text-generation-webui/server.py

309 lines
13 KiB
Python
Raw Normal View History

2022-12-21 11:27:31 -05:00
import re
2023-01-05 23:33:21 -05:00
import time
import glob
2023-01-06 17:56:44 -05:00
from sys import exit
2022-12-21 11:27:31 -05:00
import torch
2023-01-06 17:56:44 -05:00
import argparse
from pathlib import Path
2022-12-21 11:27:31 -05:00
import gradio as gr
import transformers
2023-01-06 21:14:08 -05:00
from html_generator import *
2023-01-10 21:39:50 -05:00
from transformers import AutoTokenizer, AutoModelForCausalLM
2023-01-14 22:39:51 -05:00
import warnings
2022-12-21 11:27:31 -05:00
2023-01-06 21:14:08 -05:00
2023-01-06 17:56:44 -05:00
parser = argparse.ArgumentParser()
2023-01-06 18:22:26 -05:00
parser.add_argument('--model', type=str, help='Name of the model to load by default.')
parser.add_argument('--notebook', action='store_true', help='Launch the webui in notebook mode, where the output is written to the same text box as the input.')
2023-01-07 20:52:46 -05:00
parser.add_argument('--chat', action='store_true', help='Launch the webui in chat mode.')
2023-01-09 08:58:46 -05:00
parser.add_argument('--cpu', action='store_true', help='Use the CPU to generate text.')
parser.add_argument('--auto-devices', action='store_true', help='Automatically split the model across the available GPU(s) and CPU.')
parser.add_argument('--load-in-8bit', action='store_true', help='Load the model with 8-bit precision.')
2023-01-10 23:10:11 -05:00
parser.add_argument('--no-listen', action='store_true', help='Make the webui unreachable from your local network.')
2023-01-06 17:56:44 -05:00
args = parser.parse_args()
2023-01-14 22:39:51 -05:00
2023-01-06 00:06:59 -05:00
loaded_preset = None
available_models = sorted(set(map(lambda x : str(x.name).replace('.pt', ''), list(Path('models/').glob('*'))+list(Path('torch-dumps/').glob('*')))))
2023-01-06 21:14:08 -05:00
available_models = [item for item in available_models if not item.endswith('.txt')]
2023-01-10 23:17:20 -05:00
available_models = sorted(available_models, key=str.lower)
available_presets = sorted(set(map(lambda x : str(x.name).split('.')[0], list(Path('presets').glob('*.txt')))))
2023-01-05 23:33:21 -05:00
2023-01-14 22:39:51 -05:00
transformers.logging.set_verbosity_error()
2022-12-21 11:27:31 -05:00
def load_model(model_name):
2023-01-05 23:41:52 -05:00
print(f"Loading {model_name}...")
2022-12-21 11:27:31 -05:00
t0 = time.time()
2023-01-05 23:41:52 -05:00
# Default settings
if not (args.cpu or args.auto_devices or args.load_in_8bit):
if Path(f"torch-dumps/{model_name}.pt").exists():
print("Loading in .pt format...")
model = torch.load(Path(f"torch-dumps/{model_name}.pt"))
elif model_name.lower().startswith(('gpt-neo', 'opt-', 'galactica')) and any(size in model_name.lower() for size in ('13b', '20b', '30b')):
model = AutoModelForCausalLM.from_pretrained(Path(f"models/{model_name}"), device_map='auto', load_in_8bit=True)
else:
model = AutoModelForCausalLM.from_pretrained(Path(f"models/{model_name}"), low_cpu_mem_usage=True, torch_dtype=torch.float16).cuda()
# Custom
2023-01-06 00:54:33 -05:00
else:
settings = ["low_cpu_mem_usage=True"]
cuda = ""
2023-01-10 21:39:50 -05:00
command = "AutoModelForCausalLM.from_pretrained"
2023-01-09 14:28:04 -05:00
if args.cpu:
settings.append("torch_dtype=torch.float32")
2023-01-09 14:28:04 -05:00
else:
if args.load_in_8bit:
settings.append("device_map='auto'")
settings.append("load_in_8bit=True")
else:
settings.append("torch_dtype=torch.float16")
if args.auto_devices:
settings.append("device_map='auto'")
else:
cuda = ".cuda()"
settings = ', '.join(settings)
command = f"{command}(Path(f'models/{model_name}'), {settings}){cuda}"
model = eval(command)
2022-12-21 11:27:31 -05:00
2023-01-06 00:54:33 -05:00
# Loading the tokenizer
2023-01-10 23:10:11 -05:00
if model_name.lower().startswith(('gpt4chan', 'gpt-4chan', '4chan')) and Path(f"models/gpt-j-6B/").exists():
tokenizer = AutoTokenizer.from_pretrained(Path("models/gpt-j-6B/"))
2022-12-21 11:27:31 -05:00
else:
tokenizer = AutoTokenizer.from_pretrained(Path(f"models/{model_name}/"))
2022-12-21 11:27:31 -05:00
2023-01-06 00:06:59 -05:00
print(f"Loaded the model in {(time.time()-t0):.2f} seconds.")
