automatic/cli/train/train.py

323 lines
14 KiB
Python
Executable File

#!/bin/env python
# system imports
import os
import re
import gc
import sys
import json
import shutil
import pathlib
import asyncio
import tempfile
import argparse
# 3rd party imports
import filetype
from tqdm.rich import tqdm
# local imports
import util
import sdapi
import process
import latents
import options
# console handler
from rich import print # pylint: disable=redefined-builtin
from rich.pretty import install as pretty_install
from rich.traceback import install as traceback_install
from rich.console import Console
console = Console(log_time=True, log_time_format='%H:%M:%S-%f')
pretty_install(console=console)
import torch, accelerate, diffusers, requests, urllib3, http
traceback_install(console=console, extra_lines=1, width=console.width, word_wrap=False, indent_guides=False, suppress=[torch,accelerate,diffusers,asyncio,http,urllib3,requests])
# lora imports
lora_path = os.path.abspath(os.path.join(os.path.dirname(__file__), os.pardir, 'modules', 'lora'))
sys.path.append(lora_path)
lycoris_path = os.path.abspath(os.path.join(os.path.dirname(__file__), os.pardir, 'modules', 'lycoris'))
sys.path.append(lycoris_path)
import train_network
# globals
args = None
valid_steps = ['original', 'face', 'body', 'blur', 'range', 'upscale', 'restore', 'interrogate', 'resize', 'square', 'segment']
# methods
def mem_stats():
gc.collect()
if torch.cuda.is_available():
with torch.no_grad():
torch.cuda.empty_cache()
with torch.cuda.device('cuda'):
torch.cuda.empty_cache()
torch.cuda.ipc_collect()
mem = util.get_memory()
peak = { 'active': mem['gpu-active']['peak'], 'allocated': mem['gpu-allocated']['peak'], 'reserved': mem['gpu-reserved']['peak'] }
console.log(f"memory cpu: {mem.ram} gpu current: {mem.gpu} gpu peak: {peak}")
def parse_args():
global args
parser = argparse.ArgumentParser(description = 'train lora')
# basic section
parser.add_argument('--output', '--name', type=str, default=None, required=True, help='output filename')
parser.add_argument('--type', type=str, choices=['embedding', 'lora', 'lycoris', 'dreambooth'], default=None, required=True, help='training type')
parser.add_argument('--tag', type=str, default='person', required=False, help='primary tag, default: %(default)s')
parser.add_argument('--process', type=str, default='original,interrogate,resize,square', required=False, help=f'list of possible processing steps: {valid_steps}, default: %(default)s')
parser.add_argument('--dir', type=str, default='', required=False, help='where to store processed images, default is system temp/train')
parser.add_argument('--input', '--dataset', type=str, default=None, required=True, help='input folder with training images')
parser.add_argument('--overwrite', default = False, action='store_true', help = "overwrite existing training, default: %(default)s")
# global params
parser.add_argument('--gradient', type=int, default=1, required=False, help='gradient accumulation steps, default: %(default)s')
parser.add_argument('--steps', type=int, default=2500, required=False, help='training steps, default: %(default)s')
parser.add_argument('--batch', type=int, default=1, required=False, help='batch size, default: %(default)s')
parser.add_argument('--lr', type=float, default=1e-04, required=False, help='model learning rate, default: %(default)s')
parser.add_argument('--dim', '--vectors', type=int, default=40, required=False, help='network dimension, default: %(default)s')
# lora params
parser.add_argument('--repeats', type=int, default=10, required=False, help='number of repeats per image, default: %(default)s')
parser.add_argument('--alpha', type=float, default=0, required=False, help='alpha for weights scaling, default: half of dim')
args = parser.parse_args()
def prepare_server():
try:
server_status = util.Map(sdapi.progress())
server_state = server_status['state']
except:
console.log('server error:', server_status)
exit(1)
if server_state['job_count'] > 0:
console.log('server not idle:', server_state)
exit(1)
server_options = util.Map(sdapi.options())
server_options.options.save_training_settings_to_txt = False
server_options.options.training_enable_tensorboard = False
server_options.options.training_tensorboard_save_images = False
server_options.options.pin_memory = True
