FLUX-VisionReply / train /train_personalized.py
gokaygokay's picture
full_files
2f4febc
raw
history blame
39.8 kB
import torch
import json
import yaml
import torchvision
from torch import nn, optim
from transformers import AutoTokenizer, CLIPTextModelWithProjection, CLIPVisionModelWithProjection
from warmup_scheduler import GradualWarmupScheduler
import torch.multiprocessing as mp
import os
import numpy as np
import re
import sys
sys.path.append(os.path.abspath('./'))
from dataclasses import dataclass
from torch.distributed import init_process_group, destroy_process_group, barrier
from gdf import GDF_dual_fixlrt as GDF
from gdf import EpsilonTarget, CosineSchedule
from gdf import VPScaler, CosineTNoiseCond, DDPMSampler, P2LossWeight, AdaptiveLossWeight
from torchtools.transforms import SmartCrop
from fractions import Fraction
from modules.effnet import EfficientNetEncoder
from modules.model_4stage_lite import StageC, ResBlock, AttnBlock, TimestepBlock, FeedForwardBlock
from modules.common_ckpt import GlobalResponseNorm
from modules.previewer import Previewer
from core.data import Bucketeer
from train.base import DataCore, TrainingCore
from tqdm import tqdm
from core import WarpCore
from core.utils import EXPECTED, EXPECTED_TRAIN, load_or_fail
from accelerate import init_empty_weights
from accelerate.utils import set_module_tensor_to_device
from contextlib import contextmanager
from train.dist_core import *
import glob
from torch.utils.data import DataLoader, Dataset
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.utils.data.distributed import DistributedSampler
from PIL import Image
from core.utils import EXPECTED, EXPECTED_TRAIN, update_weights_ema, create_folder_if_necessary
from core.utils import Base
import torch.nn.functional as F
import functools
import math
import copy
import random
from modules.lora import apply_lora, apply_retoken, LoRA, ReToken
Image.MAX_IMAGE_PIXELS = None
torch.manual_seed(23)
random.seed(23)
np.random.seed(23)
#7978026
class Null_Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
pass
def identity(x):
if isinstance(x, bytes):
x = x.decode('utf-8')
return x
def check_nan_inmodel(model, meta=''):
for name, param in model.named_parameters():
if torch.isnan(param).any():
print(f"nan detected in {name}", meta)
return True
print('no nan', meta)
return False
class mydist_dataset(Dataset):
def __init__(self, rootpath, tmp_prompt, img_processor=None):
self.img_pathlist = glob.glob(os.path.join(rootpath, '*.jpg'))
self.img_pathlist = self.img_pathlist * 100000
self.img_processor = img_processor
self.length = len( self.img_pathlist)
self.caption = tmp_prompt
def __getitem__(self, idx):
imgpath = self.img_pathlist[idx]
txt = self.caption
try:
img = Image.open(imgpath).convert('RGB')
w, h = img.size
if self.img_processor is not None:
img = self.img_processor(img)
except:
print('exception', imgpath)
return self.__getitem__(random.randint(0, self.length -1 ) )
return dict(captions=txt, images=img)
def __len__(self):
return self.length
class WurstCore(TrainingCore, DataCore, WarpCore):
@dataclass(frozen=True)
class Config(TrainingCore.Config, DataCore.Config, WarpCore.Config):
# TRAINING PARAMS
lr: float = EXPECTED_TRAIN
warmup_updates: int = EXPECTED_TRAIN
dtype: str = None
# MODEL VERSION
model_version: str = EXPECTED # 3.6B or 1B
clip_image_model_name: str = 'openai/clip-vit-large-patch14'
clip_text_model_name: str = 'laion/CLIP-ViT-bigG-14-laion2B-39B-b160k'
# CHECKPOINT PATHS
effnet_checkpoint_path: str = EXPECTED
previewer_checkpoint_path: str = EXPECTED
generator_checkpoint_path: str = None
ultrapixel_path: str = EXPECTED
# gdf customization
adaptive_loss_weight: str = None
# LoRA STUFF
module_filters: list = EXPECTED
rank: int = EXPECTED
train_tokens: list = EXPECTED
use_ddp: bool=EXPECTED
tmp_prompt: str=EXPECTED
@dataclass(frozen=True)
class Data(Base):
