From 0c3feb202c5714abd50d879c1db2cd9a71ce93e3 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Tue, 10 Jan 2023 14:08:29 +0300 Subject: [PATCH 1/3] disable torch weight initialization and CLIP downloading/reading checkpoint to speedup creating sd model from config --- modules/sd_disable_initialization.py | 44 ++++++++++++++++++++++++++++ modules/sd_models.py | 5 ++-- 2 files changed, 47 insertions(+), 2 deletions(-) create mode 100644 modules/sd_disable_initialization.py diff --git a/modules/sd_disable_initialization.py b/modules/sd_disable_initialization.py new file mode 100644 index 00000000..c9a3b5e4 --- /dev/null +++ b/modules/sd_disable_initialization.py @@ -0,0 +1,44 @@ +import ldm.modules.encoders.modules +import open_clip +import torch + + +class DisableInitialization: + """ + When an object of this class enters a `with` block, it starts preventing torch's layer initialization + functions from working, and changes CLIP and OpenCLIP to not download model weights. When it leaves, + reverts everything to how it was. + + Use like this: + ``` + with DisableInitialization(): + do_things() + ``` + """ + + def __enter__(self): + def do_nothing(*args, **kwargs): + pass + + def create_model_and_transforms_without_pretrained(*args, pretrained=None, **kwargs): + return self.create_model_and_transforms(*args, pretrained=None, **kwargs) + + def CLIPTextModel_from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs): + return self.CLIPTextModel_from_pretrained(None, *model_args, config=pretrained_model_name_or_path, state_dict={}, **kwargs) + + self.init_kaiming_uniform = torch.nn.init.kaiming_uniform_ + self.init_no_grad_normal = torch.nn.init._no_grad_normal_ + self.create_model_and_transforms = open_clip.create_model_and_transforms + self.CLIPTextModel_from_pretrained = ldm.modules.encoders.modules.CLIPTextModel.from_pretrained + + torch.nn.init.kaiming_uniform_ = do_nothing + torch.nn.init._no_grad_normal_ = do_nothing + open_clip.create_model_and_transforms = create_model_and_transforms_without_pretrained + ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = CLIPTextModel_from_pretrained + + def __exit__(self, exc_type, exc_val, exc_tb): + torch.nn.init.kaiming_uniform_ = self.init_kaiming_uniform + torch.nn.init._no_grad_normal_ = self.init_no_grad_normal + open_clip.create_model_and_transforms = self.create_model_and_transforms + ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = self.CLIPTextModel_from_pretrained + diff --git a/modules/sd_models.py b/modules/sd_models.py index 0a6d55ca..ee241032 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -13,7 +13,7 @@ import ldm.modules.midas as midas from ldm.util import instantiate_from_config -from modules import shared, modelloader, devices, script_callbacks, sd_vae +from modules import shared, modelloader, devices, script_callbacks, sd_vae, sd_disable_initialization from modules.paths import models_path from modules.sd_hijack_inpainting import do_inpainting_hijack, should_hijack_inpainting @@ -319,7 +319,8 @@ def load_model(checkpoint_info=None): if shared.cmd_opts.no_half: sd_config.model.params.unet_config.params.use_fp16 = False - sd_model = instantiate_from_config(sd_config.model) + with sd_disable_initialization.DisableInitialization(): + sd_model = instantiate_from_config(sd_config.model) load_model_weights(sd_model, checkpoint_info) From ce3f639ec8758ce2bc90483336361d2dc25acd3a Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Tue, 10 Jan 2023 16:51:04 +0300 Subject: [PATCH 2/3] add more stuff to ignore when creating model from config prevent .vae.safetensors files from being listed as stable diffusion models --- modules/modelloader.py | 4 +++- modules/sd_disable_initialization.py | 29 +++++++++++++++++++++---- modules/sd_models.py | 32 ++++++++++++++++++++++++---- 3 files changed, 56 insertions(+), 9 deletions(-) diff --git a/modules/modelloader.py b/modules/modelloader.py index 6a1a7ac8..e9aa514e 100644 --- a/modules/modelloader.py +++ b/modules/modelloader.py @@ -10,7 +10,7 @@ from modules.upscaler import Upscaler from modules.paths import script_path, models_path -def load_models(model_path: str, model_url: str = None, command_path: str = None, ext_filter=None, download_name=None) -> list: +def load_models(model_path: str, model_url: str = None, command_path: str = None, ext_filter=None, download_name=None, ext_blacklist=None) -> list: """ A one-and done loader to try finding the desired models in specified directories. @@ -45,6 +45,8 @@ def load_models(model_path: str, model_url: str = None, command_path: str = None full_path = file if os.path.isdir(full_path): continue + if ext_blacklist is not None and any([full_path.endswith(x) for x in ext_blacklist]): + continue if len(ext_filter) != 0: model_name, extension = os.path.splitext(file) if extension not in ext_filter: diff --git a/modules/sd_disable_initialization.py b/modules/sd_disable_initialization.py index c9a3b5e4..9942bd7e 100644 --- a/modules/sd_disable_initialization.py +++ b/modules/sd_disable_initialization.py @@ -1,15 +1,19 @@ import ldm.modules.encoders.modules import open_clip import torch +import transformers.utils.hub class DisableInitialization: """ - When an object of this class enters a `with` block, it starts preventing torch's layer initialization - functions from working, and changes CLIP and OpenCLIP to not download model weights. When it leaves, - reverts everything to how it was. + When an object of this class enters a `with` block, it starts: + - preventing torch's layer initialization functions from working + - changes CLIP and OpenCLIP to not download model weights + - changes CLIP to not make requests to check if there is a new version of a file you already have - Use like this: + When it leaves the block, it reverts everything to how it was before. + + Use it like this: ``` with DisableInitialization(): do_things() @@ -26,19 +30,36 @@ class DisableInitialization: def CLIPTextModel_from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs): return self.CLIPTextModel_from_pretrained(None, *model_args, config=pretrained_model_name_or_path, state_dict={}, **kwargs) + def transformers_utils_hub_get_from_cache(url, *args, local_files_only=False, **kwargs): + + # this file is always 404, prevent making request + if url == 'https://huggingface.co/openai/clip-vit-large-patch14/resolve/main/added_tokens.json': + raise transformers.utils.hub.EntryNotFoundError + + try: + return self.transformers_utils_hub_get_from_cache(url, *args, local_files_only=True, **kwargs) + except Exception as e: + return self.transformers_utils_hub_get_from_cache(url, *args, local_files_only=False, **kwargs) + self.init_kaiming_uniform = torch.nn.init.kaiming_uniform_ self.init_no_grad_normal = torch.nn.init._no_grad_normal_ + self.init_no_grad_uniform_ = torch.nn.init._no_grad_uniform_ self.create_model_and_transforms = open_clip.create_model_and_transforms self.CLIPTextModel_from_pretrained = ldm.modules.encoders.modules.CLIPTextModel.from_pretrained + self.transformers_utils_hub_get_from_cache = transformers.utils.hub.get_from_cache torch.nn.init.kaiming_uniform_ = do_nothing torch.nn.init._no_grad_normal_ = do_nothing + torch.nn.init._no_grad_uniform_ = do_nothing open_clip.create_model_and_transforms = create_model_and_transforms_without_pretrained ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = CLIPTextModel_from_pretrained + transformers.utils.hub.get_from_cache = transformers_utils_hub_get_from_cache def __exit__(self, exc_type, exc_val, exc_tb): torch.nn.init.kaiming_uniform_ = self.init_kaiming_uniform torch.nn.init._no_grad_normal_ = self.init_no_grad_normal + torch.nn.init._no_grad_uniform_ = self.init_no_grad_uniform_ open_clip.create_model_and_transforms = self.create_model_and_transforms ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = self.CLIPTextModel_from_pretrained + transformers.utils.hub.get_from_cache = self.transformers_utils_hub_get_from_cache diff --git a/modules/sd_models.py b/modules/sd_models.py index ee241032..1bb9088b 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -2,6 +2,7 @@ import collections import os.path import sys import gc +import time from collections import namedtuple import torch import re @@ -61,7 +62,7 @@ def find_checkpoint_config(info): def list_models(): checkpoints_list.clear() - model_list = modelloader.load_models(model_path=model_path, command_path=shared.cmd_opts.ckpt_dir, ext_filter=[".ckpt", ".safetensors"]) + model_list = modelloader.load_models(model_path=model_path, command_path=shared.cmd_opts.ckpt_dir, ext_filter=[".ckpt", ".safetensors"], ext_blacklist=[".vae.safetensors"]) def modeltitle(path, shorthash): abspath = os.path.abspath(path) @@ -288,6 +289,17 @@ def enable_midas_autodownload(): midas.api.load_model = load_model_wrapper +class Timer: + def __init__(self): + self.start = time.time() + + def elapsed(self): + end = time.time() + res = end - self.start + self.start = end + return res + + def load_model(checkpoint_info=None): from modules import lowvram, sd_hijack checkpoint_info = checkpoint_info or select_checkpoint() @@ -319,11 +331,17 @@ def load_model(checkpoint_info=None): if shared.cmd_opts.no_half: sd_config.model.params.unet_config.params.use_fp16 = False + timer = Timer() + with sd_disable_initialization.DisableInitialization(): sd_model = instantiate_from_config(sd_config.model) + elapsed_create = timer.elapsed() + load_model_weights(sd_model, checkpoint_info) + elapsed_load_weights = timer.elapsed() + if shared.cmd_opts.lowvram or shared.cmd_opts.medvram: lowvram.setup_for_low_vram(sd_model, shared.cmd_opts.medvram) else: @@ -338,7 +356,9 @@ def load_model(checkpoint_info=None): script_callbacks.model_loaded_callback(sd_model) - print("Model loaded.") + elapsed_the_rest = timer.elapsed() + + print(f"Model loaded in {elapsed_create + elapsed_load_weights + elapsed_the_rest:.1f}s ({elapsed_create:.1f}s create model, {elapsed_load_weights:.1f}s load weights).") return sd_model @@ -349,7 +369,7 @@ def reload_model_weights(sd_model=None, info=None): if not sd_model: sd_model = shared.sd_model - if sd_model is None: # previous model load failed + if sd_model is None: # previous model load failed current_checkpoint_info = None else: current_checkpoint_info = sd_model.sd_checkpoint_info @@ -371,6 +391,8 @@ def reload_model_weights(sd_model=None, info=None): sd_hijack.model_hijack.undo_hijack(sd_model) + timer = Timer() + try: load_model_weights(sd_model, checkpoint_info) except Exception as e: @@ -384,6 +406,8 @@ def reload_model_weights(sd_model=None, info=None): if not shared.cmd_opts.lowvram and not shared.cmd_opts.medvram: sd_model.to(devices.device) - print("Weights loaded.") + elapsed = timer.elapsed() + + print(f"Weights loaded in {elapsed:.1f}s.") return sd_model From 0f8603a55988d22616b17140e6c4a7e9d0736af5 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Tue, 10 Jan 2023 17:46:59 +0300 Subject: [PATCH 3/3] add support for transformers==4.25.1 add fallback for when quick model creation fails --- modules/sd_disable_initialization.py | 42 ++++++++++++++++++++++++---- modules/sd_models.py | 8 ++++-- 2 files changed, 42 insertions(+), 8 deletions(-) diff --git a/modules/sd_disable_initialization.py b/modules/sd_disable_initialization.py index 9942bd7e..088ac24b 100644 --- a/modules/sd_disable_initialization.py +++ b/modules/sd_disable_initialization.py @@ -30,30 +30,53 @@ class DisableInitialization: def CLIPTextModel_from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs): return self.CLIPTextModel_from_pretrained(None, *model_args, config=pretrained_model_name_or_path, state_dict={}, **kwargs) - def transformers_utils_hub_get_from_cache(url, *args, local_files_only=False, **kwargs): + def transformers_modeling_utils_load_pretrained_model(*args, **kwargs): + args = args[0:3] + ('/', ) + args[4:] # resolved_archive_file; must set it to something to prevent what seems to be a bug + return self.transformers_modeling_utils_load_pretrained_model(*args, **kwargs) + + def transformers_utils_hub_get_file_from_cache(original, url, *args, **kwargs): # this file is always 404, prevent making request if url == 'https://huggingface.co/openai/clip-vit-large-patch14/resolve/main/added_tokens.json': raise transformers.utils.hub.EntryNotFoundError try: - return self.transformers_utils_hub_get_from_cache(url, *args, local_files_only=True, **kwargs) + return original(url, *args, local_files_only=True, **kwargs) except Exception as e: - return self.transformers_utils_hub_get_from_cache(url, *args, local_files_only=False, **kwargs) + return original(url, *args, local_files_only=False, **kwargs) + + def transformers_utils_hub_get_from_cache(url, *args, local_files_only=False, **kwargs): + return transformers_utils_hub_get_file_from_cache(self.transformers_utils_hub_get_from_cache, url, *args, **kwargs) + + def transformers_tokenization_utils_base_cached_file(url, *args, local_files_only=False, **kwargs): + return transformers_utils_hub_get_file_from_cache(self.transformers_tokenization_utils_base_cached_file, url, *args, **kwargs) + + def