aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--README.md23
-rw-r--r--launch.py17
-rw-r--r--modules/api/api.py10
-rw-r--r--modules/deepbooru.py258
-rw-r--r--modules/deepbooru_model.py676
-rw-r--r--modules/images.py2
-rw-r--r--modules/lowvram.py8
-rw-r--r--modules/paths.py2
-rw-r--r--modules/processing.py6
-rw-r--r--modules/sd_hijack.py294
-rw-r--r--modules/sd_hijack_clip.py301
-rw-r--r--modules/sd_hijack_inpainting.py20
-rw-r--r--modules/sd_hijack_open_clip.py37
-rw-r--r--modules/sd_samplers.py14
-rw-r--r--modules/shared.py43
-rw-r--r--modules/textual_inversion/preprocess.py12
-rw-r--r--modules/textual_inversion/textual_inversion.py7
-rw-r--r--modules/ui.py38
-rw-r--r--modules/ui_tempdir.py62
-rw-r--r--requirements.txt1
-rw-r--r--requirements_versions.txt1
-rw-r--r--v1-inference.yaml70
-rw-r--r--webui.py19
23 files changed, 1374 insertions, 547 deletions
diff --git a/README.md b/README.md
index 33508f31..8a4ffade 100644
--- a/README.md
+++ b/README.md
@@ -70,7 +70,7 @@ Check the [custom scripts](https://github.com/AUTOMATIC1111/stable-diffusion-web
- separate prompts using uppercase `AND`
- also supports weights for prompts: `a cat :1.2 AND a dog AND a penguin :2.2`
- No token limit for prompts (original stable diffusion lets you use up to 75 tokens)
-- DeepDanbooru integration, creates danbooru style tags for anime prompts (add --deepdanbooru to commandline args)
+- DeepDanbooru integration, creates danbooru style tags for anime prompts
- [xformers](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers), major speed increase for select cards: (add --xformers to commandline args)
- via extension: [History tab](https://github.com/yfszzx/stable-diffusion-webui-images-browser): view, direct and delete images conveniently within the UI
- Generate forever option
@@ -84,26 +84,7 @@ Check the [custom scripts](https://github.com/AUTOMATIC1111/stable-diffusion-web
- API
- Support for dedicated [inpainting model](https://github.com/runwayml/stable-diffusion#inpainting-with-stable-diffusion) by RunwayML.
- via extension: [Aesthetic Gradients](https://github.com/AUTOMATIC1111/stable-diffusion-webui-aesthetic-gradients), a way to generate images with a specific aesthetic by using clip images embds (implementation of [https://github.com/vicgalle/stable-diffusion-aesthetic-gradients](https://github.com/vicgalle/stable-diffusion-aesthetic-gradients))
-
-## Where are Aesthetic Gradients?!?!
-Aesthetic Gradients are now an extension. You can install it using git:
-
-```commandline
-git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui-aesthetic-gradients extensions/aesthetic-gradients
-```
-
-After running this command, make sure that you have `aesthetic-gradients` dir in webui's `extensions` directory and restart
-the UI. The interface for Aesthetic Gradients should appear exactly the same as it was.
-
-## Where is History/Image browser?!?!
-Image browser is now an extension. You can install it using git:
-
-```commandline
-git clone https://github.com/yfszzx/stable-diffusion-webui-images-browser extensions/images-browser
-```
-
-After running this command, make sure that you have `images-browser` dir in webui's `extensions` directory and restart
-the UI. The interface for Image browser should appear exactly the same as it was.
+- [Stable Diffusion 2.0](https://github.com/Stability-AI/stablediffusion) support - see [wiki](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#stable-diffusion-20) for instructions
## Installation and Running
Make sure the required [dependencies](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Dependencies) are met and follow the instructions available for both [NVidia](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUs) (recommended) and [AMD](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs) GPUs.
diff --git a/launch.py b/launch.py
index 0f84b5d1..b1626cb5 100644
--- a/launch.py
+++ b/launch.py
@@ -134,19 +134,19 @@ def prepare_enviroment():
gfpgan_package = os.environ.get('GFPGAN_PACKAGE', "git+https://github.com/TencentARC/GFPGAN.git@8d2447a2d918f8eba5a4a01463fd48e45126a379")
clip_package = os.environ.get('CLIP_PACKAGE', "git+https://github.com/openai/CLIP.git@d50d76daa670286dd6cacf3bcd80b5e4823fc8e1")
- deepdanbooru_package = os.environ.get('DEEPDANBOORU_PACKAGE', "git+https://github.com/KichangKim/DeepDanbooru.git@d91a2963bf87c6a770d74894667e9ffa9f6de7ff")
+ openclip_package = os.environ.get('OPENCLIP_PACKAGE', "git+https://github.com/mlfoundations/open_clip.git@bb6e834e9c70d9c27d0dc3ecedeebeaeb1ffad6b")
xformers_windows_package = os.environ.get('XFORMERS_WINDOWS_PACKAGE', 'https://github.com/C43H66N12O12S2/stable-diffusion-webui/releases/download/f/xformers-0.0.14.dev0-cp310-cp310-win_amd64.whl')
- stable_diffusion_repo = os.environ.get('STABLE_DIFFUSION_REPO', "https://github.com/CompVis/stable-diffusion.git")
+ stable_diffusion_repo = os.environ.get('STABLE_DIFFUSION_REPO', "https://github.com/Stability-AI/stablediffusion.git")
taming_transformers_repo = os.environ.get('TAMING_TRANSFORMERS_REPO', "https://github.com/CompVis/taming-transformers.git")
k_diffusion_repo = os.environ.get('K_DIFFUSION_REPO', 'https://github.com/crowsonkb/k-diffusion.git')
codeformer_repo = os.environ.get('CODEFORMER_REPO', 'https://github.com/sczhou/CodeFormer.git')
blip_repo = os.environ.get('BLIP_REPO', 'https://github.com/salesforce/BLIP.git')
- stable_diffusion_commit_hash = os.environ.get('STABLE_DIFFUSION_COMMIT_HASH', "69ae4b35e0a0f6ee1af8bb9a5d0016ccb27e36dc")
+ stable_diffusion_commit_hash = os.environ.get('STABLE_DIFFUSION_COMMIT_HASH', "47b6b607fdd31875c9279cd2f4f16b92e4ea958e")
taming_transformers_commit_hash = os.environ.get('TAMING_TRANSFORMERS_COMMIT_HASH', "24268930bf1dce879235a7fddd0b2355b84d7ea6")
- k_diffusion_commit_hash = os.environ.get('K_DIFFUSION_COMMIT_HASH', "60e5042ca0da89c14d1dd59d73883280f8fce991")
+ k_diffusion_commit_hash = os.environ.get('K_DIFFUSION_COMMIT_HASH', "5b3af030dd83e0297272d861c19477735d0317ec")
codeformer_commit_hash = os.environ.get('CODEFORMER_COMMIT_HASH', "c5b4593074ba6214284d6acd5f1719b6c5d739af")
blip_commit_hash = os.environ.get('BLIP_COMMIT_HASH', "48211a1594f1321b00f14c9f7a5b4813144b2fb9")
@@ -158,7 +158,6 @@ def prepare_enviroment():
sys.argv, update_check = extract_arg(sys.argv, '--update-check')
sys.argv, run_tests = extract_arg(sys.argv, '--tests')
xformers = '--xformers' in sys.argv
- deepdanbooru = '--deepdanbooru' in sys.argv
ngrok = '--ngrok' in sys.argv
try:
@@ -181,6 +180,9 @@ def prepare_enviroment():
if not is_installed("clip"):
run_pip(f"install {clip_package}", "clip")
+ if not is_installed("open_clip"):
+ run_pip(f"install {openclip_package}", "open_clip")
+
if (not is_installed("xformers") or reinstall_xformers) and xformers:
if platform.system() == "Windows":
if platform.python_version().startswith("3.10"):
@@ -193,15 +195,12 @@ def prepare_enviroment():
elif platform.system() == "Linux":
run_pip("install xformers", "xformers")
- if not is_installed("deepdanbooru") and deepdanbooru:
- run_pip(f"install {deepdanbooru_package}#egg=deepdanbooru[tensorflow] tensorflow==2.10.0 tensorflow-io==0.27.0", "deepdanbooru")
-
if not is_installed("pyngrok") and ngrok:
run_pip("install pyngrok", "ngrok")
os.makedirs(dir_repos, exist_ok=True)
- git_clone(stable_diffusion_repo, repo_dir('stable-diffusion'), "Stable Diffusion", stable_diffusion_commit_hash)
+ git_clone(stable_diffusion_repo, repo_dir('stable-diffusion-stability-ai'), "Stable Diffusion", stable_diffusion_commit_hash)
git_clone(taming_transformers_repo, repo_dir('taming-transformers'), "Taming Transformers", taming_transformers_commit_hash)
git_clone(k_diffusion_repo, repo_dir('k-diffusion'), "K-diffusion", k_diffusion_commit_hash)
git_clone(codeformer_repo, repo_dir('CodeFormer'), "CodeFormer", codeformer_commit_hash)
diff --git a/modules/api/api.py b/modules/api/api.py
index 79b2c818..7a567be3 100644
--- a/modules/api/api.py
+++ b/modules/api/api.py
@@ -9,7 +9,7 @@ from fastapi.security import HTTPBasic, HTTPBasicCredentials
from secrets import compare_digest
import modules.shared as shared
-from modules import sd_samplers
+from modules import sd_samplers, deepbooru
from modules.api.models import *
from modules.processing import StableDiffusionProcessingTxt2Img, StableDiffusionProcessingImg2Img, process_images
from modules.extras import run_extras, run_pnginfo
@@ -18,9 +18,6 @@ from modules.sd_models import checkpoints_list
from modules.realesrgan_model import get_realesrgan_models
from typing import List
-if shared.cmd_opts.deepdanbooru:
- from modules.deepbooru import get_deepbooru_tags
-
def upscaler_to_index(name: str):
try:
return [x.name.lower() for x in shared.sd_upscalers].index(name.lower())
@@ -245,10 +242,7 @@ class Api:
if interrogatereq.model == "clip":
processed = shared.interrogator.interrogate(img)
elif interrogatereq.model == "deepdanbooru":
- if shared.cmd_opts.deepdanbooru:
- processed = get_deepbooru_tags(img)
- else:
- raise HTTPException(status_code=404, detail="Model not found. Add --deepdanbooru when launching for using the model.")