dataset: Dataset = EXPECTED
dataloader: DataLoader = EXPECTED
iterator: any = EXPECTED
sampler: DistributedSampler = EXPECTED
@dataclass(frozen=True)
class Models(TrainingCore.Models, DataCore.Models, WarpCore.Models):
effnet: nn.Module = EXPECTED
previewer: nn.Module = EXPECTED
train_norm: nn.Module = EXPECTED
train_lora: nn.Module = EXPECTED
@dataclass(frozen=True)
class Schedulers(WarpCore.Schedulers):
generator: any = None
@dataclass(frozen=True)
class Extras(TrainingCore.Extras, DataCore.Extras, WarpCore.Extras):
gdf: GDF = EXPECTED
sampling_configs: dict = EXPECTED
effnet_preprocess: torchvision.transforms.Compose = EXPECTED
info: TrainingCore.Info
config: Config
def setup_extras_pre(self) -> Extras:
gdf = GDF(
schedule=CosineSchedule(clamp_range=[0.0001, 0.9999]),
input_scaler=VPScaler(), target=EpsilonTarget(),
noise_cond=CosineTNoiseCond(),
loss_weight=AdaptiveLossWeight() if self.config.adaptive_loss_weight is True else P2LossWeight(),
)
sampling_configs = {"cfg": 5, "sampler": DDPMSampler(gdf), "shift": 1, "timesteps": 20}
if self.info.adaptive_loss is not None:
gdf.loss_weight.bucket_ranges = torch.tensor(self.info.adaptive_loss['bucket_ranges'])
gdf.loss_weight.bucket_losses = torch.tensor(self.info.adaptive_loss['bucket_losses'])
effnet_preprocess = torchvision.transforms.Compose([
torchvision.transforms.Normalize(
mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)
)
])
clip_preprocess = torchvision.transforms.Compose([
torchvision.transforms.Resize(224, interpolation=torchvision.transforms.InterpolationMode.BICUBIC),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.Normalize(
mean=(0.48145466, 0.4578275, 0.40821073), std=(0.26862954, 0.26130258, 0.27577711)
)
])
if self.config.training:
transforms = torchvision.transforms.Compose([
torchvision.transforms.ToTensor(),
torchvision.transforms.Resize(self.config.image_size[-1], interpolation=torchvision.transforms.InterpolationMode.BILINEAR, antialias=True),
SmartCrop(self.config.image_size, randomize_p=0.3, randomize_q=0.2)
])
else:
transforms = None
return self.Extras(
gdf=gdf,
sampling_configs=sampling_configs,
transforms=transforms,
effnet_preprocess=effnet_preprocess,
clip_preprocess=clip_preprocess
)
def get_conditions(self, batch: dict, models: Models, extras: Extras, is_eval=False, is_unconditional=False,
eval_image_embeds=False, return_fields=None):
conditions = super().get_conditions(
batch, models, extras, is_eval, is_unconditional,
eval_image_embeds, return_fields=return_fields or ['clip_text', 'clip_text_pooled', 'clip_img']
)
return conditions
def setup_models(self, extras: Extras) -> Models: # configure model
dtype = getattr(torch, self.config.dtype) if self.config.dtype else torch.bfloat16
# EfficientNet encoderin
effnet = EfficientNetEncoder()
effnet_checkpoint = load_or_fail(self.config.effnet_checkpoint_path)
effnet.load_state_dict(effnet_checkpoint if 'state_dict' not in effnet_checkpoint else effnet_checkpoint['state_dict'])
effnet.eval().requires_grad_(False).to(self.device)
del effnet_checkpoint
# Previewer
previewer = Previewer()
previewer_checkpoint = load_or_fail(self.config.previewer_checkpoint_path)
previewer.load_state_dict(previewer_checkpoint if 'state_dict' not in previewer_checkpoint else previewer_checkpoint['state_dict'])
previewer.eval().requires_grad_(False).to(self.device)
del previewer_checkpoint
@contextmanager
def dummy_context():
yield None
loading_context = dummy_context if self.config.training else init_empty_weights
# Diffusion models
with loading_context():
generator_ema = None
if self.config.model_version == '3.6B':
generator = StageC()
if self.config.ema_start_iters is not None: # default setting
generator_ema = StageC()
elif self.config.model_version == '1B':
print('in line 155 1b light model', self.config.model_version )
generator = StageC(c_cond=1536, c_hidden=[1536, 1536], nhead=[24, 24], blocks=[[4, 12], [12, 4]])
