diff options
33 files changed, 1149 insertions, 731 deletions
@@ -84,26 +84,7 @@ Check the [custom scripts](https://github.com/AUTOMATIC1111/stable-diffusion-web - API
- Support for dedicated [inpainting model](https://github.com/runwayml/stable-diffusion#inpainting-with-stable-diffusion) by RunwayML.
- via extension: [Aesthetic Gradients](https://github.com/AUTOMATIC1111/stable-diffusion-webui-aesthetic-gradients), a way to generate images with a specific aesthetic by using clip images embds (implementation of [https://github.com/vicgalle/stable-diffusion-aesthetic-gradients](https://github.com/vicgalle/stable-diffusion-aesthetic-gradients))
-
-## Where are Aesthetic Gradients?!?!
-Aesthetic Gradients are now an extension. You can install it using git:
-
-```commandline
-git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui-aesthetic-gradients extensions/aesthetic-gradients
-```
-
-After running this command, make sure that you have `aesthetic-gradients` dir in webui's `extensions` directory and restart
-the UI. The interface for Aesthetic Gradients should appear exactly the same as it was.
-
-## Where is History/Image browser?!?!
-Image browser is now an extension. You can install it using git:
-
-```commandline
-git clone https://github.com/yfszzx/stable-diffusion-webui-images-browser extensions/images-browser
-```
-
-After running this command, make sure that you have `images-browser` dir in webui's `extensions` directory and restart
-the UI. The interface for Image browser should appear exactly the same as it was.
+- [Stable Diffusion 2.0](https://github.com/Stability-AI/stablediffusion) support - see [wiki](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#stable-diffusion-20) for instructions
## Installation and Running
Make sure the required [dependencies](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Dependencies) are met and follow the instructions available for both [NVidia](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUs) (recommended) and [AMD](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs) GPUs.
diff --git a/javascript/hints.js b/javascript/hints.js index 623bc25c..ac417ff6 100644 --- a/javascript/hints.js +++ b/javascript/hints.js @@ -62,8 +62,8 @@ titles = { "Interrogate": "Reconstruct prompt from existing image and put it into the prompt field.", - "Images filename pattern": "Use following tags to define how filenames for images are chosen: [steps], [cfg], [prompt], [prompt_no_styles], [prompt_spaces], [width], [height], [styles], [sampler], [seed], [model_hash], [prompt_words], [date], [datetime], [datetime<Format>], [datetime<Format><Time Zone>], [job_timestamp]; leave empty for default.", - "Directory name pattern": "Use following tags to define how subdirectories for images and grids are chosen: [steps], [cfg], [prompt], [prompt_no_styles], [prompt_spaces], [width], [height], [styles], [sampler], [seed], [model_hash], [prompt_words], [date], [datetime], [datetime<Format>], [datetime<Format><Time Zone>], [job_timestamp]; leave empty for default.", + "Images filename pattern": "Use following tags to define how filenames for images are chosen: [steps], [cfg], [prompt], [prompt_no_styles], [prompt_spaces], [width], [height], [styles], [sampler], [seed], [model_hash], [model_name], [prompt_words], [date], [datetime], [datetime<Format>], [datetime<Format><Time Zone>], [job_timestamp]; leave empty for default.", + "Directory name pattern": "Use following tags to define how subdirectories for images and grids are chosen: [steps], [cfg], [prompt], [prompt_no_styles], [prompt_spaces], [width], [height], [styles], [sampler], [seed], [model_hash], [model_name], [prompt_words], [date], [datetime], [datetime<Format>], [datetime<Format><Time