From 6021f7a75f7b5208a2be15cda5526028152f922d Mon Sep 17 00:00:00 2001 From: discus0434 Date: Wed, 19 Oct 2022 00:51:36 +0900 Subject: add options to custom hypernetwork layer structure --- modules/hypernetworks/hypernetwork.py | 88 ++++++++++++++++++++++++++--------- 1 file changed, 67 insertions(+), 21 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 4905710e..cadb9911 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -1,52 +1,98 @@ +import csv import datetime import glob import html import os import sys import traceback -import tqdm -import csv +import modules.textual_inversion.dataset import torch - -from ldm.util import default -from modules import devices, shared, processing, sd_models -import torch -from torch import einsum +import tqdm from einops import rearrange, repeat -import modules.textual_inversion.dataset +from ldm.util import default +from modules import devices, processing, sd_models, shared from modules.textual_inversion import textual_inversion from modules.textual_inversion.learn_schedule import LearnRateScheduler +from torch import einsum + + +def parse_layer_structure(dim, state_dict): + i = 0 + res = [1] + while (key := "linear.{}.weight".format(i)) in state_dict: + weight = state_dict[key] + res.append(len(weight) // dim) + i += 1 + return res class HypernetworkModule(torch.nn.Module): multiplier = 1.0 + layer_structure = None + add_layer_norm = False def __init__(self, dim, state_dict=None): super().__init__() + if (state_dict is None or 'linear.0.weight' not in state_dict) and self.layer_structure is None: + layer_structure = (1, 2, 1) + else: + if self.layer_structure is not None: + assert self.layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" + assert self.layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" + layer_structure = self.layer_structure + else: + layer_structure = parse_layer_structure(dim, state_dict) + + linears = [] + for i in range(len(layer_structure) - 1): + linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1]))) + if self.add_layer_norm: + linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1]))) - self.linear1 = torch.nn.Linear(dim, dim * 2) - self.linear2 = torch.nn.Linear(dim * 2, dim) + self.linear = torch.nn.Sequential(*linears) if state_dict is not None: - self.load_state_dict(state_dict, strict=True) + try: + self.load_state_dict(state_dict) + except RuntimeError: + self.try_load_previous(state_dict) else: - - self.linear1.weight.data.normal_(mean=0.0, std=0.01) - self.linear1.bias.data.zero_() - self.linear2.weight.data.normal_(mean=0.0, std=0.01) - self.linear2.bias.data.zero_() + for layer in self.linear: + layer.weight.data.normal_(mean = 0.0, std = 0.01) + layer.bias.data.zero_() self.to(devices.device) + def try_load_previous(self, state_dict): + states = self.state_dict() + states['linear.0.bias'].copy_(state_dict['linear1.bias']) + states['linear.0.weight'].copy_(state_dict['linear1.weight']) + states['linear.1.bias'].copy_(state_dict['linear2.bias']) + states['linear.1.weight'].copy_(state_dict['linear2.weight']) + def forward(self, x): - return x + (self.linear2(self.linear1(x))) * self.multiplier + return x + self.linear(x) * self.multiplier + + def trainables(self): + res = [] + for layer in self.linear: + res += [layer.weight, layer.bias] + return res def apply_strength(value=None): HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength +def apply_layer_structure(value=None): + HypernetworkModule.layer_structure = value if value is not None else shared.opts.sd_hypernetwork_layer_structure + + +def apply_layer_norm(value=None): + HypernetworkModule.add_layer_norm = value if value is not None else shared.opts.sd_hypernetwork_add_layer_norm + + class Hypernetwork: filename = None name = None @@ -68,7 +114,7 @@ class Hypernetwork: for k, layers in self.layers.items(): for layer in layers: layer.train() - res += [layer.linear1.weight, layer.linear1.bias, layer.linear2.weight, layer.linear2.bias] + res += layer.trainables() return res @@ -226,7 +272,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log 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=512, height=512, 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) - + assert ds.length > 1, "Dataset should contain more than 1 images" if unload: shared.sd_model.cond_stage_model.to(devices.cpu) shared.sd_model.first_stage_model.to(devices.cpu) @@ -261,7 +307,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log 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) + 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 @@ -283,7 +329,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), { "loss": f"{mean_loss:.7f}", - "learn_rate": scheduler.learn_rate + "learn_rate": f"{scheduler.learn_rate:.7f}" }) if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0: -- cgit v1.2.3 From a5611ea5026bd8e12d8e84023384c369d0511dda Mon Sep 17 00:00:00 2001 From: discus0434 Date: Wed, 19 Oct 2022 01:00:01 +0900 Subject: update --- .gitignore | 1 - modules/hypernetworks/hypernetwork.py | 14 ++++++++------ 2 files changed, 8 insertions(+), 7 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/.gitignore b/.gitignore index 4794865c..69785b3e 100644 --- a/.gitignore +++ b/.gitignore @@ -27,4 +27,3 @@ __pycache__ notification.mp3 /SwinIR /textual_inversion -/hypernetwork diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index cadb9911..c5835bce 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -1,20 +1,22 @@ -import csv import datetime import glob import html import os import sys import traceback +import tqdm +import csv -import modules.textual_inversion.dataset import torch -import tqdm -from einops import rearrange, repeat + from ldm.util import default -from modules import devices, processing, sd_models, shared +from modules import devices, shared, processing, sd_models +import torch +from torch import einsum +from einops import rearrange, repeat +import modules.textual_inversion.dataset from modules.textual_inversion import textual_inversion from modules.textual_inversion.learn_schedule import LearnRateScheduler -from torch import einsum def parse_layer_structure(dim, state_dict): -- cgit v1.2.3 From e40ba281f1b419cf99552962ea01d87d699840a5 Mon Sep 17 00:00:00 2001 From: discus0434 Date: Wed, 19 Oct 2022 01:03:58 +0900 Subject: update --- modules/hypernetworks/hypernetwork.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index c5835bce..082165f4 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -309,7 +309,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log 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) + # 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 @@ -331,7 +331,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), { "loss": f"{mean_loss:.7f}", - "learn_rate": f"{scheduler.learn_rate:.7f}" + "learn_rate": scheduler.learn_rate }) if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0: -- cgit v1.2.3 From da72becb13e4b750fbcb3d158c3f843311ef9938 Mon Sep 17 00:00:00 2001 From: Silent <16026653+s-ilent@users.noreply.github.com> Date: Wed, 19 Oct 2022 16:14:33 +1030 Subject: Use training width/height when training hypernetworks. --- modules/hypernetworks/hypernetwork.py | 4 ++-- modules/ui.py | 2 ++ 2 files changed, 4 insertions(+), 2 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 4905710e..b8695fc1 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -196,7 +196,7 @@ def stack_conds(conds): return torch.stack(conds) -def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log_directory, 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, 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): assert hypernetwork_name, 'hypernetwork not selected' path = shared.hypernetworks.get(hypernetwork_name, None) @@ -225,7 +225,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log 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=512, height=512, 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) + 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) if unload: shared.sd_model.cond_stage_model.to(devices.cpu) diff --git a/modules/ui.py b/modules/ui.py index fb6eb5a0..ca46343f 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -1341,6 +1341,8 @@ def create_ui(wrap_gradio_gpu_call): batch_size, dataset_directory, log_directory, + training_width, + training_height, steps, create_image_every, save_embedding_every, -- cgit v1.2.3 From 42fbda83bb9830af18187fddb50c1bedd01da502 Mon Sep 17 00:00:00 2001 From: discus0434 Date: Wed, 19 Oct 2022 14:30:33 +0000 Subject: layer options moves into create hnet ui --- modules/hypernetworks/hypernetwork.py | 64 +++++++++++++++++------------------ modules/hypernetworks/ui.py | 9 +++-- modules/shared.py | 2 -- modules/ui.py | 8 +++-- webui.py | 8 ++--- 5 files changed, 48 insertions(+), 43 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 583ada31..7d519cd9 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -19,37 +19,21 @@ from modules.textual_inversion import textual_inversion from modules.textual_inversion.learn_schedule import LearnRateScheduler -def parse_layer_structure(dim, state_dict): - i = 0 - res = [1] - while (key := "linear.