server_options.options.save_optimizer_state = False
server_options.options.training_image_repeats_per_epoch = args.repeats
server_options.options.training_write_csv_every = 0
sdapi.postsync('/sdapi/v1/options', server_options.options)
console.log(f'updated server options')
def verify_args():
global args
server_options = util.Map(sdapi.options())
args.model = server_options.options['sd_model_checkpoint'].split(' [')[0]
args.lora_dir = server_options.flags['lora_dir']
if not os.path.isabs(args.model) and not os.path.exists(args.model):
args.model = os.path.abspath(os.path.join(args.lora_dir, os.pardir, 'Stable-diffusion', args.model))
if not os.path.exists(args.model) or not os.path.isfile(args.model):
console.log('cannot find model:', args.model)
exit(1)
if not os.path.exists(args.input) or not os.path.isdir(args.input):
console.log('cannot find training folder:', args.input)
exit(1)
if not os.path.exists(args.lora_dir) or not os.path.isdir(args.lora_dir):
console.log('cannot find lora folder:', args.dir)
exit(1)
if args.dir != '':
args.process_dir = args.dir
else:
args.process_dir = os.path.join(tempfile.gettempdir(), 'train', args.output)
console.log(f'args: {vars(args)}')
async def training_loop():
async def async_train():
res = await sdapi.post('/sdapi/v1/train/embedding', options.embedding)
console.log(f'train embedding result: {res}')
async def async_monitor():
await asyncio.sleep(3)
res = util.Map(sdapi.progress())
with tqdm(desc='train embedding', total=res.state.job_count) as pbar:
while res.state.job_no < res.state.job_count and not res.state.interrupted and not res.state.skipped:
await asyncio.sleep(2)
prev_job = res.state.job_no
res = util.Map(sdapi.progress())
loss = re.search(r"Loss: (.*?)(?=\<)", res.textinfo)
if loss:
pbar.set_postfix({ 'loss': loss.group(0) })
pbar.update(res.state.job_no - prev_job)
a = asyncio.create_task(async_train())
b = asyncio.create_task(async_monitor())
await asyncio.gather(a, b) # wait for both pipeline and monitor to finish
def train_embedding():
console.log(f'{args.type} options: {options.embedding}')
create_options = util.Map({
"name": args.output,
"num_vectors_per_token": args.dim,
"overwrite_old": False,
"init_text": args.tag,
})
server_options = util.Map(sdapi.options())
fn = os.path.join(server_options.flags.embeddings_dir, args.output) + '.pt'
if os.path.exists(fn) and args.overwrite:
console.log(f'delete existing embedding {fn}')
os.remove(fn)
else:
console.log(f'embedding exists {fn}')
return
console.log(f'create embedding {create_options}')
res = sdapi.postsync('/sdapi/v1/create/embedding', create_options)
if 'info' in res and 'error' in res['info']: # formatted error
console.log(res.info)
elif 'info' in res: # no error
asyncio.run(training_loop())
else: # unknown error
console.log(f'create embedding error {res}')
def train_lora():
fn = os.path.join(args.lora_dir, args.output)
for ext in ['.ckpt', '.pt', '.safetensors']:
if os.path.exists(fn + ext):
if args.overwrite:
console.log(f'delete existing lora: {fn + ext}')
os.remove(fn + ext)
else:
console.log(f'lora exists: {fn + ext}')
return
console.log(f'{args.type} options: {options.lora}')
train_network.train(options.lora)
def prepare_options():
# lora specific
options.lora.pretrained_model_name_or_path = args.model
options.lora.output_dir = args.lora_dir
options.lora.output_name = args.output
options.lora.max_train_steps = args.steps
options.lora.network_dim = args.dim
options.lora.network_alpha = args.dim // 2 if args.alpha == 0 else args.alpha
options.lora.gradient_accumulation_steps = args.gradient
options.lora.learning_rate = args.lr
options.lora.train_batch_size = args.batch
options.lora.network_alpha = args.dim // 2 if args.alpha == 0 else args.alpha
options.lora.train_data_dir = args.process_dir
if args.type == 'lycoris':
console.log('train using lycoris network')
options.lora.network_module = 'lycoris.kohya'
options.lora.in_json = os.path.join(args.process_dir, args.output + '.json')
if args.type == 'dreambooth':
console.log('train using dreambooth style training')
options.lora.in_json = None
if args.type == 'lora':
console.log('train using lora style training')
options.lora.in_json = os.path.join(args.process_dir, args.output + '.json')
if args.type == 'embedding':
console.log('train embedding')
options.lora.in_json = None