if self.config.ema_start_iters is not None and self.config.training:
generator_ema = StageC(c_cond=1536, c_hidden=[1536, 1536], nhead=[24, 24], blocks=[[4, 12], [12, 4]])
else:
raise ValueError(f"Unknown model version {self.config.model_version}")
if loading_context is dummy_context:
generator.load_state_dict( load_or_fail(self.config.generator_checkpoint_path))
else:
for param_name, param in load_or_fail(self.config.generator_checkpoint_path).items():
set_module_tensor_to_device(generator, param_name, "cpu", value=param)
generator._init_extra_parameter()
generator = generator.to(torch.bfloat16).to(self.device)
train_norm = nn.ModuleList()
cnt_norm = 0
for mm in generator.modules():
if isinstance(mm, GlobalResponseNorm):
train_norm.append(Null_Model())
cnt_norm += 1
train_norm.append(generator.agg_net)
train_norm.append(generator.agg_net_up)
sdd = torch.load(self.config.ultrapixel_path, map_location='cpu')
collect_sd = {}
for k, v in sdd.items():
collect_sd[k[7:]] = v
train_norm.load_state_dict(collect_sd)
# CLIP encoders
tokenizer = AutoTokenizer.from_pretrained(self.config.clip_text_model_name)
text_model = CLIPTextModelWithProjection.from_pretrained( self.config.clip_text_model_name).requires_grad_(False).to(dtype).to(self.device)
image_model = CLIPVisionModelWithProjection.from_pretrained(self.config.clip_image_model_name).requires_grad_(False).to(dtype).to(self.device)
# PREPARE LORA
train_lora = nn.ModuleList()
update_tokens = []
for tkn_regex, aggr_regex in self.config.train_tokens:
if (tkn_regex.startswith('[') and tkn_regex.endswith(']')) or (tkn_regex.startswith('<') and tkn_regex.endswith('>')):
# Insert new token
tokenizer.add_tokens([tkn_regex])
# add new zeros embedding
new_embedding = torch.zeros_like(text_model.text_model.embeddings.token_embedding.weight.data)[:1]
if aggr_regex is not None: # aggregate embeddings to provide an interesting baseline
aggr_tokens = [v for k, v in tokenizer.vocab.items() if re.search(aggr_regex, k) is not None]
if len(aggr_tokens) > 0:
new_embedding = text_model.text_model.embeddings.token_embedding.weight.data[aggr_tokens].mean(dim=0, keepdim=True)
elif self.is_main_node:
print(f"WARNING: No tokens found for aggregation regex {aggr_regex}. It will be initialized as zeros.")
text_model.text_model.embeddings.token_embedding.weight.data = torch.cat([
text_model.text_model.embeddings.token_embedding.weight.data, new_embedding
], dim=0)
selected_tokens = [len(tokenizer.vocab) - 1]
else:
selected_tokens = [v for k, v in tokenizer.vocab.items() if re.search(tkn_regex, k) is not None]
update_tokens += selected_tokens
update_tokens = list(set(update_tokens)) # remove duplicates
apply_retoken(text_model.text_model.embeddings.token_embedding, update_tokens)
apply_lora(generator, filters=self.config.module_filters, rank=self.config.rank)
for module in generator.modules():
if isinstance(module, LoRA) or (hasattr(module, '_fsdp_wrapped_module') and isinstance(module._fsdp_wrapped_module, LoRA)):
train_lora.append(module)
train_lora.append(text_model.text_model.embeddings.token_embedding.parametrizations.weight[0])
if os.path.exists(os.path.join(self.config.output_path, self.config.experiment_id, 'train_lora.safetensors')):
sdd = torch.load(os.path.join(self.config.output_path, self.config.experiment_id, 'train_lora.safetensors'), map_location='cpu')
collect_sd = {}
for k, v in sdd.items():
collect_sd[k[7:]] = v
train_lora.load_state_dict(collect_sd, strict=True)
train_norm.to(self.device).train().requires_grad_(True)
if generator_ema is not None:
generator_ema.load_state_dict(load_or_fail(self.config.generator_checkpoint_path))
generator_ema._init_extra_parameter()
pretrained_pth = os.path.join(self.config.output_path, self.config.experiment_id, 'generator.safetensors')
if os.path.exists(pretrained_pth):