+ processed = deepbooru.model.tag(img)
else:
raise HTTPException(status_code=404, detail="Model not found")
diff --git a/modules/deepbooru.py b/modules/deepbooru.py
index 8bbc90a4..b9066d81 100644
--- a/modules/deepbooru.py
+++ b/modules/deepbooru.py
@@ -1,173 +1,97 @@
-import os.path
-from concurrent.futures import ProcessPoolExecutor
-import multiprocessing
-import time
+import os
import re
+import torch
+from PIL import Image
+import numpy as np
+
+from modules import modelloader, paths, deepbooru_model, devices, images, shared
+
re_special = re.compile(r'([\\()])')
-def get_deepbooru_tags(pil_image):
- """
- This method is for running only one image at a time for simple use. Used to the img2img interrogate.
- """
- from modules import shared # prevents circular reference
-
- try:
- create_deepbooru_process(shared.opts.interrogate_deepbooru_score_threshold, create_deepbooru_opts())
- return get_tags_from_process(pil_image)
- finally:
- release_process()
-
-
-OPT_INCLUDE_RANKS = "include_ranks"
-def create_deepbooru_opts():
- from modules import shared
-
- return {
- "use_spaces": shared.opts.deepbooru_use_spaces,
- "use_escape": shared.opts.deepbooru_escape,
- "alpha_sort": shared.opts.deepbooru_sort_alpha,
- OPT_INCLUDE_RANKS: shared.opts.interrogate_return_ranks,
- }
-
-
-def deepbooru_process(queue, deepbooru_process_return, threshold, deepbooru_opts):
- model, tags = get_deepbooru_tags_model()
- while True: # while process is running, keep monitoring queue for new image
- pil_image = queue.get()
- if pil_image == "QUIT":
- break
- else:
- deepbooru_process_return["value"] = get_deepbooru_tags_from_model(model, tags, pil_image, threshold, deepbooru_opts)
-
-
-def create_deepbooru_process(threshold, deepbooru_opts):
- """
- Creates deepbooru process. A queue is created to send images into the process. This enables multiple images
- to be processed in a row without reloading the model or creating a new process. To return the data, a shared
- dictionary is created to hold the tags created. To wait for tags to be returned, a value of -1 is assigned
- to the dictionary and the method adding the image to the queue should wait for this value to be updated with
- the tags.
- """
- from modules import shared # prevents circular reference
- context = multiprocessing.get_context("spawn")
- shared.deepbooru_process_manager = context.Manager()
- shared.deepbooru_process_queue = shared.deepbooru_process_manager.Queue()
- shared.deepbooru_process_return = shared.deepbooru_process_manager.dict()
- shared.deepbooru_process_return["value"] = -1
- shared.deepbooru_process = context.Process(target=deepbooru_process, args=(shared.deepbooru_process_queue, shared.deepbooru_process_return, threshold, deepbooru_opts))
- shared.deepbooru_process.start()
-
-
-def get_tags_from_process(image):
- from modules import shared
-
- shared.deepbooru_process_return["value"] = -1
- shared.deepbooru_process_queue.put(image)
- while shared.deepbooru_process_return["value"] == -1:
- time.sleep(0.2)
- caption = shared.deepbooru_process_return["value"]
- shared.deepbooru_process_return["value"] = -1
-
- return caption
-
-
-def release_process():
- """
- Stops the deepbooru process to return used memory
- """
- from modules import shared # prevents circular reference
- shared.deepbooru_process_queue.put("QUIT")
- shared.deepbooru_process.join()
- shared.deepbooru_process_queue = None
- shared.deepbooru_process = None
- shared.deepbooru_process_return = None
- shared.deepbooru_process_manager = None
-
-def get_deepbooru_tags_model():
- import deepdanbooru as dd
- import tensorflow as tf
- import numpy as np
- this_folder = os.path.dirname(__file__)
- model_path = os.path.abspath(os.path.join(this_folder, '..', 'models', 'deepbooru'))
- if not os.path.exists(os.path.join(model_path, 'project.json')):
- # there is no point importing these every time
- import zipfile
- from basicsr.utils.download_util import load_file_from_url
- load_file_from_url(
- r"https://github.com/KichangKim/DeepDanbooru/releases/download/v3-20211112-sgd-e28/deepdanbooru-v3-20211112-sgd-e28.zip",
- model_path)
- with zipfile.ZipFile(os.path.join(model_path, "deepdanbooru-v3-20211112-sgd-e28.zip"), "r") as zip_ref:
- zip_ref.extractall(model_path)
- os.remove(os.path.join(model_path, "deepdanbooru-v3-20211112-sgd-e28.zip"))
-
- tags = dd.project.load_tags_from_project(model_path)
- model = dd.project.load_model_from_project(
- model_path, compile_model=False
- )
- return model, tags
-
-
-def get_deepbooru_tags_from_model(model, tags, pil_image, threshold, deepbooru_opts):
- import deepdanbooru as dd
- import tensorflow as tf
- import numpy as np
-
- alpha_sort = deepbooru_opts['alpha_sort']
- use_spaces = deepbooru_opts['use_spaces']
- use_escape = deepbooru_opts['use_escape']
- include_ranks = deepbooru_opts['include_ranks']
-
- width = model.input_shape[2]
- height = model.input_shape[1]
- image = np.array(pil_image)
- image = tf.image.resize(
- image,
- size=(height, width),
- method=tf.image.ResizeMethod.AREA,
- preserve_aspect_ratio=True,
- )
- image = image.numpy() # EagerTensor to np.array
- image = dd.image.transform_and_pad_image(image, width, height)
- image = image / 255.0
- image_shape = image.shape
- image = image.reshape((1, image_shape[0], image_shape[1], image_shape[2]))
-
- y = model.predict(image)[0]
-
- result_dict = {}
-
- for i, tag in enumerate(tags):
- result_dict[tag] = y[i]
-
- unsorted_tags_in_theshold = []
- result_tags_print = []
- for tag in tags:
- if result_dict[tag] >= threshold:
+
+class DeepDanbooru:
+ def __init__(self):
+ self.model = None
+
+ def load(self):
+ if self.model is not None:
+ return
+
+ files = modelloader.load_models(
+ model_path=os.path.join(paths.models_path, "torch_deepdanbooru"),
+ model_url='https://github.com/AUTOMATIC1111/TorchDeepDanbooru/releases/download/v1/model-resnet_custom_v3.pt',
+ ext_filter=".pt",
+ download_name='model-resnet_custom_v3.pt',
+ )
+
+ self.model = deepbooru_model.DeepDanbooruModel()
+ self.model.load_state_dict(torch.load(files[0], map_location="cpu"))
+
+ self.model.eval()
+ self.model.to(devices.cpu, devices.dtype)
+
+ def start(self):
+ self.load()
+ self.model.to(devices.device)
+
+ def stop(self):
+ if not shared.opts.interrogate_keep_models_in_memory:
+ self.model.to(devices.cpu)
+ devices.torch_gc()
+
+ def tag(self, pil_image):
+ self.start()
+ res = self.tag_multi(pil_image)
+ self.stop()
+
+ return res
+
+ def tag_multi(self, pil_image, force_disable_ranks=False):
+ threshold = shared.opts.interrogate_deepbooru_score_threshold
+ use_spaces = shared.opts.deepbooru_use_spaces
+ use_escape = shared.opts.deepbooru_escape
+ alpha_sort = shared.opts.deepbooru_sort_alpha
+ include_ranks = shared.opts.interrogate_return_ranks and not force_disable_ranks
+
+ pic = images.resize_image(2, pil_image.convert("RGB"), 512, 512)
+ a = np.expand_dims(np.array(pic, dtype=np.float32), 0) / 255
+
+ with torch.no_grad(), devices.autocast():
+ x = torch.from_numpy(a).cuda()
+ y = self.model(x)[0].detach().cpu().numpy()
+
+ probability_dict = {}
+
+ for tag, probability in zip(self.model.tags, y):
+ if probability < threshold:
+ continue
+
if tag.startswith("rating:"):
continue
- unsorted_tags_in_theshold.append((result_dict[tag], tag))
- result_tags_print.append(f'{result_dict[tag]} {tag}')
-
- # sort tags
- result_tags_out = []
- sort_ndx = 0
- if alpha_sort:
- sort_ndx = 1
-
- # sort by reverse by likelihood and normal for alpha, and format tag text as requested
- unsorted_tags_in_theshold.sort(key=lambda y: y[sort_ndx], reverse=(not alpha_sort))
- for weight, tag in unsorted_tags_in_theshold:
- tag_outformat = tag
- if use_spaces:
- tag_outformat = tag_outformat.replace('_', ' ')
- if use_escape:
- tag_outformat = re.sub(re_special, r'\\\1', tag_outformat)
- if include_ranks:
- tag_outformat = f"({tag_outformat}:{weight:.3f})"
-
- result_tags_out.append(tag_outformat)
-
- print('\n'.join(sorted(result_tags_print, reverse=True)))
-
- return ', '.join(result_tags_out)
+
+ probability_dict[tag] = probability
+
+ if alpha_sort:
+ tags = sorted(probability_dict)
+ else:
+ tags = [tag for tag, _ in sorted(probability_dict.items(), key=lambda x: -x[1])]
+
+ res = []
+
+ for tag in tags:
+ probability = probability_dict[tag]
+ tag_outformat = tag
+ if use_spaces:
+ tag_outformat = tag_outformat.replace('_', ' ')
+ if use_escape:
+ tag_outformat = re.sub(re_special, r'\\\1', tag_outformat)
+ if include_ranks:
+ tag_outformat = f"({tag_outformat}:{probability:.3f})"
+
