From a68f4690307b0f94404b85513429e18ee3936589 Mon Sep 17 00:00:00 2001 From: catboxanon <122327233+catboxanon@users.noreply.github.com> Date: Mon, 24 Jul 2023 17:54:59 -0400 Subject: Fix to parse TE in some LoRAs --- extensions-builtin/Lora/networks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index af8188e3..3a8cfa3b 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -161,7 +161,7 @@ def load_network(name, network_on_disk): sd_module = shared.sd_model.network_layer_mapping.get(key, None) elif sd_module is None and "lora_te1_text_model" in key_network_without_network_parts: key = key_network_without_network_parts.replace("lora_te1_text_model", "0_transformer_text_model") - sd_module = shared.sd_model.network_layer_mapping.get(key, None) + sd_module = shared.sd_model.network_layer_mapping.get(key, None) or shared.sd_model.network_layer_mapping.get(key[2:], None) if sd_module is None: keys_failed_to_match[key_network] = key -- cgit v1.2.3 From d0bf509fa14babebedbaef121ef54599003aa457 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Tue, 25 Jul 2023 16:18:10 +0300 Subject: fix for #11963 --- extensions-builtin/Lora/networks.py | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 3a8cfa3b..17cbe1bb 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -161,7 +161,12 @@ def load_network(name, network_on_disk): sd_module = shared.sd_model.network_layer_mapping.get(key, None) elif sd_module is None and "lora_te1_text_model" in key_network_without_network_parts: key = key_network_without_network_parts.replace("lora_te1_text_model", "0_transformer_text_model") - sd_module = shared.sd_model.network_layer_mapping.get(key, None) or shared.sd_model.network_layer_mapping.get(key[2:], None) + sd_module = shared.sd_model.network_layer_mapping.get(key, None) + + # some SD1 Loras also have correct compvis keys + if sd_module is None: + key = key_network_without_network_parts.replace("lora_te1_text_model", "transformer_text_model") + sd_module = shared.sd_model.network_layer_mapping.get(key, None) if sd_module is None: keys_failed_to_match[key_network] = key -- cgit v1.2.3 From 91a131aa6ca26d35f7879d8240e4f6e321130160 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 27 Jul 2023 09:00:47 +0300 Subject: update lora extension to work with python 3.8 --- extensions-builtin/Lora/network.py | 1 + 1 file changed, 1 insertion(+) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/network.py b/extensions-builtin/Lora/network.py index 8ecfa29a..0a18d69e 100644 --- a/extensions-builtin/Lora/network.py +++ b/extensions-builtin/Lora/network.py @@ -1,3 +1,4 @@ +from __future__ import annotations import os from collections import namedtuple import enum -- cgit v1.2.3 From 362789a3793025c698fa42372fd66c3c4f2d6413 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Fri, 4 Aug 2023 07:50:17 +0300 Subject: gradio 3.39 --- extensions-builtin/Lora/ui_edit_user_metadata.py | 2 +- modules/gradio_extensons.py | 60 ++++++++++++++++++++++++ modules/scripts.py | 60 ------------------------ modules/shared.py | 3 +- modules/ui.py | 24 +++++----- modules/ui_checkpoint_merger.py | 2 +- modules/ui_common.py | 2 +- modules/ui_components.py | 2 +- modules/ui_extensions.py | 8 ++-- modules/ui_postprocessing.py | 2 +- requirements.txt | 2 +- requirements_versions.txt | 2 +- style.css | 12 ++++- 13 files changed, 95 insertions(+), 86 deletions(-) create mode 100644 modules/gradio_extensons.py (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/ui_edit_user_metadata.py b/extensions-builtin/Lora/ui_edit_user_metadata.py index 2ca997f7..390d9dde 100644 --- a/extensions-builtin/Lora/ui_edit_user_metadata.py +++ b/extensions-builtin/Lora/ui_edit_user_metadata.py @@ -167,7 +167,7 @@ class LoraUserMetadataEditor(ui_extra_networks_user_metadata.UserMetadataEditor) random_prompt = gr.Textbox(label='Random prompt', lines=4, max_lines=4, interactive=False) with gr.Column(scale=1, min_width=120): - generate_random_prompt = gr.Button('Generate').style(full_width=True, size="lg") + generate_random_prompt = gr.Button('Generate', size="lg", scale=1) self.edit_notes = gr.TextArea(label='Notes', lines=4) diff --git a/modules/gradio_extensons.py b/modules/gradio_extensons.py new file mode 100644 index 00000000..5af7fd8e --- /dev/null +++ b/modules/gradio_extensons.py @@ -0,0 +1,60 @@ +import gradio as gr + +from modules import scripts + +def add_classes_to_gradio_component(comp): + """ + this adds gradio-* to the component for css styling (ie gradio-button to gr.Button), as well as some others + """ + + comp.elem_classes = [f"gradio-{comp.get_block_name()}", *(comp.elem_classes or [])] + + if getattr(comp, 'multiselect', False): + comp.elem_classes.append('multiselect') + + +def IOComponent_init(self, *args, **kwargs): + self.webui_tooltip = kwargs.pop('tooltip', None) + + if scripts.scripts_current is not None: + scripts.scripts_current.before_component(self, **kwargs) + + scripts.script_callbacks.before_component_callback(self, **kwargs) + + res = original_IOComponent_init(self, *args, **kwargs) + + add_classes_to_gradio_component(self) + + scripts.script_callbacks.after_component_callback(self, **kwargs) + + if scripts.scripts_current is not None: + scripts.scripts_current.after_component(self, **kwargs) + + return res + + +def Block_get_config(self): + config = original_Block_get_config(self) + + webui_tooltip = getattr(self, 'webui_tooltip', None) + if webui_tooltip: + config["webui_tooltip"] = webui_tooltip + + return config + + +def BlockContext_init(self, *args, **kwargs): + res = original_BlockContext_init(self, *args, **kwargs) + + add_classes_to_gradio_component(self) + + return res + + +original_IOComponent_init = gr.components.IOComponent.__init__ +original_Block_get_config = gr.blocks.Block.get_config +original_BlockContext_init = gr.blocks.BlockContext.__init__ + +gr.components.IOComponent.__init__ = IOComponent_init +gr.blocks.Block.get_config = Block_get_config +gr.blocks.BlockContext.__init__ = BlockContext_init diff --git a/modules/scripts.py b/modules/scripts.py index edf7347e..f7d060aa 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -631,63 +631,3 @@ def reload_script_body_only(): reload_scripts = load_scripts # compatibility alias - - -def add_classes_to_gradio_component(comp): - """ - this adds gradio-* to the component for css styling (ie gradio-button to gr.Button), as well as some others - """ - - comp.elem_classes = [f"gradio-{comp.get_block_name()}", *(comp.elem_classes or [])] - - if getattr(comp, 'multiselect', False): - comp.elem_classes.append('multiselect') - - - -def IOComponent_init(self, *args, **kwargs): - self.webui_tooltip = kwargs.pop('tooltip', None) - - if scripts_current is not None: - scripts_current.before_component(self, **kwargs) - - script_callbacks.before_component_callback(self, **kwargs) - - res = original_IOComponent_init(self, *args, **kwargs) - - add_classes_to_gradio_component(self) - - script_callbacks.after_component_callback(self, **kwargs) - - if scripts_current is not None: - scripts_current.after_component(self, **kwargs) - - return res - - -def Block_get_config(self): - config = original_Block_get_config(self) - - webui_tooltip = getattr(self, 'webui_tooltip', None) - if webui_tooltip: - config["webui_tooltip"] = webui_tooltip - - return config - - -original_IOComponent_init = gr.components.IOComponent.__init__ -original_Block_get_config = gr.components.Block.get_config -gr.components.IOComponent.__init__ = IOComponent_init -gr.components.Block.get_config = Block_get_config - - -def BlockContext_init(self, *args, **kwargs): - res = original_BlockContext_init(self, *args, **kwargs) - - add_classes_to_gradio_component(self) - - return res - - -original_BlockContext_init = gr.blocks.BlockContext.__init__ -gr.blocks.BlockContext.__init__ = BlockContext_init diff --git a/modules/shared.py b/modules/shared.py index 7103b4ca..cec030f7 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -385,7 +385,8 @@ options_templates.update(options_section(('face-restoration', "Face restoration" })) options_templates.update(options_section(('system', "System"), { - "show_warnings": OptionInfo(False, "Show warnings in console."), + "show_warnings": OptionInfo(False, "Show warnings in console.").needs_restart(), + "show_gradio_deprecation_warnings": OptionInfo(True, "Show gradio deprecation warnings in console.").needs_restart(), "memmon_poll_rate": OptionInfo(8, "VRAM usage polls per second during generation.", gr.Slider, {"minimum": 0, "maximum": 40, "step": 1}).info("0 = disable"), "samples_log_stdout": OptionInfo(False, "Always print all generation info to standard output"), "multiple_tqdm": OptionInfo(True, "Add a second progress bar to the console that shows progress for an entire job."), diff --git a/modules/ui.py b/modules/ui.py index 03306ba9..822a7660 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -12,6 +12,7 @@ import numpy as np from PIL import Image, PngImagePlugin # noqa: F401 from modules.call_queue import wrap_gradio_gpu_call, wrap_queued_call, wrap_gradio_call +from modules import gradio_extensons # noqa: F401 from modules import sd_hijack, sd_models, script_callbacks, ui_extensions, deepbooru, extra_networks, ui_common, ui_postprocessing, progress, ui_loadsave, errors, shared_items, ui_settings, timer, sysinfo, ui_checkpoint_merger, ui_prompt_styles, scripts