generator_ema.load_state_dict(torch.load(pretrained_pth, map_location='cpu'))
generator_ema.eval().requires_grad_(False)
check_nan_inmodel(generator, 'generator')
if self.config.use_ddp and self.config.training:
train_lora = DDP(train_lora, device_ids=[self.device], find_unused_parameters=True)
return self.Models(
effnet=effnet, previewer=previewer, train_norm = train_norm,
generator=generator, generator_ema=generator_ema,
tokenizer=tokenizer, text_model=text_model, image_model=image_model,
train_lora=train_lora
)
def setup_optimizers(self, extras: Extras, models: Models) -> TrainingCore.Optimizers:
params = []
params += list(models.train_lora.module.parameters())
optimizer = optim.AdamW(params, lr=self.config.lr)
return self.Optimizers(generator=optimizer)
def ema_update(self, ema_model, source_model, beta):
for param_src, param_ema in zip(source_model.parameters(), ema_model.parameters()):
param_ema.data.mul_(beta).add_(param_src.data, alpha = 1 - beta)
def sync_ema(self, ema_model):
print('sync ema', torch.distributed.get_world_size())
for param in ema_model.parameters():
torch.distributed.all_reduce(param.data, op=torch.distributed.ReduceOp.SUM)
param.data /= torch.distributed.get_world_size()
def setup_optimizers_backup(self, extras: Extras, models: Models) -> TrainingCore.Optimizers:
optimizer = optim.AdamW(
models.generator.up_blocks.parameters() ,
lr=self.config.lr)
optimizer = self.load_optimizer(optimizer, 'generator_optim',
fsdp_model=models.generator if self.config.use_fsdp else None)
return self.Optimizers(generator=optimizer)
def setup_schedulers(self, extras: Extras, models: Models, optimizers: TrainingCore.Optimizers) -> Schedulers:
scheduler = GradualWarmupScheduler(optimizers.generator, multiplier=1, total_epoch=self.config.warmup_updates)
scheduler.last_epoch = self.info.total_steps
return self.Schedulers(generator=scheduler)
def setup_data(self, extras: Extras) -> WarpCore.Data:
# SETUP DATASET
dataset_path = self.config.webdataset_path
dataset = mydist_dataset(dataset_path, self.config.tmp_prompt, \
torchvision.transforms.ToTensor() if self.config.multi_aspect_ratio is not None \
else extras.transforms)
# SETUP DATALOADER
real_batch_size = self.config.batch_size // (self.world_size * self.config.grad_accum_steps)
sampler = DistributedSampler(dataset, rank=self.process_id, num_replicas = self.world_size, shuffle=True)
dataloader = DataLoader(
dataset, batch_size=real_batch_size, num_workers=4, pin_memory=True,
collate_fn=identity if self.config.multi_aspect_ratio is not None else None,
sampler = sampler
)
if self.is_main_node:
print(f"Training with batch size {self.config.batch_size} ({real_batch_size}/GPU)")
if self.config.multi_aspect_ratio is not None:
aspect_ratios = [float(Fraction(f)) for f in self.config.multi_aspect_ratio]
dataloader_iterator = Bucketeer(dataloader, density=[ss*ss for ss in self.config.image_size] , factor=32,
ratios=aspect_ratios, p_random_ratio=self.config.bucketeer_random_ratio,
interpolate_nearest=False) # , use_smartcrop=True)
else:
dataloader_iterator = iter(dataloader)
return self.Data(dataset=dataset, dataloader=dataloader, iterator=dataloader_iterator, sampler=sampler)
def setup_ddp(self, experiment_id, single_gpu=False, rank=0):
if not single_gpu:
local_rank = rank
process_id = rank
world_size = get_world_size()
self.process_id = process_id
self.is_main_node = process_id == 0
self.device = torch.device(local_rank)
self.world_size = world_size
os.environ['MASTER_ADDR'] = 'localhost'
os.environ['MASTER_PORT'] = '14443'
torch.cuda.set_device(local_rank)
init_process_group(
backend="nccl",
rank=local_rank,
world_size=world_size,
# init_method=init_method,
)
print(f"[GPU {process_id}] READY")
else:
self.is_main_node = rank == 0
self.process_id = rank
self.device = torch.device('cuda:0')
self.world_size = 1
print("Running in single thread, DDP not enabled.")