+ res.append(tag_outformat)
+
+ return ", ".join(res)
+
+
+model = DeepDanbooru()
diff --git a/modules/deepbooru_model.py b/modules/deepbooru_model.py
new file mode 100644
index 00000000..edd40c81
--- /dev/null
+++ b/modules/deepbooru_model.py
@@ -0,0 +1,676 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+# see https://github.com/AUTOMATIC1111/TorchDeepDanbooru for more
+
+
+class DeepDanbooruModel(nn.Module):
+ def __init__(self):
+ super(DeepDanbooruModel, self).__init__()
+
+ self.tags = []
+
+ self.n_Conv_0 = nn.Conv2d(kernel_size=(7, 7), in_channels=3, out_channels=64, stride=(2, 2))
+ self.n_MaxPool_0 = nn.MaxPool2d(kernel_size=(3, 3), stride=(2, 2))
+ self.n_Conv_1 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256)
+ self.n_Conv_2 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=64)
+ self.n_Conv_3 = nn.Conv2d(kernel_size=(3, 3), in_channels=64, out_channels=64)
+ self.n_Conv_4 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256)
+ self.n_Conv_5 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=64)
+ self.n_Conv_6 = nn.Conv2d(kernel_size=(3, 3), in_channels=64, out_channels=64)
+ self.n_Conv_7 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256)
+ self.n_Conv_8 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=64)
+ self.n_Conv_9 = nn.Conv2d(kernel_size=(3, 3), in_channels=64, out_channels=64)
+ self.n_Conv_10 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256)
+ self.n_Conv_11 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=512, stride=(2, 2))
+ self.n_Conv_12 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=128)
+ self.n_Conv_13 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128, stride=(2, 2))
+ self.n_Conv_14 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_15 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_16 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_17 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_18 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_19 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_20 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_21 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_22 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_23 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_24 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_25 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_26 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_27 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_28 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_29 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_30 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_31 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_32 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_33 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128)
+ self.n_Conv_34 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128)
+ self.n_Conv_35 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512)
+ self.n_Conv_36 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=1024, stride=(2, 2))
+ self.n_Conv_37 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=256)
+ self.n_Conv_38 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256, stride=(2, 2))
+ self.n_Conv_39 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_40 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_41 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_42 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_43 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_44 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_45 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_46 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_47 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_48 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_49 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_50 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_51 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_52 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_53 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_54 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_55 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_56 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_57 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_58 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_59 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_60 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_61 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_62 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_63 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_64 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_65 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_66 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_67 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_68 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_69 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_70 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_71 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_72 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_73 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_74 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_75 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_76 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_77 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_78 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_79 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_80 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_81 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_82 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_83 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_84 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_85 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_86 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_87 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_88 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_89 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_90 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_91 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_92 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_93 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_94 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_95 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_96 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_97 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_98 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256, stride=(2, 2))
+ self.n_Conv_99 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_100 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=1024, stride=(2, 2))
+ self.n_Conv_101 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_102 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_103 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_104 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_105 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_106 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_107 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_108 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_109 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_110 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_111 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_112 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_113 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_114 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_115 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_116 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_117 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_118 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_119 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_120 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_121 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_122 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_123 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_124 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_125 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_126 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_127 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_128 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_129 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_130 