2022-12-21 11:27:31 -05:00
return model, tokenizer
2023-01-06 00:26:33 -05:00
# Removes empty replies from gpt4chan outputs
2022-12-21 11:27:31 -05:00
def fix_gpt4chan(s):
for i in range(10):
s = re.sub("--- [0-9]*\n>>[0-9]*\n---", "---", s)
s = re.sub("--- [0-9]*\n *\n---", "---", s)
s = re.sub("--- [0-9]*\n\n\n---", "---", s)
return s
2023-01-10 23:10:11 -05:00
# Fix the LaTeX equations in GALACTICA
2023-01-06 23:56:21 -05:00
def fix_galactica(s):
s = s.replace(r'\[', r'$')
s = s.replace(r'\]', r'$')
2023-01-07 10:13:09 -05:00
s = s.replace(r'\(', r'$')
s = s.replace(r'\)', r'$')
s = s.replace(r'$$', r'$')
2023-01-06 23:56:21 -05:00
return s
2023-01-10 23:10:11 -05:00
def generate_html(s):
s = '\n'.join([f'<p style="margin-bottom: 20px">{line}</p>' for line in s.split('\n')])
s = f'<div style="max-width: 600px; margin-left: auto; margin-right: auto; background-color:#eef2ff; color:#0b0f19; padding:3em; font-size:1.2em;">{s}</div>'
return s
2023-01-13 12:28:53 -05:00
def generate_reply(question, tokens, inference_settings, selected_model, eos_token=None):
2023-01-06 00:06:59 -05:00
global model, tokenizer, model_name, loaded_preset, preset
2022-12-21 11:27:31 -05:00
if selected_model != model_name:
model_name = selected_model
model = None
2023-01-08 12:37:43 -05:00
tokenizer = None
2023-01-09 08:58:46 -05:00
if not args.cpu:
torch.cuda.empty_cache()
2022-12-21 11:27:31 -05:00
model, tokenizer = load_model(model_name)
2023-01-06 00:06:59 -05:00
if inference_settings != loaded_preset:
with open(Path(f'presets/{inference_settings}.txt'), 'r') as infile:
2023-01-05 23:33:21 -05:00
preset = infile.read()
2023-01-06 00:06:59 -05:00
loaded_preset = inference_settings
2022-12-21 11:27:31 -05:00
2023-01-09 08:58:46 -05:00
if not args.cpu:
torch.cuda.empty_cache()
input_ids = tokenizer.encode(str(question), return_tensors='pt').cuda()
cuda = ".cuda()"
else:
input_ids = tokenizer.encode(str(question), return_tensors='pt')
cuda = ""
2022-12-21 11:27:31 -05:00
if eos_token is None:
2023-01-09 08:58:46 -05:00
output = eval(f"model.generate(input_ids, {preset}){cuda}")
else:
2023-01-09 10:56:54 -05:00
n = tokenizer.encode(eos_token, return_tensors='pt')[0][-1]
2023-01-09 08:58:46 -05:00
output = eval(f"model.generate(input_ids, eos_token_id={n}, {preset}){cuda}")
2023-01-06 21:14:08 -05:00
2023-01-09 08:58:46 -05:00
reply = tokenizer.decode(output[0], skip_special_tokens=True)
2023-01-10 23:10:11 -05:00
reply = reply.replace(r'<|endoftext|>', '')
2023-01-06 18:22:26 -05:00
if model_name.lower().startswith('galactica'):
2023-01-06 23:56:21 -05:00
reply = fix_galactica(reply)
2023-01-10 23:10:11 -05:00
return reply, reply, generate_html(reply)
2023-01-06 21:14:08 -05:00
elif model_name.lower().startswith('gpt4chan'):
2023-01-06 23:56:21 -05:00
reply = fix_gpt4chan(reply)
2023-01-10 23:10:11 -05:00
return reply, 'Only applicable for galactica models.', generate_4chan_html(reply)
2023-01-06 18:22:26 -05:00
else:
2023-01-10 23:10:11 -05:00
return reply, 'Only applicable for galactica models.', generate_html(reply)
2022-12-21 11:27:31 -05:00
2023-01-06 17:56:44 -05:00
# Choosing the default model
if args.model is not None:
model_name = args.model
else:
2023-01-06 20:05:37 -05:00
if len(available_models) == 0:
2023-01-06 17:56:44 -05:00
print("No models are available! Please download at least one.")