{}.weight".format(i)) in state_dict: - weight = state_dict[key] - res.append(len(weight) // dim) - i += 1 - return res - - class HypernetworkModule(torch.nn.Module): multiplier = 1.0 - layer_structure = None - add_layer_norm = False - def __init__(self, dim, state_dict=None): + def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False): super().__init__() - if (state_dict is None or 'linear.0.weight' not in state_dict) and self.layer_structure is None: - layer_structure = (1, 2, 1) + if layer_structure is not None: + assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" + assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" else: - if self.layer_structure is not None: - assert self.layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" - assert self.layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" - layer_structure = self.layer_structure - else: - layer_structure = parse_layer_structure(dim, state_dict) + layer_structure = parse_layer_structure(dim, state_dict) linears = [] for i in range(len(layer_structure) - 1): linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1]))) - if self.add_layer_norm: + if add_layer_norm: linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1]))) self.linear = torch.nn.Sequential(*linears) @@ -77,38 +61,47 @@ class HypernetworkModule(torch.nn.Module): return x + self.linear(x) * self.multiplier def trainables(self): - res = [] + layer_structure = [] for layer in self.linear: - res += [layer.weight, layer.bias] - return res + layer_structure += [layer.weight, layer.bias] + return layer_structure def apply_strength(value=None): HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength -def apply_layer_structure(value=None): - HypernetworkModule.layer_structure = value if value is not None else shared.opts.sd_hypernetwork_layer_structure +def parse_layer_structure(dim, state_dict): + i = 0 + layer_structure = [1] + while (key := "linear.{}.weight".format(i)) in state_dict: + weight = state_dict[key] + layer_structure.append(len(weight) // dim) + i += 1 -def apply_layer_norm(value=None): - HypernetworkModule.add_layer_norm = value if value is not None else shared.opts.sd_hypernetwork_add_layer_norm + return layer_structure class Hypernetwork: filename = None name = None - def __init__(self, name=None, enable_sizes=None): + def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False): self.filename = None self.name = name self.layers = {} self.step = 0 self.sd_checkpoint = None self.sd_checkpoint_name = None + self.layer_structure = layer_structure + self.add_layer_norm = add_layer_norm for size in enable_sizes or []: - self.layers[size] = (HypernetworkModule(size), HypernetworkModule(size)) + self.layers[size] = ( + HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm), + HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm), + ) def weights(self): res = [] @@ -128,6 +121,8 @@ class Hypernetwork: state_dict['step'] = self.step state_dict['name'] = self.name + state_dict['layer_structure'] = self.layer_structure + state_dict['is_layer_norm'] = self.add_layer_norm state_dict['sd_checkpoint'] = self.sd_checkpoint state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name @@ -142,10 +137,15 @@ class Hypernetwork: for size, sd in state_dict.items(): if type(size) == int: - self.layers[size] = (HypernetworkModule(size, sd[0]), HypernetworkModule(size, sd[1])) + self.layers[size] = ( + HypernetworkModule(size, sd[0], state_dict["layer_structure"], state_dict["is_layer_norm"]), + HypernetworkModule(size, sd[1], state_dict["layer_structure"], state_dict["is_layer_norm"]), + ) self.name = state_dict.get('name', self.name) self.step = state_dict.get('step', 0) + self.layer_structure = state_dict.get('layer_structure', None) + self.add_layer_norm = state_dict.get('is_layer_norm', False) self.sd_checkpoint = state_dict.get('sd_checkpoint', None) self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None) diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py index dfa599af..7e8ea95e 100644 --- a/modules/hypernetworks/ui.py +++ b/modules/hypernetworks/ui.py @@ -9,11 +9,16 @@ from modules import sd_hijack, shared, devices from modules.hypernetworks import hypernetwork -def create_hypernetwork(name, enable_sizes): +def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False): fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt") assert not os.path.exists(fn), f"file {fn} already exists" - hypernet = modules.hypernetworks.hypernetwork.Hypernetwork(name=name, enable_sizes=[int(x) for x in enable_sizes]) + hypernet = modules.hypernetworks.hypernetwork.Hypernetwork( + name=name, + enable_sizes=[int(x) for x in enable_sizes], + layer_structure=layer_structure, + add_layer_norm=add_layer_norm, + ) hypernet.save(fn) shared.reload_hypernetworks() diff --git a/modules/shared.py b/modules/shared.py index 0540cae9..faede821 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -260,8 +260,6 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), { "sd_model_checkpoint": OptionInfo(None, "Stable Diffusion checkpoint", gr.Dropdown, lambda: {"choices": modules.sd_models.checkpoint_tiles()}, refresh=sd_models.list_models), "sd_checkpoint_cache": OptionInfo(0, "Checkpoints to cache in RAM", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}), "sd_hypernetwork": OptionInfo("None", "Hypernetwork", gr.Dropdown, lambda: {"choices": ["None"] + [x for x in hypernetworks.keys()]}, refresh=reload_hypernetworks), - "sd_hypernetwork_layer_structure": OptionInfo(None, "Hypernetwork layer structure Default: (1,2,1).", gr.Dropdown, lambda: {"choices": [(1, 2, 1), (1, 2, 2, 1), (1, 2, 4, 2, 1)]}), - "sd_hypernetwork_add_layer_norm": OptionInfo(False, "Add layer normalization to hypernetwork architecture."), "sd_hypernetwork_strength": OptionInfo(1.0, "Hypernetwork strength", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.001}), "img2img_color_correction": OptionInfo(False, "Apply color correction to img2img results to match original colors."), "save_images_before_color_correction": OptionInfo(False, "Save a copy of image before applying color correction to img2img results"), diff --git a/modules/ui.py b/modules/ui.py index ca46343f..d9ee462f 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -458,14 +458,14 @@ def create_toprow(is_img2img): with gr.Row(): with gr.Column(scale=80): with gr.Row(): - prompt = gr.Textbox(label="Prompt", elem_id=f"{id_part}_prompt", show_label=False, lines=2, + prompt = gr.Textbox(label="Prompt", elem_id=f"{id_part}_prompt", show_label=False, lines=2, placeholder="Prompt (press Ctrl+Enter or Alt+Enter to generate)" ) with gr.Row(): with gr.Column(scale=80): with gr.Row(): - negative_prompt = gr.Textbox(label="Negative prompt", elem_id=f"{id_part}_neg_prompt", show_label=False, lines=2, + negative_prompt = gr.Textbox(label="Negative prompt", elem_id=f"{id_part}_neg_prompt", show_label=False, lines=2, placeholder="Negative prompt (press Ctrl+Enter or Alt+Enter to generate)" ) @@ -1198,6 +1198,8 @@ def create_ui(wrap_gradio_gpu_call): with gr.Tab(label="Create hypernetwork"): new_hypernetwork_name = gr.Textbox(label="Name") new_hypernetwork_sizes = gr.CheckboxGroup(label="Modules", value=["768", "320", "640", "1280"], choices=["768", "320", "640", "1280"]) + new_hypernetwork_layer_structure = gr.Dropdown(label="Hypernetwork layer structure", choices=[(1, 2, 1), (1, 2, 2, 1), (1, 2, 4, 2, 1)]) + new_hypernetwork_add_layer_norm = gr.Checkbox(label="Add layer normalization") with gr.Row(): with gr.Column(scale=3): @@ -1280,6 +1282,8 @@ def create_ui(wrap_gradio_gpu_call): inputs=[ new_hypernetwork_name, new_hypernetwork_sizes, + new_hypernetwork_layer_structure, + new_hypernetwork_add_layer_norm, ], outputs=[ train_hypernetwork_name, diff --git a/webui.py b/webui.py index c7260c7a..177bef74 100644 --- a/webui.py +++ b/webui.py @@ -85,9 +85,7 @@ def initialize(): shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: modules.sd_models.reload_model_weights(shared.sd_model))) shared.opts.onchange("sd_hypernetwork", wrap_queued_call(lambda: modules.hypernetworks.hypernetwork.load_hypernetwork(shared.opts.sd_hypernetwork))) shared.opts.onchange("sd_hypernetwork_strength", modules.hypernetworks.hypernetwork.apply_strength) - shared.opts.onchange("sd_hypernetwork_layer_structure", modules.hypernetworks.hypernetwork.apply_layer_structure) - shared.opts.onchange("sd_hypernetwork_add_layer_norm", modules.hypernetworks.hypernetwork.apply_layer_norm) - + # make the program just exit at ctrl+c without waiting for anything def sigint_handler(sig, frame): print(f'Interrupted with signal {sig} in {frame}') @@ -142,7 +140,7 @@ def webui(launch_api=False): create_api(app) wait_on_server(demo) - + sd_samplers.set_samplers() print('Reloading Custom Scripts') @@ -160,4 +158,4 @@ if __name__ == "__main__": if cmd_opts.nowebui: api_only() else: - webui(cmd_opts.api) \ No newline at end of file + webui(cmd_opts.api) -- cgit v1.2.3 From 2ce52d32e41fb523d1494f45073fd18496e52d35 Mon Sep 17 00:00:00 2001 From: discus0434 Date: Wed, 19 Oct 2022 16:31:12 +0000 Subject: fix for #3086 failing to load any previous hypernet --- modules/hypernetworks/hypernetwork.py | 60 ++++++++++++++++------------------- 1 file changed, 28 insertions(+), 32 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 7d519cd9..74300122 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -24,11 +24,10 @@ class HypernetworkModule(torch.nn.Module): def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False): super().