rows))[0]) >= max_length:
rows.pop(1) rows.pop(1)
prompt = ''.join(rows)
prompt = ''.join(rows)
if also_return_rows: if also_return_rows:
return prompt, rows return prompt, rows
else: else:
@ -81,6 +79,7 @@ def get_stopping_strings(state):
stopping_strings = [f"\n{state['name1']}", f"\n{state['name2']}"] stopping_strings = [f"\n{state['name1']}", f"\n{state['name2']}"]
else: else:
stopping_strings = [f"\n{state['name1']}:", f"\n{state['name2']}:"] stopping_strings = [f"\n{state['name1']}:", f"\n{state['name2']}:"]
stopping_strings += ast.literal_eval(f"[{state['custom_stopping_strings']}]") stopping_strings += ast.literal_eval(f"[{state['custom_stopping_strings']}]")
return stopping_strings return stopping_strings
@ -111,13 +110,13 @@ def extract_message_from_reply(reply, state):
break break
else: else:
continue continue
break break
return reply, next_character_found return reply, next_character_found
def chatbot_wrapper(text, state, regenerate=False, _continue=False): def chatbot_wrapper(text, state, regenerate=False, _continue=False):
if shared.model_name == 'None' or shared.model is None: if shared.model_name == 'None' or shared.model is None:
print("No model is loaded! Select one in the Model tab.") print("No model is loaded! Select one in the Model tab.")
yield shared.history['visible'] yield shared.history['visible']
@ -125,18 +124,30 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
# Defining some variables # Defining some variables
cumulative_reply = '' cumulative_reply = ''
last_reply = [shared.history['internal'][-1][1], shared.history['visible'][-1][1]] if _continue else None
just_started = True just_started = True
visible_text = None visible_text = None
eos_token = '\n' if state['stop_at_newline'] else None eos_token = '\n' if state['stop_at_newline'] else None
stopping_strings = get_stopping_strings(state) stopping_strings = get_stopping_strings(state)
text, visible_text = apply_extensions('input_hijack', text, visible_text) # Preparing the input
if not any((regenerate, _continue)):
text, visible_text = apply_extensions('input_hijack', text, visible_text)
if visible_text is None:
visible_text = text
if visible_text is None: text = apply_extensions('input', text)
visible_text = text # *Is typing...*
if not _continue: yield shared.history['visible'] + [[visible_text, shared.processing_message]]
text = apply_extensions("input", text) else:
text, visible_text = shared.history['internal'][-1][0], shared.history['visible'][-1][0]
if regenerate:
shared.history['visible'].pop()
shared.history['internal'].pop()
# *Is typing...*
yield shared.history['visible'] + [[visible_text, shared.processing_message]]
elif _continue:
last_reply = [shared.history['internal'][-1][1], shared.history['visible'][-1][1]]
yield shared.history['visible'][:-1] + [[visible_text, last_reply[1] + '...']]
# Generating the prompt # Generating the prompt
kwargs = {'_continue': _continue} kwargs = {'_continue': _continue}
@ -144,10 +155,6 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
if prompt is None: if prompt is None:
prompt = generate_chat_prompt(text, state, **kwargs) prompt = generate_chat_prompt(text, state, **kwargs)
# Yield *Is typing...*
if not any((regenerate, _continue)):
yield shared.history['visible'] + [[visible_text, shared.processing_message]]
# Generate # Generate
for i in range(state['chat_generation_attempts']): for i in range(state['chat_generation_attempts']):
reply = None reply = None
@ -158,26 +165,26 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
reply, next_character_found = extract_message_from_reply(reply, state) reply, next_character_found = extract_message_from_reply(reply, state)
visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply) visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply)
visible_reply = apply_extensions("output", visible_reply) visible_reply = apply_extensions("output", visible_reply)
if _continue:
sep = ' ' if last_reply[0][-1] not in [' ', '\n'] else ''
reply = last_reply[0] + sep + reply
sep = ' ' if last_reply[1][-1] not in [' ', '\n'] else ''
visible_reply = last_reply[1] + sep + visible_reply
# We need this global variable to handle the Stop event, # We need this global variable to handle the Stop event,
# otherwise gradio gets confused # otherwise gradio gets confused
if shared.stop_everything: if shared.stop_everything:
return shared.history['visible'] return shared.history['visible']
if just_started: if just_started:
just_started = False just_started = False
if not _continue: if not _continue:
shared.history['internal'].append(['', '']) shared.history['internal'].append(['', ''])
shared.history['visible'].append(['', '']) shared.history['visible'].append(['', ''])
if _continue: shared.history['internal'][-1] = [text, reply]
sep = list(map(lambda x: ' ' if len(x) > 0 and x[-1] != ' ' else '', last_reply)) shared.history['visible'][-1] = [visible_text, visible_reply]
shared.history['internal'][-1] = [text, f'{last_reply[0]}{sep[0]}{reply}'] yield shared.history['visible']
shared.history['visible'][-1] = [visible_text, f'{last_reply[1]}{sep[1]}{visible_reply}']
else:
shared.history['internal'][-1] = [text, reply]
shared.history['visible'][-1] = [visible_text, visible_reply]
if not shared.args.no_stream:
yield shared.history['visible']
if next_character_found: if next_character_found:
break break
@ -188,7 +195,6 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False):
def impersonate_wrapper(text, state): def impersonate_wrapper(text, state):
if shared.model_name == 'None' or shared.model is None: if shared.model_name == 'None' or shared.model is None:
print("No model is loaded! Select one in the Model tab.") print("No model is loaded! Select one in the Model tab.")