from modules.ui_components import FormRow, FormGroup, ToolButton, FormHTML from modules.paths import script_path @@ -34,6 +35,7 @@ from modules.generation_parameters_copypaste import image_from_url_text create_setting_component = ui_settings.create_setting_component warnings.filterwarnings("default" if opts.show_warnings else "ignore", category=UserWarning) +warnings.filterwarnings("default" if opts.show_gradio_deprecation_warnings else "ignore", category=gr.deprecation.GradioDeprecationWarning) # 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() @@ -146,7 +148,6 @@ def interrogate_deepbooru(image): def create_seed_inputs(target_interface): with FormRow(elem_id=f"{target_interface}_seed_row", variant="compact"): seed = (gr.Textbox if cmd_opts.use_textbox_seed else gr.Number)(label='Seed', value=-1, elem_id=f"{target_interface}_seed") - seed.style(container=False) random_seed = ToolButton(random_symbol, elem_id=f"{target_interface}_random_seed", label='Random seed') reuse_seed = ToolButton(reuse_symbol, elem_id=f"{target_interface}_reuse_seed", label='Reuse seed') @@ -158,7 +159,6 @@ def create_seed_inputs(target_interface): with FormRow(visible=False, elem_id=f"{target_interface}_subseed_row") as seed_extra_row_1: seed_extras.append(seed_extra_row_1) subseed = gr.Number(label='Variation seed', value=-1, elem_id=f"{target_interface}_subseed") - subseed.style(container=False) random_subseed = ToolButton(random_symbol, elem_id=f"{target_interface}_random_subseed") reuse_subseed = ToolButton(reuse_symbol, elem_id=f"{target_interface}_reuse_subseed") subseed_strength = gr.Slider(label='Variation strength', value=0.0, minimum=0, maximum=1, step=0.01, elem_id=f"{target_interface}_subseed_strength") @@ -408,7 +408,7 @@ def create_ui(): from modules import ui_extra_networks extra_networks_ui = ui_extra_networks.create_ui(extra_networks, toprow.extra_networks_button, 'txt2img') - with gr.Row().style(equal_height=False): + with gr.Row(equal_height=False): with gr.Column(variant='compact', elem_id="txt2img_settings"): scripts.scripts_txt2img.prepare_ui() @@ -636,7 +636,7 @@ def create_ui(): from modules import ui_extra_networks extra_networks_ui_img2img = ui_extra_networks.create_ui(extra_networks, toprow.extra_networks_button, 'img2img') - with FormRow().style(equal_height=False): + with FormRow(equal_height=False): with gr.Column(variant='compact', elem_id="img2img_settings"): copy_image_buttons = [] copy_image_destinations = {} @@ -658,19 +658,19 @@ def create_ui(): img2img_selected_tab = gr.State(0) with gr.TabItem('img2img', id='img2img', elem_id="img2img_img2img_tab") as tab_img2img: - init_img = gr.Image(label="Image for img2img", elem_id="img2img_image", show_label=False, source="upload", interactive=True, type="pil", tool="editor", image_mode="RGBA").style(height=opts.img2img_editor_height) + init_img = gr.Image(label="Image for img2img", elem_id="img2img_image", show_label=False, source="upload", interactive=True, type="pil", tool="editor", image_mode="RGBA", height=opts.img2img_editor_height) add_copy_image_controls('img2img', init_img) with gr.TabItem('Sketch', id='img2img_sketch', elem_id="img2img_img2img_sketch_tab") as tab_sketch: - sketch = gr.Image(label="Image for img2img", elem_id="img2img_sketch", show_label=False, source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGBA").style(height=opts.img2img_editor_height) + sketch = gr.Image(label="Image for img2img", elem_id="img2img_sketch", show_label=False, source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGBA", height=opts.img2img_editor_height) add_copy_image_controls('sketch', sketch) with gr.TabItem('Inpaint', id='inpaint', elem_id="img2img_inpaint_tab") as tab_inpaint: - init_img_with_mask = gr.Image(label="Image for inpainting with mask", show_label=False, elem_id="img2maskimg", source="upload", interactive=True, type="pil", tool="sketch", image_mode="RGBA").style(height=opts.img2img_editor_height) + init_img_with_mask = gr.Image(label="Image for inpainting with mask", show_label=False, elem_id="img2maskimg", source="upload", interactive=True, type="pil", tool="sketch", image_mode="RGBA", height=opts.img2img_editor_height) add_copy_image_controls('inpaint', init_img_with_mask) with gr.TabItem('Inpaint sketch', id='inpaint_sketch', elem_id="img2img_inpaint_sketch_tab") as tab_inpaint_color: - inpaint_color_sketch = gr.Image(label="Color sketch inpainting", show_label=False, elem_id="inpaint_sketch", source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGBA").style(height=opts.img2img_editor_height) + inpaint_color_sketch = gr.Image(label="Color sketch inpainting", show_label=False, elem_id="inpaint_sketch", source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGBA", height=opts.img2img_editor_height) inpaint_color_sketch_orig = gr.State(None) add_copy_image_controls('inpaint_sketch', inpaint_color_sketch) @@ -993,7 +993,7 @@ def create_ui(): ui_postprocessing.create_ui() with gr.Blocks(analytics_enabled=False) as pnginfo_interface: - with gr.Row().style(equal_height=False): + with gr.Row(equal_height=False): with gr.Column(variant='panel'): image = gr.Image(elem_id="pnginfo_image", label="Source", source="upload", interactive=True, type="pil") @@ -1018,10 +1018,10 @@ def create_ui(): modelmerger_ui = ui_checkpoint_merger.UiCheckpointMerger() with gr.Blocks(analytics_enabled=False) as train_interface: - with gr.Row().style(equal_height=False): + with gr.Row(equal_height=False): gr.HTML(value="

See wiki for detailed explanation.

") - with gr.Row(variant="compact").style(equal_height=False): + with gr.Row(variant="compact", equal_height=False): with gr.Tabs(elem_id="train_tabs"): with gr.Tab(label="Create embedding", id="create_embedding"): @@ -1181,7 +1181,7 @@ def create_ui(): with gr.Column(elem_id='ti_gallery_container'): ti_output = gr.Text(elem_id="ti_output", value="", show_label=False) - gr.Gallery(label='Output', show_label=False, elem_id='ti_gallery').style(columns=4) + gr.Gallery(label='Output', show_label=False, elem_id='ti_gallery', columns=4) gr.HTML(elem_id="ti_progress", value="") ti_outcome = gr.HTML(elem_id="ti_error", value="") diff --git a/modules/ui_checkpoint_merger.py b/modules/ui_checkpoint_merger.py index 4863d861..f9c5dd6b 100644 --- a/modules/ui_checkpoint_merger.py +++ b/modules/ui_checkpoint_merger.py @@ -29,7 +29,7 @@ def modelmerger(*args): class UiCheckpointMerger: def __init__(self): with gr.Blocks(analytics_enabled=False) as modelmerger_interface: - with gr.Row().style(equal_height=False): + with gr.Row(equal_height=False): with gr.Column(variant='compact'): self.interp_description = gr.HTML(value=update_interp_description("Weighted sum"), elem_id="modelmerger_interp_description") diff --git a/modules/ui_common.py b/modules/ui_common.py index ba75fa73..eefe0c0e 100644 --- a/modules/ui_common.py +++ b/modules/ui_common.py @@ -134,7 +134,7 @@ Requested path was: {f} with gr.Column(variant='panel', elem_id=f"{tabname}_results"): with gr.Group(elem_id=f"{tabname}_gallery_container"): - result_gallery = gr.Gallery(label='Output', show_label=False, elem_id=f"{tabname}_gallery").style(columns=4) + result_gallery = gr.Gallery(label='Output', show_label=False, elem_id=f"{tabname}_gallery", columns=4) generation_info = None with gr.Column(): diff --git a/modules/ui_components.py b/modules/ui_components.py index 64451df7..8f8a7088 100644 --- a/modules/ui_components.py +++ b/modules/ui_components.py @@ -35,7 +35,7 @@ class FormColumn(FormComponent, gr.Column): class FormGroup(FormComponent, gr.Group): - """Same as gr.Row but fits inside gradio forms""" + """Same as gr.Group but fits inside gradio forms""" def get_block_name(self): return "group" diff --git a/modules/ui_extensions.py b/modules/ui_extensions.py index bd28bfcf..15a8b0bf 100644 --- a/modules/ui_extensions.py +++ b/modules/ui_extensions.py @@ -533,8 +533,8 @@ def create_ui(): apply = gr.Button(value=apply_label, variant="primary") check = gr.Button(value="Check for updates") extensions_disable_all = gr.Radio(label="Disable all extensions", choices=["none", "extra", "all"], value=shared.opts.disable_all_extensions, elem_id="extensions_disable_all") - extensions_disabled_list = gr.Text(elem_id="extensions_disabled_list", visible=False).style(container=False) - extensions_update_list = gr.Text(elem_id="extensions_update_list", visible=False).style(container=False) + extensions_disabled_list = gr.Text(elem_id="extensions_disabled_list", visible=False, container=False) + extensions_update_list = gr.Text(elem_id="extensions_update_list", visible=False, container=False) html = "" @@ -569,7 +569,7 @@ def create_ui(): with gr.Row(): refresh_available_extensions_button = gr.Button(value="Load from:", variant="primary") extensions_index_url = os.environ.get('WEBUI_EXTENSIONS_INDEX', "https://raw.githubusercontent.com/AUTOMATIC1111/stable-diffusion-webui-extensions/master/index.json") - available_extensions_index = gr.Text(value=extensions_index_url, label="Extension index URL").style(container=False) + available_extensions_index = gr.Text(value=extensions_index_url, label="Extension index URL", container=False) extension_to_install = gr.Text(elem_id="extension_to_install", visible=False) install_extension_button = gr.Button(elem_id="install_extension_button", visible=False) @@ -578,7 +578,7 @@ def create_ui(): sort_column = gr.Radio(value="newest first", label="Order", choices=["newest first", "oldest first", "a-z", "z-a", "internal order",'update time', 'create time', "stars"], type="index") with gr.Row(): - search_extensions_text = gr.Text(label="Search").style(container=False) + search_extensions_text = gr.Text(label="Search", container=False) install_result = gr.HTML() available_extensions_table = gr.HTML() diff --git a/modules/ui_postprocessing.py b/modules/ui_postprocessing.py index c7dc1154..802e1ce7 100644 --- a/modules/ui_postprocessing.py +++ b/modules/ui_postprocessing.py @@ -6,7 +6,7 @@ import modules.generation_parameters_copypaste as parameters_copypaste def create_ui(): tab_index = gr.State(value=0) - with gr.Row().style(equal_height=False, variant='compact'): + with gr.Row(equal_height=False, variant='compact'): with gr.Column(variant='compact'): with gr.Tabs(elem_id="mode_extras"): with gr.TabItem('Single Image', id="single_image", elem_id="extras_single_tab") as tab_single: diff --git a/requirements.txt b/requirements.txt index b3f8a7f4..afdc6ee2 100644 --- a/requirements.txt +++ b/requirements.txt @@ -7,7 +7,7 @@ blendmodes clean-fid einops gfpgan -gradio==3.32.0 +gradio==3.39.0 inflection jsonmerge kornia diff --git a/requirements_versions.txt b/requirements_versions.txt index d07ab456..82b8732d 100644 --- a/requirements_versions.txt +++ b/requirements_versions.txt @@ -7,7 +7,7 @@ clean-fid==0.1.35 einops==0.4.1 fastapi==0.94.0 gfpgan==1.3.8 -gradio==3.32.0 +gradio==3.39.0 httpcore==0.15 inflection==0.5.1 jsonmerge==1.8.0 diff --git a/style.css b/style.css index cf8470e4..86b4f61e 100644 --- a/style.css +++ b/style.css @@ -8,6 +8,7 @@ --checkbox-label-gap: 0.25em 0.1em; --section-header-text-size: 12pt; --block-background-fill: transparent; + } .block.padded:not(.gradio-accordion) { @@ -42,7 +43,8 @@ div.form{ .block.gradio-radio, .block.gradio-checkboxgroup, .block.gradio-number, -.block.gradio-colorpicker +.block.gradio-colorpicker, +div.gradio-group { border-width: 0 !important; box-shadow: none !important; @@ -133,6 +135,11 @@ a{ cursor: pointer; } +div.styler{ + border: none; + background: var(--background-fill-primary); +} + /* general styled components */ @@ -164,7 +171,7 @@ a{ .checkboxes-row > div{ flex: 0; white-space: nowrap; - min-width: auto; + min-width: auto !important; } button.custom-button{ @@ -388,6 +395,7 @@ div#extras_scale_to_tab div.form{ #quicksettings > div, #quicksettings > fieldset{ max-width: 24em; min-width: 24em; + width: 24em; padding: 0; border: none; box-shadow: none; -- cgit v1.2.3 From eed963e97261ee03bffe59e3d343dcf53d82dbfd Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Wed, 9 Aug 2023 16:54:49 +0300 Subject: Lora cache in memory --- extensions-builtin/Lora/networks.py | 22 +++++++++++++++++++--- extensions-builtin/Lora/scripts/lora_script.py | 3 +++ 2 files changed, 22 insertions(+), 3 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 17cbe1bb..bc722e90 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -195,6 +195,15 @@ def load_network(name, network_on_disk): return net +def purge_networks_from_memory(): + while len(networks_in_memory) > shared.opts.lora_in_memory_limit and len(networks_in_memory) > 0: + name = next(iter(networks_in_memory)) + networks_in_memory.pop(name, None) + + devices.torch_gc() + + + def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=None): already_loaded = {} @@ -212,15 +221,19 @@ def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=No failed_to_load_networks = [] - for i, name in enumerate(names): + for i, (network_on_disk, name) in enumerate(zip(networks_on_disk, names)): net = already_loaded.get(name, None) - network_on_disk = networks_on_disk[i] - if network_on_disk is not None: + if net is None: + net = networks_in_memory.get(name) + if net is None or os.path.getmtime(network_on_disk.filename) > net.mtime: try: net = load_network(name, network_on_disk) + + networks_in_memory.pop(name, None) + networks_in_memory[name] = net except Exception as e: errors.display(e, f"loading network {network_on_disk.filename}") continue @@ -242,6 +255,8 @@ def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=No if failed_to_load_networks: sd_hijack.model_hijack.comments.append("Failed to find networks: " + ", ".join(failed_to_load_networks)) + purge_networks_from_memory() + def network_restore_weights_from_backup(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.MultiheadAttention]): weights_backup = getattr(self, "network_weights_backup", None) @@ -462,6 +477,7 @@ def infotext_pasted(infotext, params): available_networks = {} available_network_aliases = {} loaded_networks = [] +networks_in_memory = {} available_network_hash_lookup = {} forbidden_network_aliases = {} diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py index cd28afc9..6ab8b6e7 100644 --- a/extensions-builtin/Lora/scripts/lora_script.py +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -65,6 +65,7 @@ shared.options_templates.update(shared.options_section(('extra_networks', "Extra "lora_add_hashes_to_infotext": shared.OptionInfo(True, "Add Lora hashes to infotext"), "lora_show_all": shared.OptionInfo(False, "Always show all networks on the Lora page").info("otherwise, those detected as for incompatible version of Stable Diffusion will be hidden"), "lora_hide_unknown_for_versions": shared.OptionInfo([], "Hide networks of unknown versions for model versions", gr.CheckboxGroup, {"choices": ["SD1", "SD2", "SDXL"]}), + "lora_in_memory_limit": shared.OptionInfo(0, "Number of Lora networks to keep cached in memory", gr.Number, {"precision": 0}), })) @@ -121,3 +122,5 @@ def infotext_pasted(infotext, d): script_callbacks.on_infotext_pasted(infotext_pasted) + +shared.opts.onchange("lora_in_memory_limit", networks.purge_networks_from_memory) -- cgit v1.2.3 From 4fafc34e498130dcbb2d1a44fbc55fdba31e32d4 Mon Sep 17 00:00:00 2001 From: catboxanon <122327233+catboxanon@users.noreply.github.com> Date: Thu, 10 Aug 2023 23:42:58 -0400 Subject: Fix to make LoRA old method setting work --- extensions-builtin/Lora/networks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index bc722e90..7e3415ac 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -357,7 +357,7 @@ def network_forward(module, input, original_forward): if module is None: continue - y = module.forward(y, input) + y = module.forward(input, y) return y -- cgit v1.2.3 From bd4da4474bef5c9c1f690c62b971704ee73d2860 Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Sun, 13 Aug 2023 02:27:39 +0800 Subject: Add extra norm module into built-in lora ext refer to LyCORIS 1.9.0.dev6 add new option and module for training norm layer (Which is reported to be good for style) --- extensions-builtin/Lora/network.py | 7 ++- extensions-builtin/Lora/network_norm.py | 29 ++++++++++++ extensions-builtin/Lora/networks.py | 64 ++++++++++++++++++++++---- extensions-builtin/Lora/scripts/lora_script.py | 16 +++++++ 4 files changed, 105 insertions(+), 11 deletions(-) create mode 100644 extensions-builtin/Lora/network_norm.py (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/network.py b/extensions-builtin/Lora/network.py index 0a18d69e..b7b89061 100644 --- a/extensions-builtin/Lora/network.py +++ b/extensions-builtin/Lora/network.py @@ -133,7 +133,7 @@ class NetworkModule: return 1.0 - def finalize_updown(self, updown, orig_weight, output_shape): + def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): if self.bias is not None: updown = updown.reshape(self.bias.shape) updown += self.bias.to(orig_weight.device, dtype=orig_weight.dtype) @@ -145,7 +145,10 @@ class NetworkModule: if orig_weight.size().numel() == updown.size().numel(): updown = updown.reshape(orig_weight.shape) - return updown * self.calc_scale() * self.multiplier() + if ex_bias is None: + ex_bias = 0 + + return updown * self.calc_scale() * self.multiplier(), ex_bias * self.multiplier() def calc_updown(self, target): raise NotImplementedError() diff --git a/extensions-builtin/Lora/network_norm.py b/extensions-builtin/Lora/network_norm.py new file mode 100644 index 00000000..dab8b684 --- /dev/null +++ b/extensions-builtin/Lora/network_norm.py @@ -0,0 +1,29 @@ +import network + + +class ModuleTypeNorm(network.ModuleType): + def create_module(self, net: network.Network, weights: network.NetworkWeights): + if all(x in weights.w for x in ["w_norm", "b_norm"]): + return NetworkModuleNorm(net, weights) + + return None + + +class NetworkModuleNorm(network.NetworkModule): + def __init__(self, net: network.Network, weights: network.NetworkWeights): + super().