Zone>], [job_timestamp]; leave empty for default.", "Max prompt words": "Set the maximum number of words to be used in the [prompt_words] option; ATTENTION: If the words are too long, they may exceed the maximum length of the file path that the system can handle", "Loopback": "Process an image, use it as an input, repeat.", diff --git a/javascript/ui.js b/javascript/ui.js index 95cfd106..2ca66d79 100644 --- a/javascript/ui.js +++ b/javascript/ui.js @@ -8,8 +8,8 @@ function set_theme(theme){ } function selected_gallery_index(){ - var buttons = gradioApp().querySelectorAll('[style="display: block;"].tabitem .gallery-item') - var button = gradioApp().querySelector('[style="display: block;"].tabitem .gallery-item.\\!ring-2') + var buttons = gradioApp().querySelectorAll('[style="display: block;"].tabitem div[id$=_gallery] .gallery-item') + var button = gradioApp().querySelector('[style="display: block;"].tabitem div[id$=_gallery] .gallery-item.\\!ring-2') var result = -1 buttons.forEach(function(v, i){ if(v==button) { result = i } }) @@ -134,18 +134,19 @@ def prepare_enviroment(): gfpgan_package = os.environ.get('GFPGAN_PACKAGE', "git+https://github.com/TencentARC/GFPGAN.git@8d2447a2d918f8eba5a4a01463fd48e45126a379")
clip_package = os.environ.get('CLIP_PACKAGE', "git+https://github.com/openai/CLIP.git@d50d76daa670286dd6cacf3bcd80b5e4823fc8e1")
+ openclip_package = os.environ.get('OPENCLIP_PACKAGE', "git+https://github.com/mlfoundations/open_clip.git@bb6e834e9c70d9c27d0dc3ecedeebeaeb1ffad6b")
xformers_windows_package = os.environ.get('XFORMERS_WINDOWS_PACKAGE', 'https://github.com/C43H66N12O12S2/stable-diffusion-webui/releases/download/f/xformers-0.0.14.dev0-cp310-cp310-win_amd64.whl')
- stable_diffusion_repo = os.environ.get('STABLE_DIFFUSION_REPO', "https://github.com/CompVis/stable-diffusion.git")
+ stable_diffusion_repo = os.environ.get('STABLE_DIFFUSION_REPO', "https://github.com/Stability-AI/stablediffusion.git")
taming_transformers_repo = os.environ.get('TAMING_TRANSFORMERS_REPO', "https://github.com/CompVis/taming-transformers.git")
k_diffusion_repo = os.environ.get('K_DIFFUSION_REPO', 'https://github.com/crowsonkb/k-diffusion.git')
codeformer_repo = os.environ.get('CODEFORMER_REPO', 'https://github.com/sczhou/CodeFormer.git')
blip_repo = os.environ.get('BLIP_REPO', 'https://github.com/salesforce/BLIP.git')
- stable_diffusion_commit_hash = os.environ.get('STABLE_DIFFUSION_COMMIT_HASH', "69ae4b35e0a0f6ee1af8bb9a5d0016ccb27e36dc")
+ stable_diffusion_commit_hash = os.environ.get('STABLE_DIFFUSION_COMMIT_HASH', "47b6b607fdd31875c9279cd2f4f16b92e4ea958e")
taming_transformers_commit_hash = os.environ.get('TAMING_TRANSFORMERS_COMMIT_HASH', "24268930bf1dce879235a7fddd0b2355b84d7ea6")
- k_diffusion_commit_hash = os.environ.get('K_DIFFUSION_COMMIT_HASH', "60e5042ca0da89c14d1dd59d73883280f8fce991")
+ k_diffusion_commit_hash = os.environ.get('K_DIFFUSION_COMMIT_HASH', "5b3af030dd83e0297272d861c19477735d0317ec")
codeformer_commit_hash = os.environ.get('CODEFORMER_COMMIT_HASH', "c5b4593074ba6214284d6acd5f1719b6c5d739af")
blip_commit_hash = os.environ.get('BLIP_COMMIT_HASH', "48211a1594f1321b00f14c9f7a5b4813144b2fb9")
@@ -179,6 +180,9 @@ def prepare_enviroment(): if not is_installed("clip"):
run_pip(f"install {clip_package}", "clip")
+ if not is_installed("open_clip"):
+ run_pip(f"install {openclip_package}", "open_clip")
+
if (not is_installed("xformers") or reinstall_xformers) and xformers:
if platform.system() == "Windows":
if platform.python_version().startswith("3.10"):
@@ -196,7 +200,7 @@ def prepare_enviroment(): os.makedirs(dir_repos, exist_ok=True)