exit(0)
elif len(available_models) == 1:
i = 0
else:
print("The following models are available:\n")
for i,model in enumerate(available_models):
print(f"{i+1}. {model}")
print(f"\nWhich one do you want to load? 1-{len(available_models)}\n")
i = int(input())-1
2023-01-09 10:56:54 -05:00
print()
2023-01-06 17:56:44 -05:00
model_name = available_models[i]
2022-12-21 11:27:31 -05:00
model, tokenizer = load_model(model_name)
2023-01-06 17:56:44 -05:00
# UI settings
2023-01-07 17:11:21 -05:00
if model_name.lower().startswith('gpt4chan'):
2022-12-21 11:27:31 -05:00
default_text = "-----\n--- 865467536\nInput text\n--- 865467537\n"
else:
default_text = "Common sense questions and answers\n\nQuestion: \nFactual answer:"
description = f"""
2023-01-07 17:11:21 -05:00
2023-01-06 18:22:26 -05:00
# Text generation lab
Generate text using Large Language Models.
"""
css=".my-4 {margin-top: 0} .py-6 {padding-top: 2.5rem}"
2023-01-06 18:22:26 -05:00
if args.notebook:
with gr.Blocks(css=css, analytics_enabled=False) as interface:
gr.Markdown(description)
2023-01-06 20:05:37 -05:00
with gr.Tab('Raw'):
textbox = gr.Textbox(value=default_text, lines=23)
with gr.Tab('Markdown'):
markdown = gr.Markdown()
2023-01-06 21:14:08 -05:00
with gr.Tab('HTML'):
html = gr.HTML()
2023-01-06 18:22:26 -05:00
btn = gr.Button("Generate")
2023-01-06 20:05:37 -05:00
2023-01-13 12:28:53 -05:00
length_slider = gr.Slider(minimum=1, maximum=2000, step=1, label='max_new_tokens', value=200)
2023-01-06 20:05:37 -05:00
with gr.Row():
with gr.Column():
model_menu = gr.Dropdown(choices=available_models, value=model_name, label='Model')
2023-01-13 12:28:53 -05:00
with gr.Column():
preset_menu = gr.Dropdown(choices=available_presets, value="NovelAI-Sphinx Moth", label='Settings preset')
2023-01-06 18:22:26 -05:00
btn.click(generate_reply, [textbox, length_slider, preset_menu, model_menu], [textbox, markdown, html], show_progress=True, api_name="textgen")
textbox.submit(generate_reply, [textbox, length_slider, preset_menu, model_menu], [textbox, markdown, html], show_progress=True)
2023-01-07 20:52:46 -05:00
elif args.chat:
history = []
2023-01-14 22:39:51 -05:00
# This gets the new line characters right.
def chat_response_cleaner(text):
2023-01-14 21:50:34 -05:00
text = text.replace('\n', '\n\n')
text = re.sub(r"\n{3,}", "\n\n", text)
text = text.strip()
2023-01-14 22:39:51 -05:00
return text
def chatbot_wrapper(text, tokens, inference_settings, selected_model, name1, name2, context, check):
text = chat_response_cleaner(text)
2023-01-14 21:50:34 -05:00
2023-01-07 20:52:46 -05:00
question = context+'\n\n'
for i in range(len(history)):
question += f"{name1}: {history[i][0][3:-5].strip()}\n"
question += f"{name2}: {history[i][1][3:-5].strip()}\n"
2023-01-14 21:50:34 -05:00
question += f"{name1}: {text}\n"
2023-01-07 20:52:46 -05:00
question += f"{name2}:"
if check:
reply = generate_reply(question, tokens, inference_settings, selected_model, eos_token='\n')[0]
reply = reply[len(question):].split('\n')[0].strip()
else:
reply = generate_reply(question, tokens, inference_settings, selected_model)[0]
2023-01-14 21:26:14 -05:00
reply = reply[len(question):]
idx = reply.find(f"\n{name1}:")
if idx != -1:
reply = reply[:idx]
2023-01-14 22:39:51 -05:00
reply = chat_response_cleaner(response)
2023-01-07 20:52:46 -05:00
history.append((text, reply))
return history
def remove_last_message():
history.pop()
return history
2023-01-07 20:52:46 -05:00
def clear():
global history
history = []
if 'pygmalion' in model_name.lower():
context_str = "This is a conversation between two people.\n<START>"
name1_str = "You"
name2_str = "Kawaii"
else:
context_str = "This is a conversation between two people."