__init__(net, weights) + print("NetworkModuleNorm") + + self.w_norm = weights.w.get("w_norm") + self.b_norm = weights.w.get("b_norm") + + def calc_updown(self, orig_weight): + output_shape = self.w_norm.shape + updown = self.w_norm.to(orig_weight.device, dtype=orig_weight.dtype) + + if self.b_norm is not None: + ex_bias = self.b_norm.to(orig_weight.device, dtype=orig_weight.dtype) + else: + ex_bias = None + + return self.finalize_updown(updown, orig_weight, output_shape, ex_bias) diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 7e3415ac..74cefe43 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -7,6 +7,7 @@ import network_hada import network_ia3 import network_lokr import network_full +import network_norm import torch from typing import Union @@ -19,6 +20,7 @@ module_types = [ network_ia3.ModuleTypeIa3(), network_lokr.ModuleTypeLokr(), network_full.ModuleTypeFull(), + network_norm.ModuleTypeNorm(), ] @@ -31,6 +33,8 @@ suffix_conversion = { "resnets": { "conv1": "in_layers_2", "conv2": "out_layers_3", + "norm1": "in_layers_0", + "norm2": "out_layers_0", "time_emb_proj": "emb_layers_1", "conv_shortcut": "skip_connection", } @@ -258,20 +262,25 @@ def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=No purge_networks_from_memory() -def network_restore_weights_from_backup(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.MultiheadAttention]): +def network_restore_weights_from_backup(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.GroupNorm, torch.nn.LayerNorm, torch.nn.MultiheadAttention]): weights_backup = getattr(self, "network_weights_backup", None) + bias_backup = getattr(self, "network_bias_backup", None) - if weights_backup is None: + if weights_backup is None and bias_backup is None: return - if isinstance(self, torch.nn.MultiheadAttention): - self.in_proj_weight.copy_(weights_backup[0]) - self.out_proj.weight.copy_(weights_backup[1]) - else: - self.weight.copy_(weights_backup) + if weights_backup is not None: + if isinstance(self, torch.nn.MultiheadAttention): + self.in_proj_weight.copy_(weights_backup[0]) + self.out_proj.weight.copy_(weights_backup[1]) + else: + self.weight.copy_(weights_backup) + if bias_backup is not None: + self.bias.copy_(bias_backup) -def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.MultiheadAttention]): + +def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.GroupNorm, torch.nn.LayerNorm, torch.nn.MultiheadAttention]): """ Applies the currently selected set of networks to the weights of torch layer self. If weights already have this particular set of networks applied, does nothing. @@ -294,6 +303,11 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn self.network_weights_backup = weights_backup + bias_backup = getattr(self, "network_bias_backup", None) + if bias_backup is None and getattr(self, 'bias', None) is not None: + bias_backup = self.bias.to(devices.cpu, copy=True) + self.network_bias_backup = bias_backup + if current_names != wanted_names: network_restore_weights_from_backup(self) @@ -301,13 +315,15 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn module = net.modules.get(network_layer_name, None) if module is not None and hasattr(self, 'weight'): with torch.no_grad(): - updown = module.calc_updown(self.weight) + updown, ex_bias = module.calc_updown(self.weight) if len(self.weight.shape) == 4 and self.weight.shape[1] == 9: # inpainting model. zero pad updown to make channel[1] 4 to 9 updown = torch.nn.functional.pad(updown, (0, 0, 0, 0, 0, 5)) self.weight += updown + if getattr(self, 'bias', None) is not None: + self.bias += ex_bias continue module_q = net.modules.get(network_layer_name + "_q_proj", None) @@ -397,6 +413,36 @@ def network_Conv2d_load_state_dict(self, *args, **kwargs): return torch.nn.Conv2d_load_state_dict_before_network(self, *args, **kwargs) +def network_GroupNorm_forward(self, input): + if shared.opts.lora_functional: + return network_forward(self, input, torch.nn.GroupNorm_forward_before_network) + + network_apply_weights(self) + + return torch.nn.GroupNorm_forward_before_network(self, input) + + +def network_GroupNorm_load_state_dict(self, *args, **kwargs): + network_reset_cached_weight(self) + + return torch.nn.GroupNorm_load_state_dict_before_network(self, *args, **kwargs) + + +def network_LayerNorm_forward(self, input): + if shared.opts.lora_functional: + return network_forward(self, input, torch.nn.LayerNorm_forward_before_network) + + network_apply_weights(self) + + return torch.nn.LayerNorm_forward_before_network(self, input) + + +def network_LayerNorm_load_state_dict(self, *args, **kwargs): + network_reset_cached_weight(self) + + return torch.nn.LayerNorm_load_state_dict_before_network(self, *args, **kwargs) + + def network_MultiheadAttention_forward(self, *args, **kwargs): network_apply_weights(self) diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py index 6ab8b6e7..dc307f8c 100644 --- a/extensions-builtin/Lora/scripts/lora_script.py +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -40,6 +40,18 @@ if not hasattr(torch.nn, 'Conv2d_forward_before_network'): if not hasattr(torch.nn, 'Conv2d_load_state_dict_before_network'): torch.nn.Conv2d_load_state_dict_before_network = torch.nn.Conv2d._load_from_state_dict +if not hasattr(torch.nn, 'GroupNorm_forward_before_network'): + torch.nn.GroupNorm_forward_before_network = torch.nn.GroupNorm.forward + +if not hasattr(torch.nn, 'GroupNorm_load_state_dict_before_network'): + torch.nn.GroupNorm_load_state_dict_before_network = torch.nn.GroupNorm._load_from_state_dict + +if not hasattr(torch.nn, 'LayerNorm_forward_before_network'): + torch.nn.LayerNorm_forward_before_network = torch.nn.LayerNorm.forward + +if not hasattr(torch.nn, 'LayerNorm_load_state_dict_before_network'): + torch.nn.LayerNorm_load_state_dict_before_network = torch.nn.LayerNorm._load_from_state_dict + if not hasattr(torch.nn, 'MultiheadAttention_forward_before_network'): torch.nn.MultiheadAttention_forward_before_network = torch.nn.MultiheadAttention.forward @@ -50,6 +62,10 @@ torch.nn.Linear.forward = networks.network_Linear_forward torch.nn.Linear._load_from_state_dict = networks.network_Linear_load_state_dict torch.nn.Conv2d.forward = networks.network_Conv2d_forward torch.nn.Conv2d._load_from_state_dict = networks.network_Conv2d_load_state_dict +torch.nn.GroupNorm.forward = networks.network_GroupNorm_forward +torch.nn.GroupNorm._load_from_state_dict = networks.network_GroupNorm_load_state_dict +torch.nn.LayerNorm.forward = networks.network_LayerNorm_forward +torch.nn.LayerNorm._load_from_state_dict = networks.network_LayerNorm_load_state_dict torch.nn.MultiheadAttention.forward = networks.network_MultiheadAttention_forward torch.nn.MultiheadAttention._load_from_state_dict = networks.network_MultiheadAttention_load_state_dict -- cgit v1.2.3 From a2b83050965a1a117f2762d3b5fa8b4841777e8f Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Sun, 13 Aug 2023 02:35:04 +0800 Subject: return None if no ex_bias --- extensions-builtin/Lora/network.py | 6 +++--- extensions-builtin/Lora/networks.py | 2 +- 2 files changed, 4 insertions(+), 4 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/network.py b/extensions-builtin/Lora/network.py index b7b89061..d8e8dfb7 100644 --- a/extensions-builtin/Lora/network.py +++ b/extensions-builtin/Lora/network.py @@ -145,10 +145,10 @@ class NetworkModule: if orig_weight.size().numel() == updown.size().numel(): updown = updown.reshape(orig_weight.shape) - if ex_bias is None: - ex_bias = 0 + if ex_bias is not None: + ex_bias = ex_bias * self.multiplier() - return updown * self.calc_scale() * self.multiplier(), ex_bias * self.multiplier() + return updown * self.calc_scale() * self.multiplier(), ex_bias def calc_updown(self, target): raise NotImplementedError() diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 74cefe43..ba621139 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -322,7 +322,7 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn updown = torch.nn.functional.pad(updown, (0, 0, 0, 0, 0, 5)) self.weight += updown - if getattr(self, 'bias', None) is not None: + if ex_bias is not None and getattr(self, 'bias', None) is not None: self.bias += ex_bias continue -- cgit v1.2.3 From 5881dcb8873b3f87b9c6545e9cb8d1d77023f4fe Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Sun, 13 Aug 2023 02:36:02 +0800 Subject: remove debug print --- extensions-builtin/Lora/network_norm.py | 1 - 1 file changed, 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/network_norm.py b/extensions-builtin/Lora/network_norm.py index dab8b684..ce450158 100644 --- a/extensions-builtin/Lora/network_norm.py +++ b/extensions-builtin/Lora/network_norm.py @@ -12,7 +12,6 @@ class ModuleTypeNorm(network.ModuleType): class NetworkModuleNorm(network.NetworkModule): def __init__(self, net: network.Network, weights: network.NetworkWeights): super().__init__(net, weights) - print("NetworkModuleNorm") self.w_norm = weights.w.get("w_norm") self.b_norm = weights.w.get("b_norm") -- cgit v1.2.3 From 7fa5ee54b15904bef6598800df76ba1291d44ec6 Mon Sep 17 00:00:00 2001 From: catboxanon <122327233+catboxanon@users.noreply.github.com> Date: Sun, 13 Aug 2023 02:32:54 -0400 Subject: Support search and display of hashes for all extra network items --- extensions-builtin/Lora/ui_extra_networks_lora.py | 3 ++- modules/ui_extra_networks_checkpoints.py | 1 + modules/ui_extra_networks_hypernets.py | 6 +++++- modules/ui_extra_networks_textual_inversion.py | 3 ++- modules/ui_extra_networks_user_metadata.py | 4 +++- 5 files changed, 13 insertions(+), 4 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/ui_extra_networks_lora.py b/extensions-builtin/Lora/ui_extra_networks_lora.py index 3629e5c0..55409a78 100644 --- a/extensions-builtin/Lora/ui_extra_networks_lora.py +++ b/extensions-builtin/Lora/ui_extra_networks_lora.py @@ -25,9 +25,10 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): item = { "name": name, "filename": lora_on_disk.filename, + "shorthash": lora_on_disk.shorthash, "preview": self.find_preview(path), "description": self.find_description(path), - "search_term": self.search_terms_from_path(lora_on_disk.filename), + "search_term": self.search_terms_from_path(lora_on_disk.filename) + " " + (lora_on_disk.hash or ""), "local_preview": f"{path}.