- git_clone(stable_diffusion_repo, repo_dir('stable-diffusion'), "Stable Diffusion", stable_diffusion_commit_hash)
+ git_clone(stable_diffusion_repo, repo_dir('stable-diffusion-stability-ai'), "Stable Diffusion", stable_diffusion_commit_hash)
git_clone(taming_transformers_repo, repo_dir('taming-transformers'), "Taming Transformers", taming_transformers_commit_hash)
git_clone(k_diffusion_repo, repo_dir('k-diffusion'), "K-diffusion", k_diffusion_commit_hash)
git_clone(codeformer_repo, repo_dir('CodeFormer'), "CodeFormer", codeformer_commit_hash)
diff --git a/modules/codeformer_model.py b/modules/codeformer_model.py index e6d9fa4f..ab40d842 100644 --- a/modules/codeformer_model.py +++ b/modules/codeformer_model.py @@ -36,6 +36,7 @@ def setup_model(dirname): from basicsr.utils.download_util import load_file_from_url
from basicsr.utils import imwrite, img2tensor, tensor2img
from facelib.utils.face_restoration_helper import FaceRestoreHelper
+ from facelib.detection.retinaface import retinaface
from modules.shared import cmd_opts
net_class = CodeFormer
@@ -65,6 +66,8 @@ def setup_model(dirname): net.load_state_dict(checkpoint)
net.eval()
+ if hasattr(retinaface, 'device'):
+ retinaface.device = devices.device_codeformer
face_helper = FaceRestoreHelper(1, face_size=512, crop_ratio=(1, 1), det_model='retinaface_resnet50', save_ext='png', use_parse=True, device=devices.device_codeformer)
self.net = net
diff --git a/modules/devices.py b/modules/devices.py index 67165bf6..f00079c6 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -2,9 +2,10 @@ import sys, os, shlex import contextlib import torch from modules import errors +from packaging import version -# has_mps is only available in nightly pytorch (for now) and MasOS 12.3+. +# has_mps is only available in nightly pytorch (for now) and macOS 12.3+. # check `getattr` and try it for compatibility def has_mps() -> bool: if not getattr(torch, 'has_mps', False): @@ -24,17 +25,18 @@ def extract_device_id(args, name): return None -def get_optimal_device(): - if torch.cuda.is_available(): - from modules import shared +def get_cuda_device_string(): + from modules import shared - device_id = shared.cmd_opts.device_id + if shared.cmd_opts.device_id is not None: + return f"cuda:{shared.cmd_opts.device_id}" - if device_id is not None: - cuda_device = f"cuda:{device_id}" - return torch.device(cuda_device) - else: - return torch.device("cuda") + return "cuda" + + +def get_optimal_device(): + if torch.cuda.is_available(): + return torch.device(get_cuda_device_string()) if has_mps(): return torch.device("mps") @@ -44,8 +46,9 @@ def get_optimal_device(): def torch_gc(): if torch.cuda.is_available(): - torch.cuda.empty_cache() - torch.cuda.ipc_collect() + with torch.cuda.device(get_cuda_device_string()): + torch.cuda.empty_cache() + torch.cuda.ipc_collect() def enable_tf32(): @@ -97,9 +100,25 @@ def autocast(disable=False): # MPS workaround for https://github.com/pytorch/pytorch/issues/79383 -def mps_contiguous(input_tensor, device): - return input_tensor.contiguous() if device.type == 'mps' else input_tensor - - -def mps_contiguous_to(input_tensor, device): - return mps_contiguous(input_tensor, device).to(device) +orig_tensor_to = torch.Tensor.to +def tensor_to_fix(self, *args, **kwargs): + if self.device.type != 'mps' and \ + ((len(args) > 0 and isinstance(args[0], torch.device) and args[0].type == 'mps') or \ + (isinstance(kwargs.get('device'), torch.device) and kwargs['device'].type == 'mps')): + self = self.contiguous() + return orig_tensor_to(self, *args, **kwargs) + + +# MPS workaround for https://github.com/pytorch/pytorch/issues/80800 +orig_layer_norm = torch.nn.functional.layer_norm +def layer_norm_fix(*args, **kwargs): + if len(args) > 0 and isinstance(args[0], torch.Tensor) and args[0].device.type == 'mps': + args = list(args) + args[0] = args[0].contiguous() + return orig_layer_norm(*args, **kwargs) + + +# PyTorch 1.13 doesn't need these fixes but unfortunately is slower and has regressions that prevent training from working +if has_mps() and version.parse(torch.