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_131 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_132 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_133 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_134 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_135 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_136 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_137 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_138 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_139 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_140 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_141 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_142 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_143 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_144 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_145 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_146 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_147 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_148 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_149 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_150 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_151 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_152 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_153 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_154 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_155 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256)
+ self.n_Conv_156 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256)
+ self.n_Conv_157 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024)
+ self.n_Conv_158 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=2048, stride=(2, 2))
+ self.n_Conv_159 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=512)
+ self.n_Conv_160 = nn.Conv2d(kernel_size=(3, 3), in_channels=512, out_channels=512, stride=(2, 2))
+ self.n_Conv_161 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=2048)
+ self.n_Conv_162 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=512)
+ self.n_Conv_163 = nn.Conv2d(kernel_size=(3, 3), in_channels=512, out_channels=512)
+ self.n_Conv_164 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=2048)
+ self.n_Conv_165 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=512)
+ self.n_Conv_166 = nn.Conv2d(kernel_size=(3, 3), in_channels=512, out_channels=512)
+ self.n_Conv_167 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=2048)
+ self.n_Conv_168 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=4096, stride=(2, 2))
+ self.n_Conv_169 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=1024)
+ self.n_Conv_170 = nn.Conv2d(kernel_size=(3, 3), in_channels=1024, out_channels=1024, stride=(2, 2))
+ self.n_Conv_171 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=4096)
+ self.n_Conv_172 = nn.Conv2d(kernel_size=(1, 1), in_channels=4096, out_channels=1024)
+ self.n_Conv_173 = nn.Conv2d(kernel_size=(3, 3), in_channels=1024, out_channels=1024)
+ self.n_Conv_174 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=4096)
+ self.n_Conv_175 = nn.Conv2d(kernel_size=(1, 1), in_channels=4096, out_channels=1024)
+ self.n_Conv_176 = nn.Conv2d(kernel_size=(3, 3), in_channels=1024, out_channels=1024)
+ self.n_Conv_177 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=4096)
+ self.n_Conv_178 = nn.Conv2d(kernel_size=(1, 1), in_channels=4096, out_channels=9176, bias=False)
+
+ def forward(self, *inputs):
+ t_358, = inputs
+ t_359 = t_358.permute(*[0, 3, 1, 2])
+ t_359_padded = F.pad(t_359, [2, 3, 2, 3], value=0)
+ t_360 = self.n_Conv_0(t_359_padded)
+ t_361 = F.relu(t_360)
+ t_361 = F.pad(t_361, [0, 1, 0, 1], value=float('-inf'))
+ t_362 = self.n_MaxPool_0(t_361)
+ t_363 = self.n_Conv_1(t_362)
+ t_364 = self.n_Conv_2(t_362)
+ t_365 = F.relu(t_364)
+ t_365_padded = F.pad(t_365, [1, 1, 1, 1], value=0)
+ t_366 = self.n_Conv_3(t_365_padded)
+ t_367 = F.relu(t_366)
+ t_368 = self.n_Conv_4(t_367)
+ t_369 = torch.add(t_368, t_363)
+ t_370 = F.relu(t_369)
+ t_371 = self.n_Conv_5(t_370)
+ t_372 = F.relu(t_371)
+ t_372_padded = F.pad(t_372, [1, 1, 1, 1], value=0)
+ t_373 = self.n_Conv_6(t_372_padded)
+ t_374 = F.relu(t_373)
+ t_375 = self.n_Conv_7(t_374)
+ t_376 = torch.add(t_375, t_370)
+ t_377 = F.relu(t_376)
+ t_378 = self.n_Conv_8(t_377)
+ t_379 = F.relu(t_378)
+ t_379_padded = F.pad(t_379, [1, 1, 1, 1], value=0)
+ t_380 = self.n_Conv_9(t_379_padded)
+ t_381 = F.relu(t_380)
+ t_382 = self.n_Conv_10(t_381)
+ t_383 = torch.add(t_382, t_377)
+ t_384 = F.relu(t_383)
+ t_385 = self.n_Conv_11(t_384)
+ t_386 = self.n_Conv_12(t_384)
+ t_387 = F.relu(t_386)
+ t_387_padded = F.pad(t_387, [0, 1, 0, 1], value=0)
+ t_388 = self.n_Conv_13(t_387_padded)
+ t_389 = F.relu(t_388)
+ t_390 = self.n_Conv_14(t_389)
+ t_391 = torch.add(t_390, t_385)
+ t_392 = F.relu(t_391)
+ t_393 = self.n_Conv_15(t_392)
+ t_394 = F.relu(t_393)
+ t_394_padded = F.pad(t_394, [1, 1, 1, 1], value=0)
+ t_395 = self.n_Conv_16(t_394_padded)
+ t_396 = F.relu(t_395)
+ t_397 = self.n_Conv_17(t_396)
+ t_398 = torch.add(t_397, t_392)
+ t_399 = F.relu(t_398)
+ t_400 = self.n_Conv_18(t_399)
+ t_401 = F.relu(t_400)
+ t_401_padded = F.pad(t_401, [1, 1, 1, 1], value=0)
+ t_402 = self.n_Conv_19(t_401_padded)
+ t_403 = F.relu(t_402)
+ t_404 = self.n_Conv_20(t_403)
+ t_405 = torch.add(t_404, t_399)
+ t_406 = F.relu(t_405)
+ t_407 = self.n_Conv_21(t_406)
+ t_408 = F.relu(t_407)
+ t_408_padded = F.pad(t_408, [1, 1, 1, 1], value=0)
+ t_409 = self.n_Conv_22(t_408_padded)
+ t_410 = F.relu(t_409)
+ t_411 = self.n_Conv_23(t_410)
+ t_412 = torch.add(t_411, t_406)
+ t_413 = F.relu(t_412)
+ t_414 = self.n_Conv_24(t_413)
+ t_415 = F.relu(t_414)
+ t_415_padded = F.pad(t_415, [1, 1, 1, 1], value=0)
+ t_416 = self.n_Conv_25(t_415_padded)
+ t_417 = F.relu(t_416)
+ t_418 = self.n_Conv_26(t_417)
+ t_419 = torch.add(t_418, t_413)
+ t_420 = F.relu(t_419)
+ t_421 = self.n_Conv_27(t_420)
+ t_422 = F.relu(t_421)
+ t_422_padded = F.pad(t_422, [1, 1, 1, 1], value=0)
+ t_423 = self.n_Conv_28(t_422_padded)
+ t_424 = F.relu(t_423)
+ t_425 = self.n_Conv_29(t_424)
+ t_426 = torch.add(t_425, t_420)
+ t_427 = F.relu(t_426)
+ t_428 = self.n_Conv_30(t_427)
+ t_429 = F.relu(t_428)
+ t_429_padded = F.pad(t_429, [1, 1, 1, 1], value=0)
+ t_430 = self.n_Conv_31(t_429_padded)
+ t_431 = F.relu(t_430)
+ t_432 = self.n_Conv_32(t_431)
+ t_433 = torch.add(t_432, t_427)
+ t_434 = F.relu(t_433)
+ t_435 = self.n_Conv_33(t_434)
+ t_436 = F.relu(t_435)
+ t_436_padded = F.pad(t_436, [1, 1, 1, 1], value=0)
+ t_437 = self.n_Conv_34(t_436_padded)
+ t_438 = F.relu(t_437)
+ t_439 = self.n_Conv_35(t_438)
+ t_440 = torch.add(t_439, t_434)
+ t_441 = F.relu(t_440)
+ t_442 = self.n_Conv_36(t_441)
+ t_443 = self.n_Conv_37(t_441)
+ t_444 = F.relu(t_443)
+ t_444_padded = F.pad(t_444, [0, 1, 0, 1], value=0)
+ t_445 = self.n_Conv_38(t_444_padded)
+ t_446 = F.relu(t_445)
+ t_447 = self.n_Conv_39(t_446)
+ t_448 = torch.add(t_447, t_442)
+ t_449 = F.relu(t_448)
+ t_450 = self.n_Conv_40(t_449)
+ t_451 = F.relu(t_450)
+ t_451_padded = F.pad(t_451, [1, 1, 1, 1], value=0)
+ t_452 = self.n_Conv_41(t_451_padded)
+ t_453 = F.relu(t_452)
+ t_454 = self.n_Conv_42(t_453)
+ t_455 = torch.add(t_454, t_449)
+ t_456 = F.relu(t_455)
+ t_457 = self.n_Conv_43(t_456)
+ t_458 = F.relu(t_457)
+ t_458_padded = F.pad(t_458, [1, 1, 1, 1], value=0)
+ t_459 = self.n_Conv_44(t_458_padded)
+ t_460 = F.relu(t_459)
+ t_461 = self.n_Conv_45(t_460)
+ t_462 = torch.add(t_461, t_456)
+ t_463 = F.relu(t_462)
+ t_464 = self.n_Conv_46(t_463)
+ t_465 = F.relu(t_464)
+ t_465_padded = F.pad(t_465, [1, 1, 1, 1], value=0)
+ t_466 = self.n_Conv_47(t_465_padded)
+ t_467 = F.relu(t_466)
+ t_468 = self.n_Conv_48(t_467)
+ t_469 = torch.add(t_468, t_463)
+ t_470 = F.relu(t_469)
+ t_471 = self.n_Conv_49(t_470)
+ t_472 = F.relu(t_471)
+ t_472_padded = F.pad(t_472, [1, 1, 1, 1], value=0)
+ t_473 = self.n_Conv_50(t_472_padded)
+ t_474 = F.relu(t_473)
+ t_475 = self.n_Conv_51(t_474)
+ t_476 = torch.add(t_475, t_470)
+ t_477 = F.relu(t_476)
+ t_478 = self.n_Conv_52(t_477)
+ t_479 = F.relu(t_478)
+ t_479_padded = F.pad(t_479, [1, 1, 1, 1], value=0)
+ t_480 = self.n_Conv_53(t_479_padded)
+ t_481 = F.relu(t_480)
+ t_482 = self.n_Conv_54(t_481)
+ t_483 = torch.add(t_482, t_477)
+ t_484 = F.relu(t_483)
+ t_485 = self.n_Conv_55(t_484)
+ t_486 = F.relu(t_485)
+ t_486_padded = F.pad(t_486, [1, 1, 1, 1], value=0)
+ t_487 = self.n_Conv_56(t_486_padded)
+ t_488 = F.relu(t_487)
+ t_489 = self.n_Conv_57(t_488)
+ t_490 = torch.add(t_489, t_484)