{shared.opts.samples_format}", "metadata": lora_on_disk.metadata, "sort_keys": {'default': index, **self.get_sort_keys(lora_on_disk.filename)}, diff --git a/modules/ui_extra_networks_checkpoints.py b/modules/ui_extra_networks_checkpoints.py index 77885022..ebb5249f 100644 --- a/modules/ui_extra_networks_checkpoints.py +++ b/modules/ui_extra_networks_checkpoints.py @@ -19,6 +19,7 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage): return { "name": checkpoint.name_for_extra, "filename": checkpoint.filename, + "shorthash": checkpoint.shorthash, "preview": self.find_preview(path), "description": self.find_description(path), "search_term": self.search_terms_from_path(checkpoint.filename) + " " + (checkpoint.sha256 or ""), diff --git a/modules/ui_extra_networks_hypernets.py b/modules/ui_extra_networks_hypernets.py index 514a4562..4cedf085 100644 --- a/modules/ui_extra_networks_hypernets.py +++ b/modules/ui_extra_networks_hypernets.py @@ -2,6 +2,7 @@ import os from modules import shared, ui_extra_networks from modules.ui_extra_networks import quote_js +from modules.hashes import sha256_from_cache class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): @@ -14,13 +15,16 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): def create_item(self, name, index=None, enable_filter=True): full_path = shared.hypernetworks[name] path, ext = os.path.splitext(full_path) + sha256 = sha256_from_cache(full_path, f'hypernet/{name}') + shorthash = sha256[0:10] if sha256 else None return { "name": name, "filename": full_path, + "shorthash": shorthash, "preview": self.find_preview(path), "description": self.find_description(path), - "search_term": self.search_terms_from_path(path), + "search_term": self.search_terms_from_path(path) + " " + (sha256 or ""), "prompt": quote_js(f""), "local_preview": f"{path}.preview.{shared.opts.samples_format}", "sort_keys": {'default': index, **self.get_sort_keys(path + ext)}, diff --git a/modules/ui_extra_networks_textual_inversion.py b/modules/ui_extra_networks_textual_inversion.py index 73134698..55ef0ea7 100644 --- a/modules/ui_extra_networks_textual_inversion.py +++ b/modules/ui_extra_networks_textual_inversion.py @@ -19,9 +19,10 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): return { "name": name, "filename": embedding.filename, + "shorthash": embedding.shorthash, "preview": self.find_preview(path), "description": self.find_description(path), - "search_term": self.search_terms_from_path(embedding.filename), + "search_term": self.search_terms_from_path(embedding.filename) + " " + (embedding.hash or ""), "prompt": quote_js(embedding.name), "local_preview": f"{path}.preview.{shared.opts.samples_format}", "sort_keys": {'default': index, **self.get_sort_keys(embedding.filename)}, diff --git a/modules/ui_extra_networks_user_metadata.py b/modules/ui_extra_networks_user_metadata.py index cda471e4..b11622a1 100644 --- a/modules/ui_extra_networks_user_metadata.py +++ b/modules/ui_extra_networks_user_metadata.py @@ -93,11 +93,13 @@ class UserMetadataEditor: item = self.page.items.get(name, {}) try: filename = item["filename"] + shorthash = item.get("shorthash", None) stats = os.stat(filename) params = [ ('Filename: ', os.path.basename(filename)), ('File size: ', sysinfo.pretty_bytes(stats.st_size)), + ('Hash: ', shorthash), ('Modified: ', datetime.datetime.fromtimestamp(stats.st_mtime).strftime('%Y-%m-%d %H:%M')), ] @@ -115,7 +117,7 @@ class UserMetadataEditor: errors.display(e, f"reading metadata info for {name}") params = [] - table = '' + "".join(f"" for name, value in params) + '
{name}{value}
' + table = '' + "".join(f"" for name, value in params if value is not None) + '
{name}{value}
' return html.escape(name), user_metadata.get('description', ''), table, self.get_card_html(name), user_metadata.get('notes', '') -- cgit v1.2.3 From d8419762c1454ba51baa710d9ce8e762efc056ef Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 13 Aug 2023 15:07:37 +0300 Subject: Lora: output warnings in UI rather than fail for unfitting loras; switch to logging for error output in console --- extensions-builtin/Lora/extra_networks_lora.py | 12 ++++- extensions-builtin/Lora/networks.py | 61 ++++++++++++++++---------- extensions-builtin/Lora/scripts/lora_script.py | 6 +-- modules/processing.py | 13 +++--- 4 files changed, 58 insertions(+), 34 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/extra_networks_lora.py b/extensions-builtin/Lora/extra_networks_lora.py index ba2945c6..32e32cab 100644 --- a/extensions-builtin/Lora/extra_networks_lora.py +++ b/extensions-builtin/Lora/extra_networks_lora.py @@ -1,4 +1,4 @@ -from modules import extra_networks, shared +from modules import extra_networks, shared, sd_hijack import networks @@ -6,9 +6,14 @@ class ExtraNetworkLora(extra_networks.ExtraNetwork): def __init__(self): super().__init__('lora') + self.errors = {} + """mapping of network names to the number of errors the network had during operation""" + def activate(self, p, params_list): additional = shared.opts.sd_lora + self.errors.clear() + if additional != "None" and additional in networks.available_networks and not any(x for x in params_list if x.items[0] == additional): p.all_prompts = [x + f"" for x in p.all_prompts] params_list.append(extra_networks.ExtraNetworkParams(items=[additional, shared.opts.extra_networks_default_multiplier])) @@ -56,4 +61,7 @@ class ExtraNetworkLora(extra_networks.ExtraNetwork): p.extra_generation_params["Lora hashes"] = ", ".join(network_hashes) def deactivate(self, p): - pass + if self.errors: + p.comment("Networks with errors: " + ", ".join(f"{k} ({v})" for k, v in self.errors.items())) + + self.errors.clear() diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index ba621139..c252ed9e 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -1,3 +1,4 @@ +import logging import os import re @@ -194,7 +195,7 @@ def load_network(name, network_on_disk): net.modules[key] = net_module if keys_failed_to_match: - print(f"Failed to match keys when loading network {network_on_disk.filename}: {keys_failed_to_match}") + logging.debug(f"Network {network_on_disk.filename} didn't match keys: {keys_failed_to_match}") return net @@ -207,7 +208,6 @@ def purge_networks_from_memory(): devices.torch_gc() - def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=None): already_loaded = {} @@ -248,7 +248,7 @@ def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=No if net is None: failed_to_load_networks.append(name) - print(f"Couldn't find network with name {name}") + logging.info(f"Couldn't find network with name {name}") continue net.te_multiplier = te_multipliers[i] if te_multipliers else 1.0 @@ -257,7 +257,7 @@ def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=No loaded_networks.append(net) if failed_to_load_networks: - sd_hijack.model_hijack.comments.append("Failed to find networks: " + ", ".join(failed_to_load_networks)) + sd_hijack.model_hijack.comments.append("Networks not found: " + ", ".join(failed_to_load_networks)) purge_networks_from_memory() @@ -314,17 +314,22 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn for net in loaded_networks: module = net.modules.get(network_layer_name, None) if module is not None and hasattr(self, 'weight'): - with torch.no_grad(): - updown, ex_bias = module.calc_updown(self.weight) + try: + with torch.no_grad(): + updown, ex_bias = module.calc_updown(self.weight) - if len(self.weight.shape) == 4 and self.weight.shape[1] == 9: - # inpainting model. zero pad updown to make channel[1] 4 to 9 - updown = torch.nn.functional.pad(updown, (0, 0, 0, 0, 0, 5)) + if len(self.weight.shape) == 4 and self.weight.shape[1] == 9: + # inpainting model. zero pad updown to make channel[1] 4 to 9 + updown = torch.nn.functional.pad(updown, (0, 0, 0, 0, 0, 5)) - self.weight += updown - if ex_bias is not None and getattr(self, 'bias', None) is not None: - self.bias += ex_bias - continue + self.weight += updown + if ex_bias is not None and getattr(self, 'bias', None) is not None: + self.bias += ex_bias + except RuntimeError as e: + logging.debug(f"Network {net.name} layer {network_layer_name}: {e}") + extra_network_lora.errors[net.name] = extra_network_lora.errors.get(net.name, 0) + 1 + + continue module_q = net.modules.get(network_layer_name + "_q_proj", None) module_k = net.modules.get(network_layer_name + "_k_proj", None) @@ -332,21 +337,28 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn module_out = net.modules.get(network_layer_name + "_out_proj", None) if isinstance(self, torch.nn.MultiheadAttention) and module_q and module_k and module_v and module_out: - with torch.no_grad(): - updown_q = module_q.calc_updown(self.in_proj_weight) - updown_k = module_k.calc_updown(self.in_proj_weight) - updown_v = module_v.calc_updown(self.in_proj_weight) - updown_qkv = torch.vstack([updown_q, updown_k, updown_v]) - updown_out = module_out.calc_updown(self.out_proj.weight) - - self.in_proj_weight += updown_qkv - self.out_proj.weight += updown_out - continue + try: + with torch.no_grad(): + updown_q = module_q.calc_updown(self.in_proj_weight) + updown_k = module_k.calc_updown(self.in_proj_weight) + updown_v = module_v.calc_updown(self.in_proj_weight) + updown_qkv = torch.vstack([updown_q, updown_k, updown_v]) + updown_out = module_out.calc_updown(self.out_proj.weight) + + self.in_proj_weight += updown_qkv + self.out_proj.weight += updown_out + + except RuntimeError as e: + logging.debug(f"Network {net.name} layer {network_layer_name}: {e}") + extra_network_lora.errors[net.name] = extra_network_lora.errors.get(net.name, 0) + 1 + + continue if module is None: continue - print(f'failed to calculate network weights for layer {network_layer_name}') + logging.debug(f"Network {net.name} layer {network_layer_name}: couldn't find supported operation") + extra_network_lora.errors[net.name] = extra_network_lora.errors.get(net.name, 0) + 1 self.network_current_names = wanted_names @@ -519,6 +531,7 @@ def infotext_pasted(infotext, params): if added: params["Prompt"] += "\n" + "".join(added) +extra_network_lora = None available_networks = {} available_network_aliases = {} diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py index dc307f8c..4c6e774a 100644 --- a/extensions-builtin/Lora/scripts/lora_script.py +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -23,9 +23,9 @@ def unload(): def before_ui(): ui_extra_networks.register_page(ui_extra_networks_lora.ExtraNetworksPageLora()) - extra_network = extra_networks_lora.ExtraNetworkLora() - extra_networks.register_extra_network(extra_network) - extra_networks.register_extra_network_alias(extra_network, "lyco") + networks.extra_network_lora = extra_networks_lora.ExtraNetworkLora() + extra_networks.register_extra_network(networks.extra_network_lora) + extra_networks.register_extra_network_alias(networks.extra_network_lora, "lyco") if not hasattr(torch.nn, 'Linear_forward_before_network'): diff --git a/modules/processing.py b/modules/processing.py index 007a4e05..10749aa2 100755 --- a/modules/processing.py +++ b/modules/processing.py @@ -157,6 +157,7 @@ class StableDiffusionProcessing: cached_uc = [None, None] cached_c = [None, None] + comments: dict = None sampler: sd_samplers_common.Sampler | None = field(default=None, init=False) is_using_inpainting_conditioning: bool = field(default=False, init=False) paste_to: tuple | None = field(default=None, init=False) @@ -196,6 +197,8 @@ class StableDiffusionProcessing: if self.sampler_index is not None: print("sampler_index argument for StableDiffusionProcessing does not do anything; use sampler_name", file=sys.stderr) + self.comments = {} + self.sampler_noise_scheduler_override = None self.s_min_uncond = self.s_min_uncond if self.s_min_uncond is not None else opts.s_min_uncond self.s_churn = self.s_churn if self.s_churn is not None else opts.s_churn @@ -226,6 +229,9 @@ class StableDiffusionProcessing: def sd_model(self, value): pass + def comment(self, text): + self.comments[text] = 1 + def txt2img_image_conditioning(self, x, width=None, height=None): self.is_using_inpainting_conditioning = self.sd_model.model.conditioning_key in {'hybrid', 'concat'} @@ -429,7 +435,7 @@ class Processed: self.subseed = subseed self.subseed_strength = p.subseed_strength self.info = info - self.comments = comments + self.comments = "".join(f"{comment}\n" for comment in p.comments) self.width = p.width self.height = p.height self.sampler_name = p.sampler_name @@ -720,8 +726,6 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: modules.sd_hijack.model_hijack.apply_circular(p.tiling) modules.sd_hijack.model_hijack.clear_comments() - comments = {} - p.setup_prompts() if type(seed) == list: @@ -801,7 +805,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: p.setup_conds() for comment in model_hijack.comments: - comments[comment] = 1 + p.comment(comment) p.extra_generation_params.update(model_hijack.extra_generation_params) @@ -930,7 +934,6 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: images_list=output_images, seed=p.all_seeds[0], info=infotexts[0], - comments="".join(f"{comment}\n" for comment in comments), subseed=p.all_subseeds[0], index_of_first_image=index_of_first_image, infotexts=infotexts, -- cgit v1.2.3 From db40d26d086b6e980bd28ce7eb3008ec09b1f0e8 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 13 Aug 2023 16:38:10 +0300 Subject: linter --- extensions-builtin/Lora/extra_networks_lora.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/extra_networks_lora.py b/extensions-builtin/Lora/extra_networks_lora.py index 32e32cab..005ff32c 100644 --- a/extensions-builtin/Lora/extra_networks_lora.py +++ b/extensions-builtin/Lora/extra_networks_lora.py @@ -1,4 +1,4 @@ -from modules import extra_networks, shared, sd_hijack +from modules import extra_networks, shared import networks -- cgit v1.2.3 From d9cc27cb29926c9cc5dce331da8fbaf996cf4973 Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Mon, 14 Aug 2023 13:32:51 +0800 Subject: Fix MHA updown err and support ex-bias for no-bias layer --- extensions-builtin/Lora/networks.py | 37 +++++++++++++++++++++++++++++-------- 1 file changed, 29 insertions(+), 8 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index ba621139..1645b822 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -277,7 +277,15 @@ def network_restore_weights_from_backup(self: Union[torch.nn.Conv2d, torch.nn.Li self.weight.copy_(weights_backup) if bias_backup is not None: - self.bias.copy_(bias_backup) + if isinstance(self, torch.nn.MultiheadAttention): + self.out_proj.bias.copy_(bias_backup) + else: + self.bias.copy_(bias_backup) + else: + if isinstance(self, torch.nn.MultiheadAttention): + self.out_proj.bias = None + else: + self.bias = None def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.GroupNorm, torch.nn.LayerNorm, torch.nn.MultiheadAttention]): @@ -305,7 +313,12 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn bias_backup = getattr(self, "network_bias_backup", None) if bias_backup is None and getattr(self, 'bias', None) is not None: - bias_backup = self.bias.to(devices.cpu, copy=True) + if isinstance(self, torch.nn.MultiheadAttention) and self.out_proj.bias is not None: + bias_backup = self.out_proj.bias.to(devices.cpu, copy=True) + elif getattr(self, 'bias', None) is not None: + bias_backup = self.bias.to(devices.cpu, copy=True) + else: + bias_backup = None self.network_bias_backup = bias_backup if current_names != wanted_names: @@ -322,8 +335,11 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn updown = torch.nn.functional.pad(updown, (0, 0, 0, 0, 0, 5)) self.weight += updown - if ex_bias is not None and getattr(self, 'bias', None) is not None: - self.bias += ex_bias + if ex_bias is not None and hasattr(self, 'bias'): + if self.bias is None: + self.bias = torch.nn.Parameter(ex_bias) + else: + self.bias += ex_bias continue module_q = net.modules.get(network_layer_name + "_q_proj", None) @@ -333,14 +349,19 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn if isinstance(self, torch.nn.MultiheadAttention) and module_q and module_k and module_v and module_out: with torch.no_grad(): - updown_q = module_q.calc_updown(self.in_proj_weight) - updown_k = module_k.calc_updown(self.in_proj_weight) - updown_v = module_v.calc_updown(self.in_proj_weight) + updown_q, _ = module_q.calc_updown(self.in_proj_weight) + updown_k, _ = module_k.calc_updown(self.in_proj_weight) + updown_v, _ = module_v.calc_updown(self.in_proj_weight) updown_qkv = torch.vstack([updown_q, updown_k, updown_v]) - updown_out = module_out.calc_updown(self.out_proj.weight) + updown_out, ex_bias = module_out.calc_updown(self.out_proj.weight) self.in_proj_weight += updown_qkv self.out_proj.weight += updown_out + if ex_bias is not None: + if self.out_proj.bias is None: + self.out_proj.bias = torch.nn.Parameter(ex_bias) + else: + self.out_proj.bias += ex_bias continue if module is None: -- cgit v1.2.3 From f70ded89365f71d42b6a60a561e8fccfdd25c159 Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Mon, 14 Aug 2023 13:53:40 +0800 Subject: remove "if bias exist" check --- extensions-builtin/Lora/networks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 96d14344..22fdff4a 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -312,7 +312,7 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn self.network_weights_backup = weights_backup bias_backup = getattr(self, "network_bias_backup", None) - if bias_backup is None and getattr(self, 'bias', None) is not None: + if bias_backup is None: if isinstance(self, torch.nn.MultiheadAttention) and self.out_proj.bias is not None: bias_backup = self.out_proj.bias.to(devices.cpu, copy=True) elif getattr(self, 'bias', None) is not None: -- cgit v1.2.3 From aa57a89a21d7c0e59a2006a642c3838264c35051 Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Tue, 15 Aug 2023 23:41:46 +0800 Subject: full module with ex_bias --- extensions-builtin/Lora/network_full.py | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/network_full.py b/extensions-builtin/Lora/network_full.py index 109b4c2c..bf6930e9 100644 --- a/extensions-builtin/Lora/network_full.py +++ b/extensions-builtin/Lora/network_full.py @@ -14,9 +14,14 @@ class NetworkModuleFull(network.NetworkModule): super().