# Training loop --------------------------------
def get_target_lr_size(self, ratio, std_size=24):
w, h = int(std_size / math.sqrt(ratio)), int(std_size * math.sqrt(ratio))
return (h * 32 , w * 32)
def forward_pass(self, data: WarpCore.Data, extras: Extras, models: Models):
batch = data
ratio = batch['images'].shape[-2] / batch['images'].shape[-1]
shape_lr = self.get_target_lr_size(ratio)
with torch.no_grad():
conditions = self.get_conditions(batch, models, extras)
latents = self.encode_latents(batch, models, extras)
latents_lr = self.encode_latents(batch, models, extras,target_size=shape_lr)
flag_lr = random.random() < 0.5 or self.info.iter <5000
if flag_lr:
noised, noise, target, logSNR, noise_cond, loss_weight = extras.gdf.diffuse(latents_lr, shift=1, loss_shift=1)
else:
noised, noise, target, logSNR, noise_cond, loss_weight = extras.gdf.diffuse(latents, shift=1, loss_shift=1)
if not flag_lr:
noised_lr, noise_lr, target_lr, logSNR_lr, noise_cond_lr, loss_weight_lr = \
extras.gdf.diffuse(latents_lr, shift=1, loss_shift=1, t=torch.ones(latents.shape[0]).to(latents.device)*0.05, )
with torch.cuda.amp.autocast(dtype=torch.bfloat16):
if not flag_lr:
with torch.no_grad():
_, lr_enc_guide, lr_dec_guide = models.generator(noised_lr, noise_cond_lr, reuire_f=True, **conditions)
pred = models.generator(noised, noise_cond, reuire_f=False, lr_guide=(lr_enc_guide, lr_dec_guide) if not flag_lr else None , **conditions)
loss = nn.functional.mse_loss(pred, target, reduction='none').mean(dim=[1, 2, 3])
loss_adjusted = (loss * loss_weight ).mean() / self.config.grad_accum_steps
if isinstance(extras.gdf.loss_weight, AdaptiveLossWeight):
extras.gdf.loss_weight.update_buckets(logSNR, loss)
return loss, loss_adjusted
def backward_pass(self, update, loss_adjusted, models: Models, optimizers: TrainingCore.Optimizers, schedulers: Schedulers):
if update:
torch.distributed.barrier()
loss_adjusted.backward()
grad_norm = nn.utils.clip_grad_norm_(models.train_lora.module.parameters(), 1.0)
optimizers_dict = optimizers.to_dict()
for k in optimizers_dict:
if k != 'training':
optimizers_dict[k].step()
schedulers_dict = schedulers.to_dict()
for k in schedulers_dict:
if k != 'training':
schedulers_dict[k].step()
for k in optimizers_dict:
if k != 'training':
optimizers_dict[k].zero_grad(set_to_none=True)
self.info.total_steps += 1
else:
loss_adjusted.backward()
grad_norm = torch.tensor(0.0).to(self.device)
return grad_norm
def models_to_save(self):
return ['generator', 'generator_ema', 'trans_inr', 'trans_inr_ema']
def encode_latents(self, batch: dict, models: Models, extras: Extras, target_size=None) -> torch.Tensor:
images = batch['images'].to(self.device)
if target_size is not None:
images = F.interpolate(images, target_size)
return models.effnet(extras.effnet_preprocess(images))
def decode_latents(self, latents: torch.Tensor, batch: dict, models: Models, extras: Extras) -> torch.Tensor:
return models.previewer(latents)
def __init__(self, rank=0, config_file_path=None, config_dict=None, device="cpu", training=True, world_size=1, ):
self.is_main_node = (rank == 0)
self.config: self.Config = self.setup_config(config_file_path, config_dict, training)
self.setup_ddp(self.config.experiment_id, single_gpu=world_size <= 1, rank=rank)
self.info: self.Info = self.setup_info()
print('in line 292', self.config.experiment_id, rank, world_size <= 1)
p = [i for i in range( 2 * 768 // 32)]
p = [num / sum(p) for num in p]
self.rand_pro = p
self.res_list = [o for o in range(800, 2336, 32)]
def __call__(self, single_gpu=False):
if self.config.allow_tf32:
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True
if self.is_main_node:
print()
print("**STARTIG JOB WITH CONFIG:**")
print(yaml.dump(self.config.to_dict(), default_flow_style=False))