+ t_491 = F.relu(t_490)
+ t_492 = self.n_Conv_58(t_491)
+ t_493 = F.relu(t_492)
+ t_493_padded = F.pad(t_493, [1, 1, 1, 1], value=0)
+ t_494 = self.n_Conv_59(t_493_padded)
+ t_495 = F.relu(t_494)
+ t_496 = self.n_Conv_60(t_495)
+ t_497 = torch.add(t_496, t_491)
+ t_498 = F.relu(t_497)
+ t_499 = self.n_Conv_61(t_498)
+ t_500 = F.relu(t_499)
+ t_500_padded = F.pad(t_500, [1, 1, 1, 1], value=0)
+ t_501 = self.n_Conv_62(t_500_padded)
+ t_502 = F.relu(t_501)
+ t_503 = self.n_Conv_63(t_502)
+ t_504 = torch.add(t_503, t_498)
+ t_505 = F.relu(t_504)
+ t_506 = self.n_Conv_64(t_505)
+ t_507 = F.relu(t_506)
+ t_507_padded = F.pad(t_507, [1, 1, 1, 1], value=0)
+ t_508 = self.n_Conv_65(t_507_padded)
+ t_509 = F.relu(t_508)
+ t_510 = self.n_Conv_66(t_509)
+ t_511 = torch.add(t_510, t_505)
+ t_512 = F.relu(t_511)
+ t_513 = self.n_Conv_67(t_512)
+ t_514 = F.relu(t_513)
+ t_514_padded = F.pad(t_514, [1, 1, 1, 1], value=0)
+ t_515 = self.n_Conv_68(t_514_padded)
+ t_516 = F.relu(t_515)
+ t_517 = self.n_Conv_69(t_516)
+ t_518 = torch.add(t_517, t_512)
+ t_519 = F.relu(t_518)
+ t_520 = self.n_Conv_70(t_519)
+ t_521 = F.relu(t_520)
+ t_521_padded = F.pad(t_521, [1, 1, 1, 1], value=0)
+ t_522 = self.n_Conv_71(t_521_padded)
+ t_523 = F.relu(t_522)
+ t_524 = self.n_Conv_72(t_523)
+ t_525 = torch.add(t_524, t_519)
+ t_526 = F.relu(t_525)
+ t_527 = self.n_Conv_73(t_526)
+ t_528 = F.relu(t_527)
+ t_528_padded = F.pad(t_528, [1, 1, 1, 1], value=0)
+ t_529 = self.n_Conv_74(t_528_padded)
+ t_530 = F.relu(t_529)
+ t_531 = self.n_Conv_75(t_530)
+ t_532 = torch.add(t_531, t_526)
+ t_533 = F.relu(t_532)
+ t_534 = self.n_Conv_76(t_533)
+ t_535 = F.relu(t_534)
+ t_535_padded = F.pad(t_535, [1, 1, 1, 1], value=0)
+ t_536 = self.n_Conv_77(t_535_padded)
+ t_537 = F.relu(t_536)
+ t_538 = self.n_Conv_78(t_537)
+ t_539 = torch.add(t_538, t_533)
+ t_540 = F.relu(t_539)
+ t_541 = self.n_Conv_79(t_540)
+ t_542 = F.relu(t_541)
+ t_542_padded = F.pad(t_542, [1, 1, 1, 1], value=0)
+ t_543 = self.n_Conv_80(t_542_padded)
+ t_544 = F.relu(t_543)
+ t_545 = self.n_Conv_81(t_544)
+ t_546 = torch.add(t_545, t_540)
+ t_547 = F.relu(t_546)
+ t_548 = self.n_Conv_82(t_547)
+ t_549 = F.relu(t_548)
+ t_549_padded = F.pad(t_549, [1, 1, 1, 1], value=0)
+ t_550 = self.n_Conv_83(t_549_padded)
+ t_551 = F.relu(t_550)
+ t_552 = self.n_Conv_84(t_551)
+ t_553 = torch.add(t_552, t_547)
+ t_554 = F.relu(t_553)
+ t_555 = self.n_Conv_85(t_554)
+ t_556 = F.relu(t_555)
+ t_556_padded = F.pad(t_556, [1, 1, 1, 1], value=0)
+ t_557 = self.n_Conv_86(t_556_padded)
+ t_558 = F.relu(t_557)
+ t_559 = self.n_Conv_87(t_558)
+ t_560 = torch.add(t_559, t_554)
+ t_561 = F.relu(t_560)
+ t_562 = self.n_Conv_88(t_561)
+ t_563 = F.relu(t_562)
+ t_563_padded = F.pad(t_563, [1, 1, 1, 1], value=0)
+ t_564 = self.n_Conv_89(t_563_padded)
+ t_565 = F.relu(t_564)
+ t_566 = self.n_Conv_90(t_565)
+ t_567 = torch.add(t_566, t_561)
+ t_568 = F.relu(t_567)
+ t_569 = self.n_Conv_91(t_568)
+ t_570 = F.relu(t_569)
+ t_570_padded = F.pad(t_570, [1, 1, 1, 1], value=0)
+ t_571 = self.n_Conv_92(t_570_padded)
+ t_572 = F.relu(t_571)
+ t_573 = self.n_Conv_93(t_572)
+ t_574 = torch.add(t_573, t_568)
+ t_575 = F.relu(t_574)
+ t_576 = self.n_Conv_94(t_575)
+ t_577 = F.relu(t_576)
+ t_577_padded = F.pad(t_577, [1, 1, 1, 1], value=0)
+ t_578 = self.n_Conv_95(t_577_padded)
+ t_579 = F.relu(t_578)
+ t_580 = self.n_Conv_96(t_579)
+ t_581 = torch.add(t_580, t_575)
+ t_582 = F.relu(t_581)
+ t_583 = self.n_Conv_97(t_582)
+ t_584 = F.relu(t_583)
+ t_584_padded = F.pad(t_584, [0, 1, 0, 1], value=0)
+ t_585 = self.n_Conv_98(t_584_padded)
+ t_586 = F.relu(t_585)
+ t_587 = self.n_Conv_99(t_586)
+ t_588 = self.n_Conv_100(t_582)
+ t_589 = torch.add(t_587, t_588)
+ t_590 = F.relu(t_589)
+ t_591 = self.n_Conv_101(t_590)
+ t_592 = F.relu(t_591)
+ t_592_padded = F.pad(t_592, [1, 1, 1, 1], value=0)
+ t_593 = self.n_Conv_102(t_592_padded)
+ t_594 = F.relu(t_593)
+ t_595 = self.n_Conv_103(t_594)
+ t_596 = torch.add(t_595, t_590)
+ t_597 = F.relu(t_596)
+ t_598 = self.n_Conv_104(t_597)
+ t_599 = F.relu(t_598)
+ t_599_padded = F.pad(t_599, [1, 1, 1, 1], value=0)
+ t_600 = self.n_Conv_105(t_599_padded)
+ t_601 = F.relu(t_600)
+ t_602 = self.n_Conv_106(t_601)
+ t_603 = torch.add(t_602, t_597)
+ t_604 = F.relu(t_603)
+ t_605 = self.n_Conv_107(t_604)
+ t_606 = F.relu(t_605)
+ t_606_padded = F.pad(t_606, [1, 1, 1, 1], value=0)
+ t_607 = self.n_Conv_108(t_606_padded)
+ t_608 = F.relu(t_607)
+ t_609 = self.n_Conv_109(t_608)
+ t_610 = torch.add(t_609, t_604)
+ t_611 = F.relu(t_610)
+ t_612 = self.n_Conv_110(t_611)
+ t_613 = F.relu(t_612)
+ t_613_padded = F.pad(t_613, [1, 1, 1, 1], value=0)
+ t_614 = self.n_Conv_111(t_613_padded)
+ t_615 = F.relu(t_614)
+ t_616 = self.n_Conv_112(t_615)
+ t_617 = torch.add(t_616, t_611)
+ t_618 = F.relu(t_617)
+ t_619 = self.n_Conv_113(t_618)
+ t_620 = F.relu(t_619)
+ t_620_padded = F.pad(t_620, [1, 1, 1, 1], value=0)
+ t_621 = self.n_Conv_114(t_620_padded)
+ t_622 = F.relu(t_621)
+ t_623 = self.n_Conv_115(t_622)
+ t_624 = torch.add(t_623, t_618)
+ t_625 = F.relu(t_624)
+ t_626 = self.n_Conv_116(t_625)
+ t_627 = F.relu(t_626)
+ t_627_padded = F.pad(t_627, [1, 1, 1, 1], value=0)
+ t_628 = self.n_Conv_117(t_627_padded)
+ t_629 = F.relu(t_628)
+ t_630 = self.n_Conv_118(t_629)
+ t_631 = torch.add(t_630, t_625)
+ t_632 = F.relu(t_631)
+ t_633 = self.n_Conv_119(t_632)
+ t_634 = F.relu(t_633)
+ t_634_padded = F.pad(t_634, [1, 1, 1, 1], value=0)
+ t_635 = self.n_Conv_120(t_634_padded)
+ t_636 = F.relu(t_635)
+ t_637 = self.n_Conv_121(t_636)
+ t_638 = torch.add(t_637, t_632)
+ t_639 = F.relu(t_638)
+ t_640 = self.n_Conv_122(t_639)
+ t_641 = F.relu(t_640)
+ t_641_padded = F.pad(t_641, [1, 1, 1, 1], value=0)
+ t_642 = self.n_Conv_123(t_641_padded)
+ t_643 = F.relu(t_642)
+ t_644 = self.n_Conv_124(t_643)
+ t_645 = torch.add(t_644, t_639)
+ t_646 = F.relu(t_645)
+ t_647 = self.n_Conv_125(t_646)
+ t_648 = F.relu(t_647)
+ t_648_padded = F.pad(t_648, [1, 1, 1, 1], value=0)
+ t_649 = self.n_Conv_126(t_648_padded)
+ t_650 = F.relu(t_649)
+ t_651 = self.n_Conv_127(t_650)
+ t_652 = torch.add(t_651, t_646)
+ t_653 = F.relu(t_652)
+ t_654 = self.n_Conv_128(t_653)
+ t_655 = F.relu(t_654)
+ t_655_padded = F.pad(t_655, [1, 1, 1, 1], value=0)
+ t_656 = self.n_Conv_129(t_655_padded)
+ t_657 = F.relu(t_656)
+ t_658 = self.n_Conv_130(t_657)
+ t_659 = torch.add(t_658, t_653)
+ t_660 = F.relu(t_659)
+ t_661 = self.n_Conv_131(t_660)
+ t_662 = F.relu(t_661)
+ t_662_padded = F.pad(t_662, [1, 1, 1, 1], value=0)
+ t_663 = self.n_Conv_132(t_662_padded)
+ t_664 = F.relu(t_663)
+ t_665 = self.n_Conv_133(t_664)
+ t_666 = torch.add(t_665, t_660)
+ t_667 = F.relu(t_666)
+ t_668 = self.n_Conv_134(t_667)
+ t_669 = F.relu(t_668)
+ t_669_padded = F.pad(t_669, [1, 1, 1, 1], value=0)
+ t_670 = self.n_Conv_135(t_669_padded)
+ t_671 = F.relu(t_670)
+ t_672 = self.n_Conv_136(t_671)
+ t_673 = torch.add(t_672, t_667)
+ t_674 = F.relu(t_673)
+ t_675 = self.n_Conv_137(t_674)
+ t_676 = F.relu(t_675)
+ t_676_padded = F.pad(t_676, [1, 1, 1, 1], value=0)
+ t_677 = self.n_Conv_138(t_676_padded)
+ t_678 = F.relu(t_677)
+ t_679 = self.n_Conv_139(t_678)
+ t_680 = torch.add(t_679, t_674)
+ t_681 = F.relu(t_680)
+ t_682 = self.n_Conv_140(t_681)
+ t_683 = F.relu(t_682)
+ t_683_padded = F.pad(t_683, [1, 1, 1, 1], value=0)
+ t_684 = self.n_Conv_141(t_683_padded)
+ t_685 = F.relu(t_684)
+ t_686 = self.n_Conv_142(t_685)
+ t_687 = torch.add(t_686, t_681)
+ t_688 = F.relu(t_687)
+ t_689 = self.n_Conv_143(t_688)
+ t_690 = F.relu(t_689)
+ t_690_padded = F.pad(t_690, [1, 1, 1, 1], value=0)
+ t_691 = self.n_Conv_144(t_690_padded)
+ t_692 = F.relu(t_691)
+ t_693 = self.n_Conv_145(t_692)
+ t_694 = torch.add(t_693, t_688)
+ t_695 = F.relu(t_694)
+ t_696 = self.n_Conv_146(t_695)
+ t_697 = F.relu(t_696)
+ t_697_padded = F.pad(t_697, [1, 1, 1, 1], value=0)
+ t_698 = self.n_Conv_147(t_697_padded)
+ t_699 = F.relu(t_698)
+ t_700 = self.n_Conv_148(t_699)
+ t_701 = torch.add(t_700, t_695)
+ t_702 = F.relu(t_701)
+ t_703 = self.n_Conv_149(t_702)
+ t_704 = F.relu(t_703)
+ t_704_padded = F.pad(t_704, [1, 1, 1, 1], value=0)
+ t_705 = self.n_Conv_150(t_704_padded)
+ t_706 = F.relu(t_705)
+ t_707 = self.n_Conv_151(t_706)
+ t_708 = torch.add(t_707, t_702)
+ t_709 = F.relu(t_708)
+ t_710 = self.n_Conv_152(t_709)
+ t_711 = F.relu(t_710)
+ t_711_padded = F.pad(t_711, [1, 1, 1, 1], value=0)
+ t_712 = self.n_Conv_153(t_711_padded)
+ t_713 = F.relu(t_712)
+ t_714 = self.n_Conv_154(t_713)
+ t_715 = torch.add(t_714, t_709)
+ t_716 = F.relu(t_715)
+ t_717 = self.n_Conv_155(t_716)