__init__(net, weights) self.weight = weights.w.get("diff") + self.ex_bias = weights.w.get("diff_b") def calc_updown(self, orig_weight): output_shape = self.weight.shape updown = self.weight.to(orig_weight.device, dtype=orig_weight.dtype) + if self.ex_bias is not None: + ex_bias = self.ex_bias.to(orig_weight.device, dtype=orig_weight.dtype) + else: + ex_bias = None - return self.finalize_updown(updown, orig_weight, output_shape) + return self.finalize_updown(updown, orig_weight, output_shape, ex_bias) -- cgit v1.2.3 From f01682ee01e81e8ef84fd6fffe8f7aa17233285d Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Tue, 15 Aug 2023 19:23:27 +0300 Subject: store patches for Lora in a specialized module --- extensions-builtin/Lora/lora_patches.py | 31 +++++++++++++ extensions-builtin/Lora/networks.py | 32 +++++++------ extensions-builtin/Lora/scripts/lora_script.py | 52 ++------------------- modules/patches.py | 64 ++++++++++++++++++++++++++ 4 files changed, 118 insertions(+), 61 deletions(-) create mode 100644 extensions-builtin/Lora/lora_patches.py create mode 100644 modules/patches.py (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/lora_patches.py b/extensions-builtin/Lora/lora_patches.py new file mode 100644 index 00000000..b394d8e9 --- /dev/null +++ b/extensions-builtin/Lora/lora_patches.py @@ -0,0 +1,31 @@ +import torch + +import networks +from modules import patches + + +class LoraPatches: + def __init__(self): + self.Linear_forward = patches.patch(__name__, torch.nn.Linear, 'forward', networks.network_Linear_forward) + self.Linear_load_state_dict = patches.patch(__name__, torch.nn.Linear, '_load_from_state_dict', networks.network_Linear_load_state_dict) + self.Conv2d_forward = patches.patch(__name__, torch.nn.Conv2d, 'forward', networks.network_Conv2d_forward) + self.Conv2d_load_state_dict = patches.patch(__name__, torch.nn.Conv2d, '_load_from_state_dict', networks.network_Conv2d_load_state_dict) + self.GroupNorm_forward = patches.patch(__name__, torch.nn.GroupNorm, 'forward', networks.network_GroupNorm_forward) + self.GroupNorm_load_state_dict = patches.patch(__name__, torch.nn.GroupNorm, '_load_from_state_dict', networks.network_GroupNorm_load_state_dict) + self.LayerNorm_forward = patches.patch(__name__, torch.nn.LayerNorm, 'forward', networks.network_LayerNorm_forward) + self.LayerNorm_load_state_dict = patches.patch(__name__, torch.nn.LayerNorm, '_load_from_state_dict', networks.network_LayerNorm_load_state_dict) + self.MultiheadAttention_forward = patches.patch(__name__, torch.nn.MultiheadAttention, 'forward', networks.network_MultiheadAttention_forward) + self.MultiheadAttention_load_state_dict = patches.patch(__name__, torch.nn.MultiheadAttention, '_load_from_state_dict', networks.network_MultiheadAttention_load_state_dict) + + def undo(self): + self.Linear_forward = patches.undo(__name__, torch.nn.Linear, 'forward') + self.Linear_load_state_dict = patches.undo(__name__, torch.nn.Linear, '_load_from_state_dict') + self.Conv2d_forward = patches.undo(__name__, torch.nn.Conv2d, 'forward') + self.Conv2d_load_state_dict = patches.undo(__name__, torch.nn.Conv2d, '_load_from_state_dict') + self.GroupNorm_forward = patches.undo(__name__, torch.nn.GroupNorm, 'forward') + self.GroupNorm_load_state_dict = patches.undo(__name__, torch.nn.GroupNorm, '_load_from_state_dict') + self.LayerNorm_forward = patches.undo(__name__, torch.nn.LayerNorm, 'forward') + self.LayerNorm_load_state_dict = patches.undo(__name__, torch.nn.LayerNorm, '_load_from_state_dict') + self.MultiheadAttention_forward = patches.undo(__name__, torch.nn.MultiheadAttention, 'forward') + self.MultiheadAttention_load_state_dict = patches.undo(__name__, torch.nn.MultiheadAttention, '_load_from_state_dict') + diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 22fdff4a..9fca36b6 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -2,6 +2,7 @@ import logging import os import re +import lora_patches import network import network_lora import network_hada @@ -418,74 +419,74 @@ def network_reset_cached_weight(self: Union[torch.nn.Conv2d, torch.nn.Linear]): def network_Linear_forward(self, input): if shared.opts.lora_functional: - return network_forward(self, input, torch.nn.Linear_forward_before_network) + return network_forward(self, input, originals.Linear_forward) network_apply_weights(self) - return torch.nn.Linear_forward_before_network(self, input) + return originals.Linear_forward(self, input) def network_Linear_load_state_dict(self, *args, **kwargs): network_reset_cached_weight(self) - return torch.nn.Linear_load_state_dict_before_network(self, *args, **kwargs) + return originals.Linear_load_state_dict(self, *args, **kwargs) def network_Conv2d_forward(self, input): if shared.opts.lora_functional: - return network_forward(self, input, torch.nn.Conv2d_forward_before_network) + return network_forward(self, input, originals.Conv2d_forward) network_apply_weights(self) - return torch.nn.Conv2d_forward_before_network(self, input) + return originals.Conv2d_forward(self, input) def network_Conv2d_load_state_dict(self, *args, **kwargs): network_reset_cached_weight(self) - return torch.nn.Conv2d_load_state_dict_before_network(self, *args, **kwargs) + return originals.Conv2d_load_state_dict(self, *args, **kwargs) def network_GroupNorm_forward(self, input): if shared.opts.lora_functional: - return network_forward(self, input, torch.nn.GroupNorm_forward_before_network) + return network_forward(self, input, originals.GroupNorm_forward) network_apply_weights(self) - return torch.nn.GroupNorm_forward_before_network(self, input) + return originals.GroupNorm_forward(self, input) def network_GroupNorm_load_state_dict(self, *args, **kwargs): network_reset_cached_weight(self) - return torch.nn.GroupNorm_load_state_dict_before_network(self, *args, **kwargs) + return originals.GroupNorm_load_state_dict(self, *args, **kwargs) def network_LayerNorm_forward(self, input): if shared.opts.lora_functional: - return network_forward(self, input, torch.nn.LayerNorm_forward_before_network) + return network_forward(self, input, originals.LayerNorm_forward) network_apply_weights(self) - return torch.nn.LayerNorm_forward_before_network(self, input) + return originals.LayerNorm_forward(self, input) def network_LayerNorm_load_state_dict(self, *args, **kwargs): network_reset_cached_weight(self) - return torch.nn.LayerNorm_load_state_dict_before_network(self, *args, **kwargs) + return originals.LayerNorm_load_state_dict(self, *args, **kwargs) def network_MultiheadAttention_forward(self, *args, **kwargs): network_apply_weights(self) - return torch.nn.MultiheadAttention_forward_before_network(self, *args, **kwargs) + return originals.MultiheadAttention_forward(self, *args, **kwargs) def network_MultiheadAttention_load_state_dict(self, *args, **kwargs): network_reset_cached_weight(self) - return torch.nn.MultiheadAttention_load_state_dict_before_network(self, *args, **kwargs) + return originals.MultiheadAttention_load_state_dict(self, *args, **kwargs) def list_available_networks(): @@ -552,6 +553,9 @@ def infotext_pasted(infotext, params): if added: params["Prompt"] += "\n" + "".join(added) + +originals: lora_patches.LoraPatches = None + extra_network_lora = None available_networks = {} diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py index 4c6e774a..546fb55e 100644 --- a/extensions-builtin/Lora/scripts/lora_script.py +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -7,17 +7,14 @@ from fastapi import FastAPI import network import networks import lora # noqa:F401 +import lora_patches import extra_networks_lora import ui_extra_networks_lora -from modules import script_callbacks, ui_extra_networks, extra_networks, shared +from modules import script_callbacks, ui_extra_networks, extra_networks, shared, patches + def unload(): - torch.nn.Linear.forward = torch.nn.Linear_forward_before_network - torch.nn.Linear._load_from_state_dict = torch.nn.Linear_load_state_dict_before_network - torch.nn.Conv2d.forward = torch.nn.Conv2d_forward_before_network - torch.nn.Conv2d._load_from_state_dict = torch.nn.Conv2d_load_state_dict_before_network - torch.nn.MultiheadAttention.forward = torch.nn.MultiheadAttention_forward_before_network - torch.nn.MultiheadAttention._load_from_state_dict = torch.nn.MultiheadAttention_load_state_dict_before_network + networks.originals.undo() def before_ui(): @@ -28,46 +25,7 @@ def before_ui(): extra_networks.register_extra_network_alias(networks.extra_network_lora, "lyco") -if not hasattr(torch.nn, 'Linear_forward_before_network'): - torch.nn.Linear_forward_before_network = torch.nn.Linear.forward - -if not hasattr(torch.nn, 'Linear_load_state_dict_before_network'): - torch.nn.Linear_load_state_dict_before_network = torch.nn.Linear._load_from_state_dict - -if not hasattr(torch.nn, 'Conv2d_forward_before_network'): - torch.nn.Conv2d_forward_before_network = torch.nn.Conv2d.forward - -if not hasattr(torch.nn, 'Conv2d_load_state_dict_before_network'): - torch.nn.Conv2d_load_state_dict_before_network = torch.nn.Conv2d._load_from_state_dict - -if not hasattr(torch.nn, 