print("------------------------------------")
print()
print("**INFO:**")
print(yaml.dump(vars(self.info), default_flow_style=False))
print("------------------------------------")
print()
print('in line 308', self.is_main_node, self.is_main_node, self.process_id, self.device )
# SETUP STUFF
extras = self.setup_extras_pre()
assert extras is not None, "setup_extras_pre() must return a DTO"
data = self.setup_data(extras)
assert data is not None, "setup_data() must return a DTO"
if self.is_main_node:
print("**DATA:**")
print(yaml.dump({k:type(v).__name__ for k, v in data.to_dict().items()}, default_flow_style=False))
print("------------------------------------")
print()
models = self.setup_models(extras)
assert models is not None, "setup_models() must return a DTO"
if self.is_main_node:
print("**MODELS:**")
print(yaml.dump({
k:f"{type(v).__name__} - {f'trainable params {sum(p.numel() for p in v.parameters() if p.requires_grad)}' if isinstance(v, nn.Module) else 'Not a nn.Module'}" for k, v in models.to_dict().items()
}, default_flow_style=False))
print("------------------------------------")
print()
optimizers = self.setup_optimizers(extras, models)
assert optimizers is not None, "setup_optimizers() must return a DTO"
if self.is_main_node:
print("**OPTIMIZERS:**")
print(yaml.dump({k:type(v).__name__ for k, v in optimizers.to_dict().items()}, default_flow_style=False))
print("------------------------------------")
print()
schedulers = self.setup_schedulers(extras, models, optimizers)
assert schedulers is not None, "setup_schedulers() must return a DTO"
if self.is_main_node:
print("**SCHEDULERS:**")
print(yaml.dump({k:type(v).__name__ for k, v in schedulers.to_dict().items()}, default_flow_style=False))
print("------------------------------------")
print()
post_extras =self.setup_extras_post(extras, models, optimizers, schedulers)
assert post_extras is not None, "setup_extras_post() must return a DTO"
extras = self.Extras.from_dict({ **extras.to_dict(),**post_extras.to_dict() })
if self.is_main_node:
print("**EXTRAS:**")
print(yaml.dump({k:f"{v}" for k, v in extras.to_dict().items()}, default_flow_style=False))
print("------------------------------------")
print()
# -------
# TRAIN
if self.is_main_node:
print("**TRAINING STARTING...**")
self.train(data, extras, models, optimizers, schedulers)
if single_gpu is False:
barrier()
destroy_process_group()
if self.is_main_node:
print()
print("------------------------------------")
print()
print("**TRAINING COMPLETE**")
if self.config.wandb_project is not None:
wandb.alert(title=f"Training {self.info.wandb_run_id} finished", text=f"Training {self.info.wandb_run_id} finished")
def train(self, data: WarpCore.Data, extras: WarpCore.Extras, models: Models, optimizers: TrainingCore.Optimizers,
schedulers: WarpCore.Schedulers):
start_iter = self.info.iter + 1
max_iters = self.config.updates * self.config.grad_accum_steps
if self.is_main_node:
print(f"STARTING AT STEP: {start_iter}/{max_iters}")
if self.is_main_node:
create_folder_if_necessary(f'{self.config.output_path}/{self.config.experiment_id}/')
if 'generator' in self.models_to_save():
models.generator.train()
iter_cnt = 0
epoch_cnt = 0
models.train_norm.train()
while True:
epoch_cnt += 1
if self.world_size > 1:
data.sampler.set_epoch(epoch_cnt)
for ggg in range(len(data.dataloader)):
iter_cnt += 1
# FORWARD PASS
loss, loss_adjusted = self.forward_pass(next(data.iterator), extras, models)
# # BACKWARD PASS
grad_norm = self.backward_pass(
iter_cnt % self.config.grad_accum_steps == 0 or iter_cnt == max_iters, loss_adjusted,
models, optimizers, schedulers
)
self.info.iter = iter_cnt
self.info.ema_loss = loss.mean().item() if self.info.ema_loss is None else self.info.ema_loss * 0.99 + loss.mean().item() * 0.01