+ t_718 = F.relu(t_717)
+ t_718_padded = F.pad(t_718, [1, 1, 1, 1], value=0)
+ t_719 = self.n_Conv_156(t_718_padded)
+ t_720 = F.relu(t_719)
+ t_721 = self.n_Conv_157(t_720)
+ t_722 = torch.add(t_721, t_716)
+ t_723 = F.relu(t_722)
+ t_724 = self.n_Conv_158(t_723)
+ t_725 = self.n_Conv_159(t_723)
+ t_726 = F.relu(t_725)
+ t_726_padded = F.pad(t_726, [0, 1, 0, 1], value=0)
+ t_727 = self.n_Conv_160(t_726_padded)
+ t_728 = F.relu(t_727)
+ t_729 = self.n_Conv_161(t_728)
+ t_730 = torch.add(t_729, t_724)
+ t_731 = F.relu(t_730)
+ t_732 = self.n_Conv_162(t_731)
+ t_733 = F.relu(t_732)
+ t_733_padded = F.pad(t_733, [1, 1, 1, 1], value=0)
+ t_734 = self.n_Conv_163(t_733_padded)
+ t_735 = F.relu(t_734)
+ t_736 = self.n_Conv_164(t_735)
+ t_737 = torch.add(t_736, t_731)
+ t_738 = F.relu(t_737)
+ t_739 = self.n_Conv_165(t_738)
+ t_740 = F.relu(t_739)
+ t_740_padded = F.pad(t_740, [1, 1, 1, 1], value=0)
+ t_741 = self.n_Conv_166(t_740_padded)
+ t_742 = F.relu(t_741)
+ t_743 = self.n_Conv_167(t_742)
+ t_744 = torch.add(t_743, t_738)
+ t_745 = F.relu(t_744)
+ t_746 = self.n_Conv_168(t_745)
+ t_747 = self.n_Conv_169(t_745)
+ t_748 = F.relu(t_747)
+ t_748_padded = F.pad(t_748, [0, 1, 0, 1], value=0)
+ t_749 = self.n_Conv_170(t_748_padded)
+ t_750 = F.relu(t_749)
+ t_751 = self.n_Conv_171(t_750)
+ t_752 = torch.add(t_751, t_746)
+ t_753 = F.relu(t_752)
+ t_754 = self.n_Conv_172(t_753)
+ t_755 = F.relu(t_754)
+ t_755_padded = F.pad(t_755, [1, 1, 1, 1], value=0)
+ t_756 = self.n_Conv_173(t_755_padded)
+ t_757 = F.relu(t_756)
+ t_758 = self.n_Conv_174(t_757)
+ t_759 = torch.add(t_758, t_753)
+ t_760 = F.relu(t_759)
+ t_761 = self.n_Conv_175(t_760)
+ t_762 = F.relu(t_761)
+ t_762_padded = F.pad(t_762, [1, 1, 1, 1], value=0)
+ t_763 = self.n_Conv_176(t_762_padded)
+ t_764 = F.relu(t_763)
+ t_765 = self.n_Conv_177(t_764)
+ t_766 = torch.add(t_765, t_760)
+ t_767 = F.relu(t_766)
+ t_768 = self.n_Conv_178(t_767)
+ t_769 = F.avg_pool2d(t_768, kernel_size=t_768.shape[-2:])
+ t_770 = torch.squeeze(t_769, 3)
+ t_770 = torch.squeeze(t_770, 2)
+ t_771 = torch.sigmoid(t_770)
+ return t_771
+
+ def load_state_dict(self, state_dict, **kwargs):
+ self.tags = state_dict.get('tags', [])
+
+ super(DeepDanbooruModel, self).load_state_dict({k: v for k, v in state_dict.items() if k != 'tags'})
+
diff --git a/modules/images.py b/modules/images.py
index 26d5b7a9..8737ccff 100644
--- a/modules/images.py
+++ b/modules/images.py
@@ -524,6 +524,8 @@ def save_image(image, path, basename, seed=None, prompt=None, extension='png', i
else:
image.save(fullfn, quality=opts.jpeg_quality)
+ image.already_saved_as = fullfn
+
target_side_length = 4000
oversize = image.width > target_side_length or image.height > target_side_length
if opts.export_for_4chan and (oversize or os.stat(fullfn).st_size > 4 * 1024 * 1024):
diff --git a/modules/lowvram.py b/modules/lowvram.py
index a4652cb1..aa464a95 100644
--- a/modules/lowvram.py
+++ b/modules/lowvram.py
@@ -51,6 +51,10 @@ def setup_for_low_vram(sd_model, use_medvram):
send_me_to_gpu(first_stage_model, None)
return first_stage_model_decode(z)
+ # for SD1, cond_stage_model is CLIP and its NN is in the tranformer frield, but for SD2, it's open clip, and it's in model field
+ if hasattr(sd_model.cond_stage_model, 'model'):
+ sd_model.cond_stage_model.transformer = sd_model.cond_stage_model.model
+
# remove three big modules, cond, first_stage, and unet from the model and then
# send the model to GPU. Then put modules back. the modules will be in CPU.
stored = sd_model.cond_stage_model.transformer, sd_model.first_stage_model, sd_model.model
@@ -65,6 +69,10 @@ def setup_for_low_vram(sd_model, use_medvram):
sd_model.first_stage_model.decode = first_stage_model_decode_wrap
parents[sd_model.cond_stage_model.transformer] = sd_model.cond_stage_model
+ if hasattr(sd_model.cond_stage_model, 'model'):
+ sd_model.cond_stage_model.model = sd_model.cond_stage_model.transformer
+ del sd_model.cond_stage_model.transformer
+
if use_medvram:
sd_model.model.register_forward_pre_hook(send_me_to_gpu)
else:
diff --git a/modules/paths.py b/modules/paths.py
index 1e7a2fbc..4dd03a35 100644
--- a/modules/paths.py
+++ b/modules/paths.py
@@ -9,7 +9,7 @@ sys.path.insert(0, script_path)
# search for directory of stable diffusion in following places
sd_path = None
-possible_sd_paths = [os.path.join(script_path, 'repositories/stable-diffusion'), '.', os.path.dirname(script_path)]
+possible_sd_paths = [os.path.join(script_path, 'repositories/stable-diffusion-stability-ai'), '.', os.path.dirname(script_path)]
for possible_sd_path in possible_sd_paths:
if os.path.exists(os.path.join(possible_sd_path, 'ldm/models/diffusion/ddpm.py')):
sd_path = os.path.abspath(possible_sd_path)
diff --git a/modules/processing.py b/modules/processing.py
index accb31d1..c310df6a 100644
--- a/modules/processing.py
+++ b/modules/processing.py
@@ -424,13 +424,15 @@ def process_images(p: StableDiffusionProcessing) -> Processed:
try:
for k, v in p.override_settings.items():
- setattr(opts, k, v) # we don't call onchange for simplicity which makes changing model, hypernet impossible
+ setattr(opts, k, v) # we don't call onchange for simplicity which makes changing model impossible
+ if k == 'sd_hypernetwork': shared.reload_hypernetworks() # make onchange call for changing hypernet since it is relatively fast to load on-change, while SD models are not
res = process_images_inner(p)
- finally:
+ finally: # restore opts to original state
for k, v in stored_opts.items():
setattr(opts, k, v)
+ if k == 'sd_hypernetwork': shared.reload_hypernetworks()
return res
diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py
index 29c8b561..b824b5bf 100644
--- a/modules/sd_hijack.py
+++ b/modules/sd_hijack.py
@@ -9,18 +9,30 @@ from torch.nn.functional import silu
import modules.textual_inversion.textual_inversion
from modules import prompt_parser, devices, sd_hijack_optimizations, shared, sd_hijack_checkpoint
+from modules.hypernetworks import hypernetwork
from modules.shared import opts, device, cmd_opts
+from modules import sd_hijack_clip, sd_hijack_open_clip
+
from modules.sd_hijack_optimizations import invokeAI_mps_available
import ldm.modules.attention
import ldm.modules.diffusionmodules.model
import ldm.models.diffusion.ddim
import ldm.models.diffusion.plms
+import ldm.modules.encoders.modules
attention_CrossAttention_forward = ldm.modules.attention.CrossAttention.forward
diffusionmodules_model_nonlinearity = ldm.modules.diffusionmodules.model.nonlinearity
diffusionmodules_model_AttnBlock_forward = ldm.modules.diffusionmodules.model.AttnBlock.forward
+# new memory efficient cross attention blocks do not support hypernets and we already
+# have memory efficient cross attention anyway, so this disables SD2.0's memory efficient cross attention
+ldm.modules.attention.MemoryEfficientCrossAttention = ldm.modules.attention.CrossAttention
+ldm.modules.attention.BasicTransformerBlock.ATTENTION_MODES["softmax-xformers"] = ldm.modules.attention.CrossAttention
+
+# silence new console spam from SD2
+ldm.modules.attention.print = lambda *args: None
+ldm.modules.diffusionmodules.model.print = lambda *args: None
def apply_optimizations():
undo_optimizations()
@@ -49,16 +61,11 @@ def apply_optimizations():
def undo_optimizations():
- from modules.hypernetworks import hypernetwork
-
ldm.modules.attention.CrossAttention.forward = hypernetwork.attention_CrossAttention_forward
ldm.modules.diffusionmodules.model.nonlinearity = diffusionmodules_model_nonlinearity
ldm.modules.diffusionmodules.model.AttnBlock.forward = diffusionmodules_model_AttnBlock_forward
-def get_target_prompt_token_count(token_count):
- return math.ceil(max(token_count, 1) / 75) * 75
-
def fix_checkpoint():
ldm.modules.attention.BasicTransformerBlock.forward = sd_hijack_checkpoint.BasicTransformerBlock_forward
ldm.modules.diffusionmodules.openaimodel.ResBlock.forward = sd_hijack_checkpoint.ResBlock_forward
@@ -74,10 +81,13 @@ class StableDiffusionModelHijack:
embedding_db = modules.textual_inversion.textual_inversion.EmbeddingDatabase(cmd_opts.embeddings_dir)
def hijack(self, m):
- model_embeddings = m.cond_stage_model.transformer.text_model.embeddings
-
- model_embeddings.token_embedding = EmbeddingsWithFixes(model_embeddings.token_embedding, self)
- m.cond_stage_model = FrozenCLIPEmbedderWithCustomWords(m.cond_stage_model, self)
+ if type(m.cond_stage_model) == ldm.modules.encoders.modules.FrozenCLIPEmbedder:
+ model_embeddings = m.cond_stage_model.transformer.text_model.embeddings