'GroupNorm_forward_before_network'): - torch.nn.GroupNorm_forward_before_network = torch.nn.GroupNorm.forward - -if not hasattr(torch.nn, 'GroupNorm_load_state_dict_before_network'): - torch.nn.GroupNorm_load_state_dict_before_network = torch.nn.GroupNorm._load_from_state_dict - -if not hasattr(torch.nn, 'LayerNorm_forward_before_network'): - torch.nn.LayerNorm_forward_before_network = torch.nn.LayerNorm.forward - -if not hasattr(torch.nn, 'LayerNorm_load_state_dict_before_network'): - torch.nn.LayerNorm_load_state_dict_before_network = torch.nn.LayerNorm._load_from_state_dict - -if not hasattr(torch.nn, 'MultiheadAttention_forward_before_network'): - torch.nn.MultiheadAttention_forward_before_network = torch.nn.MultiheadAttention.forward - -if not hasattr(torch.nn, 'MultiheadAttention_load_state_dict_before_network'): - torch.nn.MultiheadAttention_load_state_dict_before_network = torch.nn.MultiheadAttention._load_from_state_dict - -torch.nn.Linear.forward = networks.network_Linear_forward -torch.nn.Linear._load_from_state_dict = networks.network_Linear_load_state_dict -torch.nn.Conv2d.forward = networks.network_Conv2d_forward -torch.nn.Conv2d._load_from_state_dict = networks.network_Conv2d_load_state_dict -torch.nn.GroupNorm.forward = networks.network_GroupNorm_forward -torch.nn.GroupNorm._load_from_state_dict = networks.network_GroupNorm_load_state_dict -torch.nn.LayerNorm.forward = networks.network_LayerNorm_forward -torch.nn.LayerNorm._load_from_state_dict = networks.network_LayerNorm_load_state_dict -torch.nn.MultiheadAttention.forward = networks.network_MultiheadAttention_forward -torch.nn.MultiheadAttention._load_from_state_dict = networks.network_MultiheadAttention_load_state_dict +networks.originals = lora_patches.LoraPatches() script_callbacks.on_model_loaded(networks.assign_network_names_to_compvis_modules) script_callbacks.on_script_unloaded(unload) diff --git a/modules/patches.py b/modules/patches.py new file mode 100644 index 00000000..348235e7 --- /dev/null +++ b/modules/patches.py @@ -0,0 +1,64 @@ +from collections import defaultdict + + +def patch(key, obj, field, replacement): + """Replaces a function in a module or a class. + + Also stores the original function in this module, possible to be retrieved via original(key, obj, field). + If the function is already replaced by this caller (key), an exception is raised -- use undo() before that. + + Arguments: + key: identifying information for who is doing the replacement. You can use __name__. + obj: the module or the class + field: name of the function as a string + replacement: the new function + + Returns: + the original function + """ + + patch_key = (obj, field) + if patch_key in originals[key]: + raise RuntimeError(f"patch for {field} is already applied") + + original_func = getattr(obj, field) + originals[key][patch_key] = original_func + + setattr(obj, field, replacement) + + return original_func + + +def undo(key, obj, field): + """Undoes the peplacement by the patch(). + + If the function is not replaced, raises an exception. + + Arguments: + key: identifying information for who is doing the replacement. You can use __name__. + obj: the module or the class + field: name of the function as a string + + Returns: + Always None + """ + + patch_key = (obj, field) + + if patch_key not in originals[key]: + raise RuntimeError(f"there is no patch for {field} to undo") + + original_func = originals[key].pop(patch_key) + setattr(obj, field, original_func) + + return None + + +def original(key, obj, field): + """Returns the original function for the patch created by the patch() function""" + patch_key = (obj, field) + + return originals[key].get(patch_key, None) + + +originals = defaultdict(dict) -- cgit v1.2.3 From 85fcb7b8dfe7b3dd06931943f095c77f1043dc25 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Tue, 15 Aug 2023 19:24:55 +0300 Subject: lint --- extensions-builtin/Lora/scripts/lora_script.py | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py index 546fb55e..ef23968c 100644 --- a/extensions-builtin/Lora/scripts/lora_script.py +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -1,6 +1,5 @@ import re -import torch import gradio as gr from fastapi import FastAPI @@ -10,7 +9,7 @@ import lora # noqa:F401 import lora_patches import extra_networks_lora import ui_extra_networks_lora -from modules import script_callbacks, ui_extra_networks, extra_networks, shared, patches +from modules import script_callbacks, ui_extra_networks, extra_networks, shared def unload(): -- cgit v1.2.3 From 86221269f98ef9b21a6e6c9d04b86e2fb5cb33d3 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Wed, 16 Aug 2023 09:55:35 +0300 Subject: RAM optimization round 2 --- extensions-builtin/Lora/networks.py | 5 +++- modules/sd_disable_initialization.py | 51 +++++++++++++++++++++++++++++++----- 2 files changed, 48 insertions(+), 8 deletions(-) (limited to 'extensions-builtin/Lora') diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 9fca36b6..96f935b2 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -304,7 +304,10 @@ def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn wanted_names = tuple((x.name, x.te_multiplier, x.unet_multiplier, x.dyn_dim) for x in loaded_networks) weights_backup = getattr(self, "network_weights_backup", None) - if weights_backup is None: + if weights_backup is None and wanted_names != (): + if current_names != (): + raise RuntimeError("no backup weights found and current weights are not unchanged") + if isinstance(self, torch.nn.MultiheadAttention): weights_backup = (self.in_proj_weight.to(devices.cpu, copy=True), self.out_proj.weight.to(devices.cpu, copy=True)) else: diff --git a/modules/sd_disable_initialization.py b/modules/sd_disable_initialization.py index 695c5736..719eeb93 100644 --- a/modules/sd_disable_initialization.py +++ b/modules/sd_disable_initialization.py @@ -168,22 +168,59 @@ class LoadStateDictOnMeta(ReplaceHelper): device = self.device def load_from_state_dict(original, self, state_dict, prefix, *args, **kwargs): - params = [(name, param) for name, param in self._parameters.items() if param is not None and param.is_meta] + used_param_keys = [] + + for name, param in self._parameters.items(): + if param is None: + continue + + key = prefix + name + sd_param = sd.pop(key, None) + if sd_param is not None: + state_dict[key] = sd_param + used_param_keys.append(key) - for name, param in params: if param.is_meta: - self._parameters[name] = torch.nn.parameter.Parameter(torch.zeros_like(param, device=device), requires_grad=param.requires_grad) + dtype = sd_param.dtype if sd_param is not None else param.dtype + self._parameters[name] = torch.nn.parameter.Parameter(torch.zeros_like(param, device=device, dtype=dtype), requires_grad=param.requires_grad) + + for name in self._buffers: + key = prefix + name + + sd_param = sd.pop(key, None) + if sd_param is not None: + state_dict[key] = sd_param + used_param_keys.append(key) original(self, state_dict, prefix, *args, **kwargs) - for name, _ in params: - key = prefix + name - if key in sd: - del sd[key] + for key in used_param_keys: + state_dict.pop(key, None) + + def load_state_dict(original, self, state_dict, strict=True): + """torch makes a lot of copies of the dictionary with weights, so just deleting entries from state_dict does not help + because the same values are stored in multiple copies of the dict. The trick used here is to give torch a dict with + all weights on meta device, i.e. deleted, and then it doesn't matter how many copies torch makes. + + In _load_from_state_dict, the correct weight will be obtained from a single dict with the right weights (sd). + + The dangerous thing about this is if _load_from_state_dict is not called, (if some exotic module overloads + the function and does not call the original) the state dict will just fail to load because weights + would be on the meta device. + """ + + if state_dict == sd: + state_dict = {k: v.to(device="meta", dtype=v.dtype) for k, v in state_dict.items()} + + original(self, state_dict, strict=strict) + module_load_state_dict = self.replace(torch.nn.Module, 'load_state_dict', lambda *args, **kwargs: load_state_dict(module_load_state_dict, *args, **kwargs)) + module_load_from_state_dict = self.replace(torch.nn.Module, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(module_load_from_state_dict, *args, **kwargs)) linear_load_from_state_dict = self.replace(torch.nn.Linear, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(linear_load_from_state_dict, *args, **kwargs)) conv2d_load_from_state_dict = self.replace(torch.nn.Conv2d, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(conv2d_load_from_state_dict, *args, **kwargs)) mha_load_from_state_dict = self.replace(torch.nn.MultiheadAttention, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(mha_load_from_state_dict, *args, **kwargs)) + layer_norm_load_from_state_dict = self.replace(torch.nn.LayerNorm, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(layer_norm_load_from_state_dict, *args, **kwargs)) + group_norm_load_from_state_dict = self.replace(torch.nn.GroupNorm, '_load_from_state_dict', lambda *args, **kwargs: load_from_state_dict(group_norm_load_from_state_dict, *args, **kwargs)) def __exit__(self, exc_type, exc_val, exc_tb): self.restore() -- cgit v1.2.3