__init__() - if layer_structure is not None: - assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" - assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" - else: - layer_structure = parse_layer_structure(dim, state_dict) + + assert layer_structure is not None, "layer_structure mut not be None" + assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" + assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" linears = [] for i in range(len(layer_structure) - 1): @@ -39,23 +38,30 @@ class HypernetworkModule(torch.nn.Module): self.linear = torch.nn.Sequential(*linears) if state_dict is not None: - try: - self.load_state_dict(state_dict) - except RuntimeError: - self.try_load_previous(state_dict) + self.fix_old_state_dict(state_dict) + self.load_state_dict(state_dict) else: for layer in self.linear: - layer.weight.data.normal_(mean = 0.0, std = 0.01) + layer.weight.data.normal_(mean=0.0, std=0.01) layer.bias.data.zero_() self.to(devices.device) - def try_load_previous(self, state_dict): - states = self.state_dict() - states['linear.0.bias'].copy_(state_dict['linear1.bias']) - states['linear.0.weight'].copy_(state_dict['linear1.weight']) - states['linear.1.bias'].copy_(state_dict['linear2.bias']) - states['linear.1.weight'].copy_(state_dict['linear2.weight']) + def fix_old_state_dict(self, state_dict): + changes = { + 'linear1.bias': 'linear.0.bias', + 'linear1.weight': 'linear.0.weight', + 'linear2.bias': 'linear.1.bias', + 'linear2.weight': 'linear.1.weight', + } + + for fr, to in changes.items(): + x = state_dict.get(fr, None) + if x is None: + continue + + del state_dict[fr] + state_dict[to] = x def forward(self, x): return x + self.linear(x) * self.multiplier @@ -71,18 +77,6 @@ def apply_strength(value=None): HypernetworkModule.multiplier = value if value is not None else shared.opts.sd_hypernetwork_strength -def parse_layer_structure(dim, state_dict): - i = 0 - layer_structure = [1] - - while (key := "linear.{}.weight".format(i)) in state_dict: - weight = state_dict[key] - layer_structure.append(len(weight) // dim) - i += 1 - - return layer_structure - - class Hypernetwork: filename = None name = None @@ -135,17 +129,18 @@ class Hypernetwork: state_dict = torch.load(filename, map_location='cpu') + self.layer_structure = state_dict.get('layer_structure', [1, 2, 1]) + self.add_layer_norm = state_dict.get('is_layer_norm', False) + for size, sd in state_dict.items(): if type(size) == int: self.layers[size] = ( - HypernetworkModule(size, sd[0], state_dict["layer_structure"], state_dict["is_layer_norm"]), - HypernetworkModule(size, sd[1], state_dict["layer_structure"], state_dict["is_layer_norm"]), + HypernetworkModule(size, sd[0], self.layer_structure, self.add_layer_norm), + HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm), ) self.name = state_dict.get('name', self.name) self.step = state_dict.get('step', 0) - self.layer_structure = state_dict.get('layer_structure', None) - self.add_layer_norm = state_dict.get('is_layer_norm', False) self.sd_checkpoint = state_dict.get('sd_checkpoint', None) self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None) @@ -244,6 +239,7 @@ def stack_conds(conds): return torch.stack(conds) + 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): assert hypernetwork_name, 'hypernetwork not selected' -- cgit v1.2.3 From d6ea5841374a28f3f6deb73abc251c8f0bcb240f Mon Sep 17 00:00:00 2001 From: DepFA <35278260+dfaker@users.noreply.github.com> Date: Thu, 20 Oct 2022 00:07:57 +0100 Subject: change html output --- modules/hypernetworks/hypernetwork.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 7d519cd9..73c1cb80 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -380,7 +380,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log Loss: {mean_loss:.7f}
Step: {hypernetwork.step}
Last prompt: {html.escape(entries[0].cond_text)}
-Last saved embedding: {html.escape(last_saved_file)}
+Last saved hypernetwork: {html.escape(last_saved_file)}
Last saved image: {html.escape(last_saved_image)}