if self.is_main_node and np.isnan(loss.mean().item()) or np.isnan(grad_norm.item()):
print(f"gggg NaN value encountered in training run {self.info.wandb_run_id}", \
f"Loss {loss.mean().item()} - Grad Norm {grad_norm.item()}. Run {self.info.wandb_run_id}")
if self.is_main_node:
logs = {
'loss': self.info.ema_loss,
'backward_loss': loss_adjusted.mean().item(),
'ema_loss': self.info.ema_loss,
'raw_ori_loss': loss.mean().item(),
'grad_norm': grad_norm.item(),
'lr': optimizers.generator.param_groups[0]['lr'] if optimizers.generator is not None else 0,
'total_steps': self.info.total_steps,
}
print(iter_cnt, max_iters, logs, epoch_cnt, )
if iter_cnt == 1 or iter_cnt % (self.config.save_every ) == 0 or iter_cnt == max_iters:
if np.isnan(loss.mean().item()):
if self.is_main_node and self.config.wandb_project is not None:
print(f"NaN value encountered in training run {self.info.wandb_run_id}", \
f"Loss {loss.mean().item()} - Grad Norm {grad_norm.item()}. Run {self.info.wandb_run_id}")
else:
if isinstance(extras.gdf.loss_weight, AdaptiveLossWeight):
self.info.adaptive_loss = {
'bucket_ranges': extras.gdf.loss_weight.bucket_ranges.tolist(),
'bucket_losses': extras.gdf.loss_weight.bucket_losses.tolist(),
}
if self.is_main_node and iter_cnt % (self.config.save_every * self.config.grad_accum_steps) == 0:
print('save model', iter_cnt, iter_cnt % (self.config.save_every * self.config.grad_accum_steps), self.config.save_every, self.config.grad_accum_steps )
torch.save(models.train_lora.state_dict(), \
f'{self.config.output_path}/{self.config.experiment_id}/train_lora.safetensors')
torch.save(models.train_lora.state_dict(), \
f'{self.config.output_path}/{self.config.experiment_id}/train_lora_{iter_cnt}.safetensors')
if iter_cnt == 1 or iter_cnt % (self.config.save_every* self.config.grad_accum_steps) == 0 or iter_cnt == max_iters:
if self.is_main_node:
self.sample(models, data, extras)
if False:
param_changes = {name: (param - initial_params[name]).norm().item() for name, param in models.train_norm.named_parameters()}
threshold = sorted(param_changes.values(), reverse=True)[int(len(param_changes) * 0.1)] # top 10%
important_params = [name for name, change in param_changes.items() if change > threshold]
print(important_params, threshold, len(param_changes), self.process_id)
json.dump(important_params, open(f'{self.config.output_path}/{self.config.experiment_id}/param.json', 'w'), indent=4)
if self.info.iter >= max_iters:
break
def sample(self, models: Models, data: WarpCore.Data, extras: Extras):
models.generator.eval()
models.train_norm.eval()
with torch.no_grad():
batch = next(data.iterator)
ratio = batch['images'].shape[-2] / batch['images'].shape[-1]
shape_lr = self.get_target_lr_size(ratio)
conditions = self.get_conditions(batch, models, extras, is_eval=True, is_unconditional=False, eval_image_embeds=False)
unconditions = self.get_conditions(batch, models, extras, is_eval=True, is_unconditional=True, eval_image_embeds=False)
latents = self.encode_latents(batch, models, extras)
latents_lr = self.encode_latents(batch, models, extras, target_size = shape_lr)
if self.is_main_node:
with torch.cuda.amp.autocast(dtype=torch.bfloat16):
*_, (sampled, _, _, sampled_lr) = extras.gdf.sample(
models.generator, conditions,
latents.shape, latents_lr.shape,
unconditions, device=self.device, **extras.sampling_configs
)
sampled_ema = sampled
sampled_ema_lr = sampled_lr
if self.is_main_node:
print('sampling results hr latent shape ', latents.shape, 'lr latent shape', latents_lr.shape, )
noised_images = torch.cat(
[self.decode_latents(latents[i:i + 1].float(), batch, models, extras) for i in range(len(latents))], dim=0)
sampled_images = torch.cat(
[self.decode_latents(sampled[i:i + 1].float(), batch, models, extras) for i in range(len(sampled))], dim=0)
sampled_images_ema = torch.cat(