name1_str = "Person 1"
name2_str = "Person 2"
with gr.Blocks(css=css+".h-\[40vh\] {height: 50vh}", analytics_enabled=False) as interface:
gr.Markdown(description)
2023-01-09 15:23:43 -05:00
with gr.Row():
2023-01-07 20:52:46 -05:00
with gr.Column():
2023-01-13 12:28:53 -05:00
length_slider = gr.Slider(minimum=1, maximum=2000, step=1, label='max_new_tokens', value=200)
2023-01-09 15:23:43 -05:00
with gr.Row():
2023-01-07 20:52:46 -05:00
with gr.Column():
model_menu = gr.Dropdown(choices=available_models, value=model_name, label='Model')
2023-01-07 20:52:46 -05:00
with gr.Column():
2023-01-13 12:28:53 -05:00
preset_menu = gr.Dropdown(choices=available_presets, value="NovelAI-Sphinx Moth", label='Settings preset')
name1 = gr.Textbox(value=name1_str, lines=1, label='Your name')
name2 = gr.Textbox(value=name2_str, lines=1, label='Bot\'s name')
context = gr.Textbox(value=context_str, lines=2, label='Context')
with gr.Row():
check = gr.Checkbox(value=True, label='Stop generating at new line character?')
2023-01-07 20:52:46 -05:00
with gr.Column():
display1 = gr.Chatbot()
2023-01-07 20:55:54 -05:00
textbox = gr.Textbox(lines=2, label='Input')
2023-01-07 20:52:46 -05:00
btn = gr.Button("Generate")
with gr.Row():
with gr.Column():
btn3 = gr.Button("Remove last message")
with gr.Column():
btn2 = gr.Button("Clear history")
2023-01-07 20:52:46 -05:00
btn.click(chatbot_wrapper, [textbox, length_slider, preset_menu, model_menu, name1, name2, context, check], display1, show_progress=True, api_name="textgen")
textbox.submit(chatbot_wrapper, [textbox, length_slider, preset_menu, model_menu, name1, name2, context, check], display1, show_progress=True)
btn3.click(remove_last_message, [], display1, show_progress=False)
2023-01-07 20:52:46 -05:00
btn2.click(clear)
2023-01-07 23:10:02 -05:00
btn.click(lambda x: "", textbox, textbox, show_progress=False)
2023-01-07 23:33:45 -05:00
textbox.submit(lambda x: "", textbox, textbox, show_progress=False)
2023-01-07 23:10:02 -05:00
btn2.click(lambda x: "", display1, display1)
2023-01-06 18:22:26 -05:00
else:
2023-01-10 23:33:57 -05:00
2023-01-13 12:28:53 -05:00
def continue_wrapper(question, tokens, inference_settings, selected_model):
a, b, c = generate_reply(question, tokens, inference_settings, selected_model)
2023-01-10 23:33:57 -05:00
return a, a, b, c
with gr.Blocks(css=css, analytics_enabled=False) as interface:
gr.Markdown(description)
2023-01-06 20:05:37 -05:00
with gr.Row():
with gr.Column():
textbox = gr.Textbox(value=default_text, lines=15, label='Input')
2023-01-13 12:28:53 -05:00
length_slider = gr.Slider(minimum=1, maximum=2000, step=1, label='max_new_tokens', value=200)
preset_menu = gr.Dropdown(choices=available_presets, value="NovelAI-Sphinx Moth", label='Settings preset')
2023-01-06 20:05:37 -05:00
model_menu = gr.Dropdown(choices=available_models, value=model_name, label='Model')
btn = gr.Button("Generate")
2023-01-10 23:33:57 -05:00
cont = gr.Button("Continue")
2023-01-06 20:05:37 -05:00
with gr.Column():
with gr.Tab('Raw'):
2023-01-10 23:36:11 -05:00
output_textbox = gr.Textbox(lines=15, label='Output')
2023-01-06 20:05:37 -05:00
with gr.Tab('Markdown'):
markdown = gr.Markdown()
2023-01-06 21:14:08 -05:00
with gr.Tab('HTML'):
html = gr.HTML()
2023-01-06 20:05:37 -05:00
btn.click(generate_reply, [textbox, length_slider, preset_menu, model_menu], [output_textbox, markdown, html], show_progress=True, api_name="textgen")
cont.click(continue_wrapper, [output_textbox, length_slider, preset_menu, model_menu], [output_textbox, textbox, markdown, html], show_progress=True)
textbox.submit(generate_reply, [textbox, length_slider, preset_menu, model_menu], [output_textbox, markdown, html], show_progress=True)
2022-12-21 11:27:31 -05:00
2023-01-10 23:10:11 -05:00
if args.no_listen:
2023-01-09 17:05:36 -05:00
interface.launch(share=False)
2023-01-10 23:10:11 -05:00
else:
interface.launch(share=False, server_name="0.0.0.0")