""" -- cgit v1.2.3 From 6f98e89486f55b0e4657e96ce640cf1c4675d187 Mon Sep 17 00:00:00 2001 From: discus0434 Date: Thu, 20 Oct 2022 00:10:45 +0000 Subject: update --- modules/hypernetworks/hypernetwork.py | 29 +++++++++++++++-------- modules/hypernetworks/ui.py | 3 ++- modules/ui.py | 43 +++++++++++++++++++---------------- 3 files changed, 44 insertions(+), 31 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 74300122..7d617680 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -22,16 +22,20 @@ from modules.textual_inversion.learn_schedule import LearnRateScheduler class HypernetworkModule(torch.nn.Module): multiplier = 1.0 - def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False): + def __init__(self, dim, state_dict=None, layer_structure=None, add_layer_norm=False, activation_func=None): super().__init__() - assert layer_structure is not None, "layer_structure mut not be None" + assert layer_structure is not None, "layer_structure must not be None" assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" linears = [] for i in range(len(layer_structure) - 1): linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1]))) + if activation_func == "relu": + linears.append(torch.nn.ReLU()) + if activation_func == "leakyrelu": + linears.append(torch.nn.LeakyReLU()) if add_layer_norm: linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1]))) @@ -42,8 +46,9 @@ class HypernetworkModule(torch.nn.Module): self.load_state_dict(state_dict) else: for layer in self.linear: - layer.weight.data.normal_(mean=0.0, std=0.01) - layer.bias.data.zero_() + if not "ReLU" in layer.__str__(): + layer.weight.data.normal_(mean=0.0, std=0.01) + layer.bias.data.zero_() self.to(devices.device) @@ -69,7 +74,8 @@ class HypernetworkModule(torch.nn.Module): def trainables(self): layer_structure = [] for layer in self.linear: - layer_structure += [layer.weight, layer.bias] + if not "ReLU" in layer.__str__(): + layer_structure += [layer.weight, layer.bias] return layer_structure @@ -81,7 +87,7 @@ class Hypernetwork: filename = None name = None - def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False): + def __init__(self, name=None, enable_sizes=None, layer_structure=None, add_layer_norm=False, activation_func=None): self.filename = None self.name = name self.layers = {} @@ -90,11 +96,12 @@ class Hypernetwork: self.sd_checkpoint_name = None self.layer_structure = layer_structure self.add_layer_norm = add_layer_norm + self.activation_func = activation_func for size in enable_sizes or []: self.layers[size] = ( - HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm), - HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm), + HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm, self.activation_func), + HypernetworkModule(size, None, self.layer_structure, self.add_layer_norm, self.activation_func), ) def weights(self): @@ -117,6 +124,7 @@ class Hypernetwork: state_dict['name'] = self.name state_dict['layer_structure'] = self.layer_structure state_dict['is_layer_norm'] = self.add_layer_norm + state_dict['activation_func'] = self.activation_func state_dict['sd_checkpoint'] = self.sd_checkpoint state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name @@ -131,12 +139,13 @@ class Hypernetwork: self.layer_structure = state_dict.get('layer_structure', [1, 2, 1]) self.add_layer_norm = state_dict.get('is_layer_norm', False) + self.activation_func = state_dict.get('activation_func', None) for size, sd in state_dict.items(): if type(size) == int: self.layers[size] = ( - HypernetworkModule(size, sd[0], self.layer_structure, self.add_layer_norm), - HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm), + HypernetworkModule(size, sd[0], self.layer_structure, self.add_layer_norm, self.activation_func), + HypernetworkModule(size, sd[1], self.layer_structure, self.add_layer_norm, self.activation_func), ) self.name = state_dict.get('name', self.name) diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py index 08f75f15..83f9547b 100644 --- a/modules/hypernetworks/ui.py +++ b/modules/hypernetworks/ui.py @@ -10,7 +10,7 @@ from modules import sd_hijack, shared, devices from modules.hypernetworks import hypernetwork -def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False): +def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm=False, activation_func=None): fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt") assert not os.path.exists(fn), f"file {fn} already exists" @@ -22,6 +22,7 @@ def create_hypernetwork(name, enable_sizes, layer_structure=None, add_layer_norm enable_sizes=[int(x) for x in enable_sizes], layer_structure=layer_structure, add_layer_norm=add_layer_norm, + activation_func=activation_func, ) hypernet.save(fn) diff --git a/modules/ui.py