[self.decode_latents(sampled_ema[i:i + 1].float(), batch, models, extras) for i in range(len(sampled_ema))],
dim=0)
noised_images_lr = torch.cat(
[self.decode_latents(latents_lr[i:i + 1].float(), batch, models, extras) for i in range(len(latents_lr))], dim=0)
sampled_images_lr = torch.cat(
[self.decode_latents(sampled_lr[i:i + 1].float(), batch, models, extras) for i in range(len(sampled_lr))], dim=0)
sampled_images_ema_lr = torch.cat(
[self.decode_latents(sampled_ema_lr[i:i + 1].float(), batch, models, extras) for i in range(len(sampled_ema_lr))],
dim=0)
images = batch['images']
if images.size(-1) != noised_images.size(-1) or images.size(-2) != noised_images.size(-2):
images = nn.functional.interpolate(images, size=noised_images.shape[-2:], mode='bicubic')
images_lr = nn.functional.interpolate(images, size=noised_images_lr.shape[-2:], mode='bicubic')
collage_img = torch.cat([
torch.cat([i for i in images.cpu()], dim=-1),
torch.cat([i for i in noised_images.cpu()], dim=-1),
torch.cat([i for i in sampled_images.cpu()], dim=-1),
torch.cat([i for i in sampled_images_ema.cpu()], dim=-1),
], dim=-2)
collage_img_lr = torch.cat([
torch.cat([i for i in images_lr.cpu()], dim=-1),
torch.cat([i for i in noised_images_lr.cpu()], dim=-1),
torch.cat([i for i in sampled_images_lr.cpu()], dim=-1),
torch.cat([i for i in sampled_images_ema_lr.cpu()], dim=-1),
], dim=-2)
torchvision.utils.save_image(collage_img, f'{self.config.output_path}/{self.config.experiment_id}/{self.info.total_steps:06d}.jpg')
torchvision.utils.save_image(collage_img_lr, f'{self.config.output_path}/{self.config.experiment_id}/{self.info.total_steps:06d}_lr.jpg')
captions = batch['captions']
if self.config.wandb_project is not None:
log_data = [
[captions[i]] + [wandb.Image(sampled_images[i])] + [wandb.Image(sampled_images_ema[i])] + [
wandb.Image(images[i])] for i in range(len(images))]
log_table = wandb.Table(data=log_data, columns=["Captions", "Sampled", "Sampled EMA", "Orig"])
wandb.log({"Log": log_table})
if isinstance(extras.gdf.loss_weight, AdaptiveLossWeight):
plt.plot(extras.gdf.loss_weight.bucket_ranges, extras.gdf.loss_weight.bucket_losses[:-1])
plt.ylabel('Raw Loss')
plt.ylabel('LogSNR')
wandb.log({"Loss/LogSRN": plt})
models.generator.train()
models.train_norm.train()
print('finish sampling')
def sample_fortest(self, models: Models, extras: Extras, hr_shape, lr_shape, batch, eval_image_embeds=False):
models.generator.eval()
models.trans_inr.eval()
with torch.no_grad():
if self.is_main_node:
conditions = self.get_conditions(batch, models, extras, is_eval=True, is_unconditional=False, eval_image_embeds=eval_image_embeds)
unconditions = self.get_conditions(batch, models, extras, is_eval=True, is_unconditional=True, eval_image_embeds=False)
with torch.cuda.amp.autocast(dtype=torch.bfloat16):
*_, (sampled, _, _, sampled_lr) = extras.gdf.sample(
models.generator, conditions,
hr_shape, lr_shape,
unconditions, device=self.device, **extras.sampling_configs
)
if models.generator_ema is not None:
*_, (sampled_ema, _, _, sampled_ema_lr) = extras.gdf.sample(
models.generator_ema, conditions,
latents.shape, latents_lr.shape,
unconditions, device=self.device, **extras.sampling_configs
)
else:
sampled_ema = sampled
sampled_ema_lr = sampled_lr
return sampled, sampled_lr
def main_worker(rank, cfg):
print("Launching Script in main worker")
warpcore = WurstCore(
config_file_path=cfg, rank=rank, world_size = get_world_size()
)
# core.fsdp_defaults['sharding_strategy'] = ShardingStrategy.NO_SHARD
# RUN TRAINING
warpcore(get_world_size()==1)
if __name__ == '__main__':
if get_master_ip() == "127.0.0.1":
mp.spawn(main_worker, nprocs=get_world_size(), args=(sys.argv[1] if len(sys.argv) > 1 else None, ))
else:
main_worker(0, sys.argv[1] if len(sys.argv) > 1 else None, )