yield '' yield ''
@ -202,7 +208,6 @@ def impersonate_wrapper(text, state):
# Yield *Is typing...* # Yield *Is typing...*
yield shared.processing_message yield shared.processing_message
for i in range(state['chat_generation_attempts']): for i in range(state['chat_generation_attempts']):
reply = None reply = None
for reply in generate_reply(f"{prompt}{' ' if len(cumulative_reply) > 0 else ''}{cumulative_reply}", state, eos_token=eos_token, stopping_strings=stopping_strings): for reply in generate_reply(f"{prompt}{' ' if len(cumulative_reply) > 0 else ''}{cumulative_reply}", state, eos_token=eos_token, stopping_strings=stopping_strings):
@ -227,23 +232,16 @@ def regenerate_wrapper(text, state):
if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0: if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0:
yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode']) yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
else: else:
last_visible = shared.history['visible'].pop() for history in chatbot_wrapper('', state, regenerate=True):
last_internal = shared.history['internal'].pop() yield chat_html_wrapper(history, state['name1'], state['name2'], state['mode'])
# Yield '*Is typing...*'
yield chat_html_wrapper(shared.history['visible'] + [[last_visible[0], shared.processing_message]], state['name1'], state['name2'], state['mode'])
for history in chatbot_wrapper(last_internal[0], state, regenerate=True):
shared.history['visible'][-1] = [last_visible[0], history[-1][1]]
yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
def continue_wrapper(text, state): def continue_wrapper(text, state):
if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0: if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0:
yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode']) yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
else: else:
# Yield ' ...' for history in chatbot_wrapper('', state, _continue=True):
yield chat_html_wrapper(shared.history['visible'][:-1] + [[shared.history['visible'][-1][0], shared.history['visible'][-1][1] + ' ...']], state['name1'], state['name2'], state['mode']) yield chat_html_wrapper(history, state['name1'], state['name2'], state['mode'])
for history in chatbot_wrapper(shared.history['internal'][-1][0], state, _continue=True):
yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
def remove_last_message(name1, name2, mode): def remove_last_message(name1, name2, mode):
@ -281,6 +279,7 @@ def send_dummy_reply(text, name1, name2, mode):
if len(shared.history['visible']) > 0 and not shared.history['visible'][-1][1] == '': if len(shared.history['visible']) > 0 and not shared.history['visible'][-1][1] == '':
shared.history['visible'].append(['', '']) shared.history['visible'].append(['', ''])
shared.history['internal'].append(['', '']) shared.history['internal'].append(['', ''])
shared.history['visible'][-1][1] = text shared.history['visible'][-1][1] = text
shared.history['internal'][-1][1] = apply_extensions("input", text) shared.history['internal'][-1][1] = apply_extensions("input", text)
return chat_html_wrapper(shared.history['visible'], name1, name2, mode) return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
@ -300,7 +299,6 @@ def clear_chat_log(name1, name2, greeting, mode):
# Save cleared logs # Save cleared logs
save_history(mode) save_history(mode)
return chat_html_wrapper(shared.history['visible'], name1, name2, mode) return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
@ -321,8 +319,8 @@ def tokenize_dialogue(dialogue, name1, name2, mode):
for i in range(len(idx) - 1): for i in range(len(idx) - 1):
messages.append(dialogue[idx[i]:idx[i + 1]].strip()) messages.append(dialogue[idx[i]:idx[i + 1]].strip())
messages.append(dialogue[idx[-1]:].strip())
messages.append(dialogue[idx[-1]:].strip())
entry = ['', ''] entry = ['', '']
for i in messages: for i in messages:
if i.startswith(f'{name1}:'): if i.startswith(f'{name1}:'):
@ -331,6 +329,7 @@ def tokenize_dialogue(dialogue, name1, name2, mode):
entry[1] = i[len(f'{name2}:'):].strip() entry[1] = i[len(f'{name2}:'):].strip()
if not (len(entry[0]) == 0 and len(entry[1]) == 0): if not (len(entry[0]) == 0 and len(entry[1]) == 0):
history.append(entry) history.append(entry)
entry = ['', ''] entry = ['', '']
print("\033[1;32;1m\nDialogue tokenized to:\033[0;37;0m\n", end='') print("\033[1;32;1m\nDialogue tokenized to:\033[0;37;0m\n", end='')
@ -339,6 +338,7 @@ def tokenize_dialogue(dialogue, name1, name2, mode):
print("\n") print("\n")
for line in column.strip().split('\n'): for line in column.strip().split('\n'):
print("| " + line + "\n") print("| " + line + "\n")
print("|\n") print("|\n")
print("------------------------------") print("------------------------------")
@ -351,14 +351,17 @@ def save_history(mode, timestamp=False):
if mode == 'instruct': if mode == 'instruct':
if not timestamp: if not timestamp:
return return
fname = f"Instruct_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json" fname = f"Instruct_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
else: else:
if timestamp: if timestamp:
fname = f"{shared.character}_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json" fname = f"{shared.character}_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
else: else:
fname = f"{shared.character}_persistent.json" fname = f"{shared.character}_persistent.json"
if not Path('logs').exists(): if not Path('logs').exists():
Path('logs').mkdir() Path('logs').mkdir()
with open(Path(f'logs/{fname}'), 'w', encoding='utf-8') as f: with open(Path(f'logs/{fname}'), 'w', encoding='utf-8') as f:
f.write(json.dumps({'data': shared.history['internal'], 'data_visible': shared.history['visible']}, indent=2)) f.write(json.dumps({'data': shared.history['internal'], 'data_visible': shared.history['visible']}, indent=2))
@ -389,8 +392,10 @@ def build_pygmalion_style_context(data):
context = "" context = ""
if 'char_persona' in data and data['char_persona'] != '': if 'char_persona' in data and data['char_persona'] != '':
context += f"{data['char_name']}'s Persona: {data['char_persona']}\n" context += f"{data['char_name']}'s Persona: {data['char_persona']}\n"
if 'world_scenario' in data and data['world_scenario'] != '': if 'world_scenario' in data and data['world_scenario'] != '':
context += f"Scenario: {data['world_scenario']}\n" context += f"Scenario: {data['world_scenario']}\n"
context = f"{context.strip()}\n<START>\n" context = f"{context.strip()}\n<START>\n"
return context return context
@ -405,6 +410,7 @@ def generate_pfp_cache(character):
img = make_thumbnail(Image.open(path)) img = make_thumbnail(Image.open(path))
img.save(Path('cache/pfp_character.png'), format='PNG') img.save(Path('cache/pfp_character.png'), format='PNG')
return img return img
return None return None
@ -488,13 +494,17 @@ def upload_character(json_file, img, tavern=False):
while Path(f'characters/{outfile_name}.json').exists(): while Path(f'characters/{outfile_name}.json').exists():
outfile_name = f'{data["char_name"]}_{i:03d}' outfile_name = f'{data["char_name"]}_{i:03d}'
i += 1 i += 1
if tavern: if tavern:
outfile_name = f'TavernAI-{outfile_name}' outfile_name = f'TavernAI-{outfile_name}'
with open(Path(f'characters/{outfile_name}.json'), 'w', encoding='utf-8') as f: with open(Path(f'characters/{outfile_name}.json'), 'w', encoding='utf-8') as f:
f.write(json_file) f.write(json_file)
if img is not None: if img is not None:
img = Image.open(io.BytesIO(img)) img = Image.open(io.BytesIO(img))
img.save(Path(f'characters/{outfile_name}.png')) img.save(Path(f'characters/{outfile_name}.png'))
print(f'New character saved to "characters/{outfile_name}.json".') print(f'New character saved to "characters/{outfile_name}.json".')
return outfile_name return outfile_name