__version__) < version.parse("1.13"): + torch.Tensor.to = tensor_to_fix + torch.nn.functional.layer_norm = layer_norm_fix diff --git a/modules/esrgan_model.py b/modules/esrgan_model.py index c61669b4..9a9c38f1 100644 --- a/modules/esrgan_model.py +++ b/modules/esrgan_model.py @@ -199,7 +199,7 @@ def upscale_without_tiling(model, img): img = img[:, :, ::-1]
img = np.ascontiguousarray(np.transpose(img, (2, 0, 1))) / 255
img = torch.from_numpy(img).float()
- img = devices.mps_contiguous_to(img.unsqueeze(0), devices.device_esrgan)
+ img = img.unsqueeze(0).to(devices.device_esrgan)
with torch.no_grad():
output = model(img)
output = output.squeeze().float().cpu().clamp_(0, 1).numpy()
diff --git a/modules/gfpgan_model.py b/modules/gfpgan_model.py index a9452dce..1e2dbc32 100644 --- a/modules/gfpgan_model.py +++ b/modules/gfpgan_model.py @@ -36,7 +36,9 @@ def gfpgann(): else:
print("Unable to load gfpgan model!")
return None
- model = gfpgan_constructor(model_path=model_file, upscale=1, arch='clean', channel_multiplier=2, bg_upsampler=None)
+ if hasattr(facexlib.detection.retinaface, 'device'):
+ facexlib.detection.retinaface.device = devices.device_gfpgan
+ model = gfpgan_constructor(model_path=model_file, upscale=1, arch='clean', channel_multiplier=2, bg_upsampler=None, device=devices.device_gfpgan)
loaded_gfpgan_model = model
return model
diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index fbb87dd1..8466887f 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -38,7 +38,7 @@ class HypernetworkModule(torch.nn.Module): activation_dict.update({cls_name.lower(): cls_obj for cls_name, cls_obj in inspect.getmembers(torch.nn.modules.activation) if inspect.isclass(cls_obj) and cls_obj.__module__ == 'torch.nn.modules.activation'})
def __init__(self, dim, state_dict=None, layer_structure=None, activation_func=None, weight_init='Normal',
- add_layer_norm=False, use_dropout=False, activate_output=False, last_layer_dropout=True):
+ add_layer_norm=False, use_dropout=False, activate_output=False, last_layer_dropout=False):
super().__init__()
assert layer_structure is not None, "layer_structure must not be None"
@@ -154,16 +154,28 @@ class Hypernetwork: HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.weight_init,
self.add_layer_norm, self.use_dropout, self.activate_output, last_layer_dropout=self.last_layer_dropout),
)
+ self.eval_mode()
def weights(self):
res = []
+ for k, layers in self.layers.items():
+ for layer in layers:
+ res += layer.parameters()
+ return res
+ def train_mode(self):
for k, layers in self.layers.items():
for layer in layers:
layer.train()
- res += layer.trainables()
+ for param in layer.parameters():
+ param.requires_grad = True
- return res
+ def eval_mode(self):
+ for k, layers in self.layers.items():
+ for layer in layers:
+ layer.eval()
+ for param in layer.parameters():
+ param.requires_grad = False
def save(self, filename):
state_dict = {}
@@ -367,13 +379,13 @@ def report_statistics(loss_info:dict): -def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
+def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, log_directory, training_width, training_height, steps, shuffle_tags, tag_drop_out, latent_sampling_method, create_image_every, save_hypernetwork_every, template_file, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