b/modules/ui.py index d2e24880..8751fa9c 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -5,43 +5,44 @@ import json import math import mimetypes import os +import platform import random +import subprocess as sp import sys import tempfile import time import traceback -import platform -import subprocess as sp from functools import partial, reduce +import gradio as gr +import gradio.routes +import gradio.utils import numpy as np +import piexif import torch from PIL import Image, PngImagePlugin -import piexif -import gradio as gr -import gradio.utils -import gradio.routes - -from modules import sd_hijack, sd_models, localization +from modules import localization, sd_hijack, sd_models from modules.paths import script_path -from modules.shared import opts, cmd_opts, restricted_opts +from modules.shared import cmd_opts, opts, restricted_opts + if cmd_opts.deepdanbooru: from modules.deepbooru import get_deepbooru_tags -import modules.shared as shared -from modules.sd_samplers import samplers, samplers_for_img2img -from modules.sd_hijack import model_hijack + +import modules.codeformer_model +import modules.generation_parameters_copypaste +import modules.gfpgan_model +import modules.hypernetworks.ui +import modules.images_history as img_his import modules.ldsr_model import modules.scripts -import modules.gfpgan_model -import modules.codeformer_model +import modules.shared as shared import modules.styles -import modules.generation_parameters_copypaste +import modules.textual_inversion.ui from modules import prompt_parser from modules.images import save_image -import modules.textual_inversion.ui -import modules.hypernetworks.ui -import modules.images_history as img_his +from modules.sd_hijack import model_hijack +from modules.sd_samplers import samplers, samplers_for_img2img # this is a fix for Windows users. Without it, javascript files will be served with text/html content-type and the browser will not show any UI mimetypes.init() @@ -268,8 +269,8 @@ def calc_time_left(progress, threshold, label, force_display): time_since_start = time.time() - shared.state.time_start eta = (time_since_start/progress) eta_relative = eta-time_since_start - if (eta_relative > threshold and progress > 0.02) or force_display: - return label + time.strftime('%H:%M:%S', time.gmtime(eta_relative)) + if (eta_relative > threshold and progress > 0.02) or force_display: + return label + time.strftime('%H:%M:%S', time.gmtime(eta_relative)) else: return "" @@ -1219,6 +1220,7 @@ def create_ui(wrap_gradio_gpu_call): new_hypernetwork_sizes = gr.CheckboxGroup(label="Modules", value=["768", "320", "640", "1280"], choices=["768", "320", "640", "1280"]) new_hypernetwork_layer_structure = gr.Textbox("1, 2, 1", label="Enter hypernetwork layer structure", placeholder="1st and last digit must be 1. ex:'1, 2, 1'") new_hypernetwork_add_layer_norm = gr.Checkbox(label="Add layer normalization") + new_hypernetwork_activation_func = gr.Dropdown(value="relu", label="Select activation function of hypernetwork", choices=["relu", "leakyrelu"]) with gr.Row(): with gr.Column(scale=3): @@ -1303,6 +1305,7 @@ def create_ui(wrap_gradio_gpu_call): new_hypernetwork_sizes, new_hypernetwork_layer_structure, new_hypernetwork_add_layer_norm, + new_hypernetwork_activation_func, ], outputs=[ train_hypernetwork_name, -- cgit v1.2.3 From c23f666dba2b484d521d2dc4be91cf9e09312647 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Fri, 21 Oct 2022 09:47:43 +0300 Subject: a more strict check for activation type and a more reasonable check for type of layer in hypernets --- modules/hypernetworks/hypernetwork.py | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 7d617680..84e7e350 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -32,10 +32,16 @@ class HypernetworkModule(torch.nn.Module): linears = [] for i in range(len(layer_structure) - 1): linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1]))) + if activation_func == "relu": linears.append(torch.nn.ReLU()) - if activation_func == "leakyrelu": + elif activation_func == "leakyrelu": linears.append(torch.nn.LeakyReLU()) + elif activation_func == 'linear' or activation_func is None: + pass + else: + raise RuntimeError(f'hypernetwork uses an unsupported activation function: {activation_func}') + if add_layer_norm: linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1]))) @@ -46,7 +52,7 @@ class HypernetworkModule(torch.nn.Module): self.load_state_dict(state_dict) else: for layer in self.linear: - if not "ReLU" in layer.