pass
# embedding specific
options.embedding.embedding_name = args.output
options.embedding.learn_rate = str(args.lr)
options.embedding.batch_size = args.batch
options.embedding.steps = args.steps
options.embedding.data_root = args.process_dir
options.embedding.log_directory = os.path.join(args.process_dir, 'log')
options.embedding.gradient_step = args.gradient
def process_inputs():
pathlib.Path(args.process_dir).mkdir(parents=True, exist_ok=True)
processing_options = args.process.split(',') if isinstance(args.process, str) else args.process
processing_options = [opt.strip() for opt in re.split(',| ', args.process)]
console.log(f'processing steps: {processing_options}')
for step in processing_options:
if step not in valid_steps:
console.log(f'invalid processing step: {[step]}')
exit(1)
for root, _sub_dirs, folder in os.walk(args.input):
files = [os.path.join(root, f) for f in folder if filetype.is_image(os.path.join(root, f))]
console.log(f'processing input images: {len(files)}')
if os.path.exists(args.process_dir):
console.log('removing existing processed folder:', args.process_dir)
shutil.rmtree(args.process_dir, ignore_errors=True)
steps = [step for step in processing_options if step in ['face', 'body', 'original']]
process.reset()
metadata = {}
for step in steps:
if step == 'face':
opts = [step for step in processing_options if step not in ['body', 'original']]
if step == 'body':
opts = [step for step in processing_options if step not in ['face', 'original', 'upscale', 'restore']] # body does not perform upscale or restore
if step == 'original':
opts = [step for step in processing_options if step not in ['face', 'body', 'upscale', 'restore', 'blur', 'range', 'segment']] # original does not perform most steps
console.log(f'processing current step: {opts}')
tag = step
if tag == 'original' and args.tag is not None:
concept = args.tag.split(',')[0].strip()
else:
concept = step
if args.type in ['lora', 'lycoris', 'dreambooth']:
dir = os.path.join(args.process_dir, str(args.repeats) + '_' + concept) # separate concepts per folder
if args.type in ['embedding']:
dir = os.path.join(args.process_dir) # everything into same folder
console.log('processing concept:', concept)
console.log('processing output folder:', dir)
pathlib.Path(dir).mkdir(parents=True, exist_ok=True)
results = {}
for f in files:
res = process.file(filename = f, folder = dir, tag = args.tag, requested = opts)
if res.image: # valid result
results[res.type] = results.get(res.type, 0) + 1
results['total'] = results.get('total', 0) + 1
rel_path = res.basename.replace(os.path.commonpath([res.basename, args.process_dir]), '')
if rel_path.startswith(os.path.sep): rel_path = rel_path[1:]
metadata[rel_path] = { 'caption': res.caption, 'tags': ','.join(res.tags) }
if options.lora.in_json is None:
with open(res.output.replace(options.process.format, '.txt'), "w") as outfile:
outfile.write(res.caption)
console.log(f"processing {'saved' if res.image is not None else 'skipped'}: {f} => {res.output} {res.ops} {res.message}")
dirs = [os.path.join(args.process_dir, dir) for dir in os.listdir(args.process_dir) if os.path.isdir(os.path.join(args.process_dir, dir))]
console.log(f'input datasets {dirs}')
if options.lora.in_json is not None:
with open(options.lora.in_json, "w") as outfile: # write json at the end only
outfile.write(json.dumps(metadata, indent=2))
for dir in dirs: # create latents
latents.create_vae_latents(util.Map({ 'input': dir, 'json': options.lora.in_json }))
latents.unload_vae()
r = { 'inputs': len(files), 'outputs': results, 'metadata': options.lora.in_json }
console.log(f'processing steps result: {r}')
if args.gradient < 0:
console.log(f"setting gradient accumulation to number of images: {results['total']}")
options.lora.gradient_accumulation_steps = results['total']
options.embedding.gradient_step = results['total']
process.unload()
if __name__ == '__main__':
console.log('train script for stable diffusion')
parse_args()
prepare_server()
verify_args()
prepare_options()
mem_stats()
process_inputs()
mem_stats()
try:
if args.type == 'embedding':
train_embedding()
if args.type == 'lora' or args.type == 'lycoris' or args.type == 'dreambooth':
train_lora()
except KeyboardInterrupt as e:
console.log('interrupt requested')
sdapi.interrupt()
mem_stats()
console.log('done')