# images allows training previews to have infotext. Importing it at the top causes a circular import problem.
from modules import images
save_hypernetwork_every = save_hypernetwork_every or 0
create_image_every = create_image_every or 0
- textual_inversion.validate_train_inputs(hypernetwork_name, learn_rate, batch_size, data_root, template_file, steps, save_hypernetwork_every, create_image_every, log_directory, name="hypernetwork")
+ textual_inversion.validate_train_inputs(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, template_file, steps, save_hypernetwork_every, create_image_every, log_directory, name="hypernetwork")
path = shared.hypernetworks.get(hypernetwork_name, None)
shared.loaded_hypernetwork = Hypernetwork()
@@ -403,32 +415,30 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log hypernetwork = shared.loaded_hypernetwork
checkpoint = sd_models.select_checkpoint()
- ititial_step = hypernetwork.step or 0
- if ititial_step >= steps:
+ initial_step = hypernetwork.step or 0
+ if initial_step >= steps:
shared.state.textinfo = f"Model has already been trained beyond specified max steps"
return hypernetwork, filename
- scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
-
+ scheduler = LearnRateScheduler(learn_rate, steps, initial_step)
+
# dataset loading may take a while, so input validations and early returns should be done before this
shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
- with torch.autocast("cuda"):
- ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size)
+
+ pin_memory = shared.opts.pin_memory
+
+ ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, cond_model=shared.sd_model.cond_stage_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size, gradient_step=gradient_step, shuffle_tags=shuffle_tags, tag_drop_out=tag_drop_out, latent_sampling_method=latent_sampling_method)
+
+ latent_sampling_method = ds.latent_sampling_method
+
+ dl = modules.textual_inversion.dataset.PersonalizedDataLoader(ds, latent_sampling_method=latent_sampling_method, batch_size=ds.batch_size, pin_memory=pin_memory)
if unload:
shared.sd_model.cond_stage_model.to(devices.cpu)
shared.sd_model.first_stage_model.to(devices.cpu)
-
- size = len(ds.indexes)
- loss_dict = defaultdict(lambda : deque(maxlen = 1024))
- losses = torch.zeros((size,))
- previous_mean_losses = [0]
- previous_mean_loss = 0
- print("Mean loss of {} elements".format(size))
weights = hypernetwork.weights()
- for weight in weights:
- weight.requires_grad = True
+ hypernetwork.train_mode()
# Here we use optimizer from saved HN, or we can specify as UI option.
if hypernetwork.optimizer_name in optimizer_dict:
@@ -446,131 +456,156 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log print("Cannot resume from saved optimizer!")