__str__(): + if type(layer) == torch.nn.Linear: layer.weight.data.normal_(mean=0.0, std=0.01) layer.bias.data.zero_() @@ -74,7 +80,7 @@ class HypernetworkModule(torch.nn.Module): def trainables(self): layer_structure = [] for layer in self.linear: - if not "ReLU" in layer.__str__(): + if type(layer) == torch.nn.Linear: layer_structure += [layer.weight, layer.bias] return layer_structure -- cgit v1.2.3 From 5245c7a4935f67b677da0f5a1fc2b74c074aa0e2 Mon Sep 17 00:00:00 2001 From: timntorres Date: Wed, 19 Oct 2022 12:21:32 -0700 Subject: Issue #2921-Give PNG info to Hypernet previews. --- modules/hypernetworks/hypernetwork.py | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 84e7e350..68c8f26d 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -256,6 +256,9 @@ def stack_conds(conds): 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): + # images is required here to give training previews their infotext. Importing this at the very top causes a circular dependency. + from modules import images + assert hypernetwork_name, 'hypernetwork not selected' path = shared.hypernetworks.get(hypernetwork_name, None) @@ -298,6 +301,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log last_saved_file = "" last_saved_image = "" + forced_filename = "" ititial_step = hypernetwork.step or 0 if ititial_step > steps: @@ -345,7 +349,8 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log }) if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0: - last_saved_image = os.path.join(images_dir, f'{hypernetwork_name}-{hypernetwork.step}.png') + forced_filename = f'{hypernetwork_name}-{hypernetwork.step}' + last_saved_image = os.path.join(images_dir, forced_filename) optimizer.zero_grad() shared.sd_model.cond_stage_model.to(devices.device) @@ -381,7 +386,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log if image is not None: shared.state.current_image = image - image.save(last_saved_image) + last_saved_image, last_text_info = images.save_image(image, images_dir, "", p.seed, p.prompt, shared.opts.samples_format, processed.infotexts[0], p=p, forced_filename=forced_filename) last_saved_image += f", prompt: {preview_text}" shared.state.job_no = hypernetwork.step -- cgit v1.2.3 From 4ff274e1e35bb642687253ce744d2cfa738ab293 Mon Sep 17 00:00:00 2001 From: timntorres Date: Wed, 19 Oct 2022 12:32:22 -0700 Subject: Revise comments. --- modules/hypernetworks/hypernetwork.py | 2 +- modules/images.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 68c8f26d..3f96361c 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -256,7 +256,7 @@ def stack_conds(conds): 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): - # images is required here to give training previews their infotext. Importing this at the very top causes a circular dependency. + # images allows training previews to have infotext. Importing it at the top causes a circular import problem. from modules import images assert hypernetwork_name, 'hypernetwork not selected' diff --git a/modules/images.py b/modules/images.py index 550e53ae..b8834e3c 100644 --- a/modules/images.py +++ b/modules/images.py @@ -419,7 +419,7 @@ def save_image(image, path, basename, seed=None, prompt=None, extension='png', i try: os.makedirs(path, exist_ok=True) except FileExistsError: - # If the file already exists, continue and allow said file to be overwritten. + # If the file already exists, allow said file to be overwritten. pass if forced_filename is None: -- cgit v1.2.3 From 03a1e288c4973dd2dff57a97469b40f146b6fccf Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Fri, 21 Oct 2022 10:13:24 +0300 Subject: turns out LayerNorm also has weight and bias and needs to be pre-multiplied and trained for hypernets --- modules/hypernetworks/hypernetwork.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'modules/hypernetworks/hypernetwork.py') diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py index 3274a802..b1a5d0c7 100644 --- a/modules/hypernetworks/hypernetwork.py +++ b/modules/hypernetworks/hypernetwork.py @@ -52,7 +52,7 @@ class HypernetworkModule(torch.nn.Module): self.load_state_dict(state_dict) else: for layer in self.linear: - if type(layer) == torch.nn.Linear: + if type(layer) == torch.nn.Linear or type(layer) == torch.nn.LayerNorm: layer.weight.data.normal_(mean=0.0, std=0.01) layer.bias.data.zero_() @@ -80,7 +80,7 @@ class HypernetworkModule(torch.nn.Module): def trainables(self): layer_structure = [] for layer in self.linear: - if type(layer) == torch.nn.Linear: + if type(layer) == torch.nn.Linear or type(layer) == torch.nn.LayerNorm: layer_structure += [layer.weight, layer.bias] return layer_structure -- cgit v1.2.3