transformers_configuration_utils_cached_file(url, *args, local_files_only=False, **kwargs): + return transformers_utils_hub_get_file_from_cache(self.transformers_configuration_utils_cached_file, url, *args, **kwargs) self.init_kaiming_uniform = torch.nn.init.kaiming_uniform_ self.init_no_grad_normal = torch.nn.init._no_grad_normal_ self.init_no_grad_uniform_ = torch.nn.init._no_grad_uniform_ self.create_model_and_transforms = open_clip.create_model_and_transforms self.CLIPTextModel_from_pretrained = ldm.modules.encoders.modules.CLIPTextModel.from_pretrained - self.transformers_utils_hub_get_from_cache = transformers.utils.hub.get_from_cache + self.transformers_modeling_utils_load_pretrained_model = getattr(transformers.modeling_utils.PreTrainedModel, '_load_pretrained_model', None) + self.transformers_tokenization_utils_base_cached_file = getattr(transformers.tokenization_utils_base, 'cached_file', None) + self.transformers_configuration_utils_cached_file = getattr(transformers.configuration_utils, 'cached_file', None) + self.transformers_utils_hub_get_from_cache = getattr(transformers.utils.hub, 'get_from_cache', None) torch.nn.init.kaiming_uniform_ = do_nothing torch.nn.init._no_grad_normal_ = do_nothing torch.nn.init._no_grad_uniform_ = do_nothing open_clip.create_model_and_transforms = create_model_and_transforms_without_pretrained ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = CLIPTextModel_from_pretrained - transformers.utils.hub.get_from_cache = transformers_utils_hub_get_from_cache + if self.transformers_modeling_utils_load_pretrained_model is not None: + transformers.modeling_utils.PreTrainedModel._load_pretrained_model = transformers_modeling_utils_load_pretrained_model + if self.transformers_tokenization_utils_base_cached_file is not None: + transformers.tokenization_utils_base.cached_file = transformers_tokenization_utils_base_cached_file + if self.transformers_configuration_utils_cached_file is not None: + transformers.configuration_utils.cached_file = transformers_configuration_utils_cached_file + if self.transformers_utils_hub_get_from_cache is not None: + transformers.utils.hub.get_from_cache = transformers_utils_hub_get_from_cache def __exit__(self, exc_type, exc_val, exc_tb): torch.nn.init.kaiming_uniform_ = self.init_kaiming_uniform @@ -61,5 +84,12 @@ class DisableInitialization: torch.nn.init._no_grad_uniform_ = self.init_no_grad_uniform_ open_clip.create_model_and_transforms = self.create_model_and_transforms ldm.modules.encoders.modules.CLIPTextModel.from_pretrained = self.CLIPTextModel_from_pretrained - transformers.utils.hub.get_from_cache = self.transformers_utils_hub_get_from_cache + if self.transformers_modeling_utils_load_pretrained_model is not None: + transformers.modeling_utils.PreTrainedModel._load_pretrained_model = self.transformers_modeling_utils_load_pretrained_model + if self.transformers_tokenization_utils_base_cached_file is not None: + transformers.utils.hub.cached_file = self.transformers_tokenization_utils_base_cached_file + if self.transformers_configuration_utils_cached_file is not None: + transformers.utils.hub.cached_file = self.transformers_configuration_utils_cached_file + if self.transformers_utils_hub_get_from_cache is not None: + transformers.utils.hub.get_from_cache = self.transformers_utils_hub_get_from_cache diff --git a/modules/sd_models.py b/modules/sd_models.py index 1bb9088b..b5bc12f0 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -14,7 +14,7 @@ import ldm.modules.midas as midas from ldm.util import instantiate_from_config -from modules import shared, modelloader, devices, script_callbacks, sd_vae, sd_disable_initialization +from modules import shared, modelloader, devices, script_callbacks, sd_vae, sd_disable_initialization, errors from modules.paths import models_path from modules.sd_hijack_inpainting import do_inpainting_hijack, should_hijack_inpainting @@ -333,7 +333,11 @@ def load_model(checkpoint_info=None): timer = Timer() - with sd_disable_initialization.DisableInitialization(): + try: + with sd_disable_initialization.DisableInitialization(): + sd_model = instantiate_from_config(sd_config.model) + except Exception as e: + print('Failed to create model quickly; will retry using slow method.', file=sys.stderr) sd_model = instantiate_from_config(sd_config.model) elapsed_create = timer.elapsed()