print(e)
+ scaler = torch.cuda.amp.GradScaler()
+
+ batch_size = ds.batch_size
+ gradient_step = ds.gradient_step
+ # n steps = batch_size * gradient_step * n image processed
+ steps_per_epoch = len(ds) // batch_size // gradient_step
+ max_steps_per_epoch = len(ds) // batch_size - (len(ds) // batch_size) % gradient_step
+ loss_step = 0
+ _loss_step = 0 #internal
+ # size = len(ds.indexes)
+ # loss_dict = defaultdict(lambda : deque(maxlen = 1024))
+ # losses = torch.zeros((size,))
+ # previous_mean_losses = [0]
+ # previous_mean_loss = 0
+ # print("Mean loss of {} elements".format(size))
+
steps_without_grad = 0
last_saved_file = "<none>"
last_saved_image = "<none>"
forced_filename = "<none>"
- pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
- for i, entries in pbar:
- hypernetwork.step = i + ititial_step
- if len(loss_dict) > 0:
- previous_mean_losses = [i[-1] for i in loss_dict.values()]
- previous_mean_loss = mean(previous_mean_losses)
-
- scheduler.apply(optimizer, hypernetwork.step)
- if scheduler.finished:
- break
-
- if shared.state.interrupted:
- break
-
- with torch.autocast("cuda"):
- c = stack_conds([entry.cond for entry in entries]).to(devices.device)
- # c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
- x = torch.stack([entry.latent for entry in entries]).to(devices.device)
- loss = shared.sd_model(x, c)[0]
- del x
- del c
-
- losses[hypernetwork.step % losses.shape[0]] = loss.item()
- for entry in entries:
- loss_dict[entry.filename].append(loss.item())
+ pbar = tqdm.tqdm(total=steps - initial_step)
+ try:
+ for i in range((steps-initial_step) * gradient_step):
+ if scheduler.finished:
+ break
+ if shared.state.interrupted:
+ break
+ for j, batch in enumerate(dl):
+ # works as a drop_last=True for gradient accumulation
+ if j == max_steps_per_epoch:
+ break
+ scheduler.apply(optimizer, hypernetwork.step)
+ if scheduler.finished:
+ break
+ if shared.state.interrupted:
+ break
+
+ with torch.autocast("cuda"):
+ x = batch.latent_sample.to(devices.device, non_blocking=pin_memory)
+ if tag_drop_out != 0 or shuffle_tags:
+ shared.sd_model.cond_stage_model.to(devices.device)
+ c = shared.sd_model.cond_stage_model(batch.cond_text).to(devices.device, non_blocking=pin_memory)
+ shared.sd_model.cond_stage_model.to(devices.cpu)
+ else:
+ c = stack_conds(batch.cond).to(devices.device, non_blocking=pin_memory)
+ loss = shared.sd_model(x, c)[0] / gradient_step
+ del x
+ del c
+
+ _loss_step += loss.item()
+ scaler.scale(loss).backward()
+ # go back until we reach gradient accumulation steps
+ if (j + 1) % gradient_step != 0:
+ continue
+ # print(f"grad:{weights[0].grad.detach().cpu().abs().mean().item():.7f}")
+ # scaler.unscale_(optimizer)
+ # print(f"grad:{weights[0].grad.detach().cpu().abs().mean().item():.15f}")
+ # torch.nn.utils.clip_grad_norm_(weights, max_norm=1.0)
+ # print(f"grad:{weights[0].grad.detach().cpu().abs().mean().item():.15f}")
+ scaler.step(optimizer)
+ scaler.update()
+ hypernetwork.step += 1
+ pbar.update()
+ optimizer.zero_grad(set_to_none=True)
+ loss_step = _loss_step
+ _loss_step = 0
+
+ steps_done = hypernetwork.step + 1
- optimizer.zero_grad()
- weights[0].grad = None
- loss.backward()
-
- if weights[0].grad is None:
- steps_without_grad += 1
- else:
- steps_without_grad = 0
- assert steps_without_grad < 10, 'no gradient found for the trained weight after backward() for 10 steps in a row; this is a bug; training cannot continue'
-
- optimizer.step()
-
- steps_done = hypernetwork.step + 1
-
- if torch.isnan(losses[hypernetwork.step % losses.shape[0]]):
- raise RuntimeError("Loss diverged.")
-
- if len(previous_mean_losses) > 1:
- std = stdev(previous_mean_losses)
- else:
- std = 0
- dataset_loss_info = f"dataset loss:{mean(previous_mean_losses):.3f}" + u"\u00B1" + f"({std / (len(previous_mean_losses) ** 0.5):.3f})"
- pbar.set_description(dataset_loss_info)
-
- if hypernetwork_dir is not None and steps_done % save_hypernetwork_every == 0:
- # Before saving, change name to match current checkpoint.
- hypernetwork_name_every = f'{hypernetwork_name}-{steps_done}'
- last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork_name_every}.pt')
- hypernetwork.optimizer_name = optimizer_name
- if shared.opts.save_optimizer_state:
- hypernetwork.optimizer_state_dict = optimizer.state_dict()
- save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, last_saved_file)
- hypernetwork.optimizer_state_dict = None # dereference it after saving, to save memory.
-
- textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), {
- "loss": f"{previous_mean_loss:.7f}",
- "learn_rate": scheduler.learn_rate
- })
-
- if images_dir is not None and steps_done % create_image_every == 0:
- forced_filename = f'{hypernetwork_name}-{steps_done}'
- last_saved_image = os.path.join(images_dir, forced_filename)
-
- optimizer.zero_grad()
- shared.sd_model.cond_stage_model.to(devices.device)
- shared.sd_model.first_stage_model.to(devices.device)
-
- p = processing.StableDiffusionProcessingTxt2Img(
- sd_model=shared.sd_model,
- do_not_save_grid=True,
- do_not_save_samples=True,
- )
-
- if preview_from_txt2img:
- p.prompt = preview_prompt
- p.negative_prompt = preview_negative_prompt
- p.steps = preview_steps
- p.sampler_name = sd_samplers.samplers[preview_sampler_index].name
- p.cfg_scale = preview_cfg_scale
- p.seed = preview_seed
- p.width = preview_width
- p.height = preview_height
- else:
- p.prompt = entries[0].cond_text
- p.steps = 20
-
- preview_text = p.prompt
+ epoch_num = hypernetwork.step // steps_per_epoch
+ epoch_step = hypernetwork.step % steps_per_epoch
+
+ pbar.set_description(f"[Epoch {epoch_num}: {epoch_step+1}/{steps_per_epoch}]loss: {loss_step:.7f}")
+ if hypernetwork_dir is not None and steps_done % save_hypernetwork_every == 0:
+ # Before saving, change name to match current checkpoint.
+ hypernetwork_name_every = f'{hypernetwork_name}-{steps_done}'
+ last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork_name_every}.pt')
+ hypernetwork.optimizer_name = optimizer_name
+ if shared.opts.save_optimizer_state:
+ hypernetwork.optimizer_state_dict = optimizer.state_dict()
+ save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, last_saved_file)
+ hypernetwork.optimizer_state_dict = None # dereference it after saving, to save memory.
+
+ textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, steps_per_epoch, {
+ "loss": f"{loss_step:.7f}",
+ "learn_rate": scheduler.learn_rate
+ })
+
+ if images_dir is not None and steps_done % create_image_every == 0:
+ forced_filename = f'{hypernetwork_name}-{steps_done}'
+ last_saved_image = os.path.join(images_dir, forced_filename)
+ hypernetwork.eval_mode()
+ shared.sd_model.cond_stage_model.to(devices.device)
+ shared.sd_model.first_stage_model.to(devices.device)
+
+ p = processing.StableDiffusionProcessingTxt2Img(
+ sd_model=shared.sd_model,
+ do_not_save_grid=True,
+ do_not_save_samples=True,
+ )
+
+ if preview_from_txt2img:
+ p.prompt = preview_prompt
+ p.negative_prompt = preview_negative_prompt
+ p.steps = preview_steps
+ p.sampler_name = sd_samplers.samplers[preview_sampler_index].name
+ p.cfg_scale = preview_cfg_scale
+ p.seed = preview_seed
+ p.width = preview_width
+ p.height = preview_height
+ else:
+ p.prompt = batch.cond_text[0]
+ p.steps = 20
+ p.width = training_width
+ p.height = training_height
- processed = processing.process_images(p)
- image = processed.images[0] if len(processed.images)>0 else None
+ preview_text = p.prompt
|