+ model_embeddings.token_embedding = EmbeddingsWithFixes(model_embeddings.token_embedding, self)
+ m.cond_stage_model = sd_hijack_clip.FrozenCLIPEmbedderWithCustomWords(m.cond_stage_model, self)
+ elif type(m.cond_stage_model) == ldm.modules.encoders.modules.FrozenOpenCLIPEmbedder:
+ m.cond_stage_model.model.token_embedding = EmbeddingsWithFixes(m.cond_stage_model.model.token_embedding, self)
+ m.cond_stage_model = sd_hijack_open_clip.FrozenOpenCLIPEmbedderWithCustomWords(m.cond_stage_model, self)
self.clip = m.cond_stage_model
@@ -94,12 +104,15 @@ class StableDiffusionModelHijack:
self.layers = flatten(m)
def undo_hijack(self, m):
- if type(m.cond_stage_model) == FrozenCLIPEmbedderWithCustomWords:
+ if type(m.cond_stage_model) == sd_hijack_clip.FrozenCLIPEmbedderWithCustomWords:
m.cond_stage_model = m.cond_stage_model.wrapped
- model_embeddings = m.cond_stage_model.transformer.text_model.embeddings
- if type(model_embeddings.token_embedding) == EmbeddingsWithFixes:
- model_embeddings.token_embedding = model_embeddings.token_embedding.wrapped
+ model_embeddings = m.cond_stage_model.transformer.text_model.embeddings
+ if type(model_embeddings.token_embedding) == EmbeddingsWithFixes:
+ model_embeddings.token_embedding = model_embeddings.token_embedding.wrapped
+ elif type(m.cond_stage_model) == sd_hijack_open_clip.FrozenOpenCLIPEmbedderWithCustomWords:
+ m.cond_stage_model.wrapped.model.token_embedding = m.cond_stage_model.wrapped.model.token_embedding.wrapped
+ m.cond_stage_model = m.cond_stage_model.wrapped
self.apply_circular(False)
self.layers = None
@@ -119,261 +132,8 @@ class StableDiffusionModelHijack:
def tokenize(self, text):
_, remade_batch_tokens, _, _, _, token_count = self.clip.process_text([text])
- return remade_batch_tokens[0], token_count, get_target_prompt_token_count(token_count)
-
-
-class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
- def __init__(self, wrapped, hijack):
- super().__init__()
- self.wrapped = wrapped
- self.hijack: StableDiffusionModelHijack = hijack
- self.tokenizer = wrapped.tokenizer
- self.token_mults = {}
-
- self.comma_token = [v for k, v in self.tokenizer.get_vocab().items() if k == ',</w>'][0]
-
- tokens_with_parens = [(k, v) for k, v in self.tokenizer.get_vocab().items() if '(' in k or ')' in k or '[' in k or ']' in k]
- for text, ident in tokens_with_parens:
- mult = 1.0
- for c in text:
- if c == '[':
- mult /= 1.1
- if c == ']':
- mult *= 1.1
- if c == '(':
- mult *= 1.1
- if c == ')':
- mult /= 1.1
-
- if mult != 1.0:
- self.token_mults[ident] = mult
-
- def tokenize_line(self, line, used_custom_terms, hijack_comments):
- id_end = self.wrapped.tokenizer.eos_token_id
-
- if opts.enable_emphasis:
- parsed = prompt_parser.parse_prompt_attention(line)
- else:
- parsed = [[line, 1.0]]
-
- tokenized = self.wrapped.tokenizer([text for text, _ in parsed], truncation=False, add_special_tokens=False)["input_ids"]
-
- fixes = []
- remade_tokens = []
- multipliers = []
- last_comma = -1
-
- for tokens, (text, weight) in zip(tokenized, parsed):
- i = 0
- while i < len(tokens):
- token = tokens[i]
-
- embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
-
- if token == self.comma_token:
- last_comma = len(remade_tokens)
- elif opts.comma_padding_backtrack != 0 and max(len(remade_tokens), 1) % 75 == 0 and last_comma != -1 and len(remade_tokens) - last_comma <= opts.comma_padding_backtrack:
- last_comma += 1
- reloc_tokens = remade_tokens[last_comma:]
- reloc_mults = multipliers[last_comma:]
-
- remade_tokens = remade_tokens[:last_comma]
- length = len(remade_tokens)
-
- rem = int(math.ceil(length / 75)) * 75 - length
- remade_tokens += [id_end] * rem + reloc_tokens
- multipliers = multipliers[:last_comma] + [1.0] * rem + reloc_mults
-
- if embedding is None:
- remade_tokens.append(token)
- multipliers.append(weight)
- i += 1
- else:
- emb_len = int(embedding.vec.shape[0])
- iteration = len(remade_tokens) // 75
- if (len(remade_tokens) + emb_len) // 75 != iteration:
- rem = (75 * (iteration + 1) - len(remade_tokens))
- remade_tokens += [id_end] * rem
- multipliers += [1.0] * rem
- iteration += 1
- fixes.append((iteration, (len(remade_tokens) % 75, embedding)))
- remade_tokens += [0] * emb_len
- multipliers += [weight] * emb_len
- used_custom_terms.append((embedding.name, embedding.checksum()))
- i += embedding_length_in_tokens
-
- token_count = len(remade_tokens)
- prompt_target_length = get_target_prompt_token_count(token_count)
- tokens_to_add = prompt_target_length - len(remade_tokens)
-
- remade_tokens = remade_tokens + [id_end] * tokens_to_add
- multipliers = multipliers + [1.0] * tokens_to_add
-
- return remade_tokens, fixes, multipliers, token_count
-
- def process_text(self, texts):
- used_custom_terms = []
- remade_batch_tokens = []
- hijack_comments = []
- hijack_fixes = []
- token_count = 0
-
- cache = {}
- batch_multipliers = []
- for line in texts:
- if line in cache:
- remade_tokens, fixes, multipliers = cache[line]
- else:
- remade_tokens, fixes, multipliers, current_token_count = self.tokenize_line(line, used_custom_terms, hijack_comments)
- token_count = max(current_token_count, token_count)
-
- cache[line] = (remade_tokens, fixes, multipliers)
-
- remade_batch_tokens.append(remade_tokens)
- hijack_fixes.append(fixes)
- batch_multipliers.append(multipliers)
-
- return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
-
- def process_text_old(self, text):
- id_start = self.wrapped.tokenizer.bos_token_id
- id_end = self.wrapped.tokenizer.eos_token_id
- maxlen = self.wrapped.max_length # you get to stay at 77
- used_custom_terms = []
- remade_batch_tokens = []
- overflowing_words = []
- hijack_comments = []
- hijack_fixes = []
- token_count = 0
-
- cache = {}
- batch_tokens = self.wrapped.tokenizer(text, truncation=False, add_special_tokens=False)["input_ids"]
- batch_multipliers = []
- for tokens in batch_tokens:
- tuple_tokens = tuple(tokens)
-
- if tuple_tokens in cache:
- remade_tokens, fixes, multipliers = cache[tuple_tokens]
- else:
- fixes = []
- remade_tokens = []
- multipliers = []
- mult = 1.0
-
- i = 0
- while i < len(tokens):
- token = tokens[i]
-
- embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
-
- mult_change = self.token_mults.get(token) if opts.enable_emphasis else None
- if mult_change is not None:
- mult *= mult_change
- i += 1
- elif embedding is None:
- remade_tokens.append(token)
- multipliers.append(mult)
- i += 1
- else:
- emb_len = int(embedding.vec.shape[0])
- fixes.append((len(remade_tokens), embedding))
- remade_tokens += [0] * emb_len
- multipliers += [mult] * emb_len
- used_custom_terms.append((embedding.name, embedding.checksum()))
- i += embedding_length_in_tokens
-
- if len(remade_tokens) > maxlen - 2:
- vocab = {v: k for k, v in self.wrapped.tokenizer.get_vocab().items()}
- ovf = remade_tokens[maxlen - 2:]
- overflowing_words = [vocab.get(int(x), "") for x in ovf]
- overflowing_text = self.wrapped.tokenizer.convert_tokens_to_string(''.join(overflowing_words))
- hijack_comments.append(f"Warning: too many input tokens; some ({len(overflowing_words)}) have been truncated:\n{overflowing_text}\n")
-
- token_count = len(remade_tokens)
- remade_tokens = remade_tokens + [id_end] * (maxlen - 2 - len(remade_tokens))
- remade_tokens = [id_start] + remade_tokens[0:maxlen - 2] + [id_end]
- cache[tuple_tokens] = (remade_tokens, fixes, multipliers)
-
- multipliers = multipliers + [1.0] * (maxlen - 2 - len(multipliers))
- multipliers = [1.0] + multipliers[0:maxlen - 2] + [1.0]
-
- remade_batch_tokens.append(remade_tokens)
- hijack_fixes.append(fixes)
- batch_multipliers.append(multipliers)
- return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
-
- def forward(self, text):
- use_old = opts.use_old_emphasis_implementation
- if use_old:
- batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text_old(text)
- else:
- batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text(text)
-
- self.hijack.comments += hijack_comments
-
- if len(used_custom_terms) > 0:
- self.hijack.comments.append("Used embeddings: " + ", ".join([f'{word} [{checksum}]' for word, checksum in used_custom_terms]))
-
- if use_old:
- self.hijack.fixes = hijack_fixes
- return self.process_tokens(remade_batch_tokens, batch_multipliers)
-
- z = None
- i = 0
- while max(map(len, remade_batch_tokens)) != 0:
- rem_tokens = [x[75:] for x in remade_batch_tokens]
- rem_multipliers = [x[75:] for x in batch_multipliers]
-
- self.hijack.fixes = []
- for unfiltered in hijack_fixes:
- fixes = []
- for fix in unfiltered:
- if fix[0] == i:
- fixes.append(fix[1])
- self.hijack.fixes.append(fixes)
-
- tokens = []
- multipliers = []
- for j in range(len(remade_batch_tokens)):
- if len(remade_batch_tokens[j]) > 0:
- tokens.append(remade_batch_tokens[j][:75])
- multipliers.append(batch_multipliers[j][:75])
- else:
- tokens.append([self.wrapped.tokenizer.eos_token_id] * 75)
- multipliers.append([1.0] * 75)
-
- z1 = self.process_tokens(tokens, multipliers)
- z = z1 if z is None else torch.cat((z, z1), axis=-2)
-
- remade_batch_tokens = rem_tokens
- batch_multipliers = rem_multipliers
- i += 1
-
- return z
-
- def process_tokens(self, remade_batch_tokens, batch_multipliers):
- if not opts.use_old_emphasis_implementation:
- remade_batch_tokens = [[self.wrapped.tokenizer.bos_token_id] + x[:75] + [self.wrapped.tokenizer.eos_token_id] for x in remade_batch_tokens]
- batch_multipliers = [[1.0] + x[:75] + [1.0] for x in batch_multipliers]
-
- tokens = torch.asarray(remade_batch_tokens).to(device)
- outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=-opts.CLIP_stop_at_last_layers)
-
- if opts.CLIP_stop_at_last_layers > 1:
- z = outputs.hidden_states[-opts.CLIP_stop_at_last_layers]
- z = self.wrapped.transformer.text_model.final_layer_norm(z)
- else:
- z = outputs.last_hidden_state
-
- # restoring original mean is likely not correct, but it seems to work well to prevent artifacts that happen otherwise
- batch_multipliers_of_same_length = [x + [1.0] * (75 - len(x)) for x in batch_multipliers]
- batch_multipliers = torch.asarray(batch_multipliers_of_same_length).to(device)
- original_mean = z.mean()
- z *= batch_multipliers.reshape(batch_multipliers.shape + (1,)).expand(z.shape)
- new_mean = z.mean()
- z *= original_mean / new_mean
+ return remade_batch_tokens[0], token_count, sd_hijack_clip.get_target_prompt_token_count(token_count)
- return z
class EmbeddingsWithFixes(torch.nn.Module):
diff --git a/modules/sd_hijack_clip.py b/modules/sd_hijack_clip.py
new file mode 100644
index 00000000..b451d1cf
--- /dev/null
+++ b/modules/sd_hijack_clip.py
@@ -0,0 +1,301 @@
+import math
+
+import torch
+
+from modules import prompt_parser, devices
+from modules.shared import opts
+
+
+def get_target_prompt_token_count(token_count):
+ return math.ceil(max(token_count, 1) / 75) * 75
+
+
+class FrozenCLIPEmbedderWithCustomWordsBase(torch.nn.Module):
+ def __init__(self, wrapped, hijack):
+ super().__init__()
+ self.wrapped = wrapped
+ self.hijack = hijack
+
+ def tokenize(self, texts):
+ raise NotImplementedError
+
+ def encode_with_transformers(self, tokens):
+ raise NotImplementedError
+
+ def encode_embedding_init_text(self, init_text, nvpt):
+ raise NotImplementedError
+
+ def tokenize_line(self, line, used_custom_terms, hijack_comments):
+ if opts.enable_emphasis:
+ parsed = prompt_parser.parse_prompt_attention(line)
+ else:
+ parsed = [[line, 1.0]]
+
+ tokenized = self.tokenize([text for text, _ in parsed])
+
+ fixes = []
+ remade_tokens = []
+ multipliers = []
+ last_comma = -1
+
+ for tokens, (text, weight) in zip(tokenized, parsed):
+ i = 0
+ while i < len(tokens):
+ token = tokens[i]
+
+ embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
+
+ if token == self.comma_token:
+ last_comma = len(remade_tokens)
+ elif opts.comma_padding_backtrack != 0 and max(len(remade_tokens), 1) % 75 == 0 and last_comma != -1 and len(remade_tokens) - last_comma <= opts.comma_padding_backtrack:
+ last_comma += 1
+ reloc_tokens = remade_tokens[last_comma:]
+ reloc_mults = multipliers[last_comma:]
+
+ remade_tokens = remade_tokens[:last_comma]
+ length = len(remade_tokens)
+
+ rem = int(math.ceil(length / 75)) * 75 - length
+ remade_tokens += [self.id_end] * rem + reloc_tokens
+ multipliers = multipliers[:last_comma] + [1.0] * rem + reloc_mults
+
+ if embedding is None:
+ remade_tokens.append(token)
+ multipliers.append(weight)
+ i += 1
+ else:
+ emb_len = int(embedding.vec.shape[0])
+ iteration = len(remade_tokens) // 75
+ if (len(remade_tokens) + emb_len) // 75 != iteration:
+ rem = (75 * (iteration + 1) - len(remade_tokens))
+ remade_tokens += [self.id_end] * rem
+ multipliers += [1.0] * rem
+ iteration += 1
+ fixes.append((iteration, (len(remade_tokens) % 75, embedding)))
+ remade_tokens += [0] * emb_len
+ multipliers += [weight] * emb_len
+ used_custom_terms.append((embedding.name, embedding.checksum()))
+ i += embedding_length_in_tokens
+
+ token_count = len(remade_tokens)
+ prompt_target_length = get_target_prompt_token_count(token_count)
+ tokens_to_add = prompt_target_length - len(remade_tokens)
+
+ remade_tokens = remade_tokens + [self.id_end] * tokens_to_add
+ multipliers = multipliers + [1.0] * tokens_to_add
+
+ return remade_tokens, fixes, multipliers, token_count
+
+ def process_text(self, texts):
+ used_custom_terms = []
+ remade_batch_tokens = []
+ hijack_comments = []
+ hijack_fixes = []
+ token_count = 0
+
+ cache = {}
+ batch_multipliers = []
+ for line in texts:
+ if line in cache:
+ remade_tokens, fixes, multipliers = cache[line]
+ else:
+ remade_tokens, fixes, multipliers, current_token_count = self.tokenize_line(line, used_custom_terms, hijack_comments)
+ token_count = max(current_token_count, token_count)
+
+ cache[line] = (remade_tokens, fixes, multipliers)
+
+ remade_batch_tokens.append(remade_tokens)
+ hijack_fixes.append(fixes)
+ batch_multipliers.append(multipliers)
+
+ return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
+
+ def process_text_old(self, texts):
+ id_start = self.id_start
+ id_end = self.id_end
+ maxlen = self.wrapped.max_length # you get to stay at 77
+ used_custom_terms = []
+ remade_batch_tokens = []
+ hijack_comments = []
+ hijack_fixes = []
+ token_count = 0
+
+ cache = {}
+ batch_tokens = self.tokenize(texts)
+ batch_multipliers = []
+ for tokens in batch_tokens:
+ tuple_tokens = tuple(tokens)
+
+ if tuple_tokens in cache:
+ remade_tokens, fixes, multipliers = cache[tuple_tokens]
+ else:
+ fixes = []
+ remade_tokens = []
+ multipliers = []
+ mult = 1.0
+
+ i = 0
+ while i < len(tokens):
+ token = tokens[i]
+
+ embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
+
+ mult_change = self.token_mults.get(token) if opts.enable_emphasis else None
+ if mult_change is not None:
+ mult *= mult_change
+ i += 1
+ elif embedding is None:
+ remade_tokens.append(token)
+ multipliers.append(mult)
+ i += 1
+ else:
+ emb_len = int(embedding.vec.shape[0])
+ fixes.append((len(remade_tokens), embedding))
+ remade_tokens += [0] * emb_len
+ multipliers += [mult] * emb_len
+ used_custom_terms.append((embedding.name, embedding.checksum()))
+ i += embedding_length_in_tokens
+
+ if len(remade_tokens) > maxlen - 2:
+ vocab = {v: k for k, v in self.wrapped.tokenizer.get_vocab().items()}
+ ovf = remade_tokens[maxlen - 2:]
+ overflowing_words = [vocab.get(int(x), "") for x in ovf]
+ overflowing_text = self.wrapped.tokenizer.convert_tokens_to_string(''.join(overflowing_words))
+ hijack_comments.append(f"Warning: too many input tokens; some ({len(overflowing_words)}) have been truncated:\n{overflowing_text}\n")
+
+ token_count = len(remade_tokens)
+ remade_tokens = remade_tokens + [id_end] * (maxlen - 2 - len(remade_tokens))
+ remade_tokens = [id_start] + remade_tokens[0:maxlen - 2] + [id_end]
+ cache[tuple_tokens] = (remade_tokens, fixes, multipliers)
+
+ multipliers = multipliers + [1.0] * (maxlen - 2 - len(multipliers))
+ multipliers = [1.0] + multipliers[0:maxlen - 2] + [1.0]
+
+ remade_batch_tokens.append(remade_tokens)
+ hijack_fixes.append(fixes)
+ batch_multipliers.append(multipliers)
+ return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
+
+ def forward(self, text):
+ use_old = opts.use_old_emphasis_implementation
+ if use_old:
+ batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text_old(text)
+ else:
+ batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text(text)
+
+ self.hijack.comments += hijack_comments
+
+ if len(used_custom_terms) > 0:
+ self.hijack.comments.append("Used embeddings: " + ", ".join([f'{word} [{checksum}]' for word, checksum in used_custom_terms]))
+
+ if use_old:
+ self.hijack.fixes = hijack_fixes
+ return self.process_tokens(remade_batch_tokens, batch_multipliers)
+
+ z = None
+ i = 0
+ while max(map(len, remade_batch_tokens)) != 0:
+ rem_tokens = [x[75:] for x in remade_batch_tokens]
+ rem_multipliers = [x[75:] for x in batch_multipliers]
+
+ self.hijack.fixes = []
+ for unfiltered in hijack_fixes:
+ fixes = []
+ for fix in unfiltered:
+ if fix[0] == i:
+ fixes.append(fix[1])
+ self.hijack.fixes.append(fixes)
+
+ tokens = []
+ multipliers = []
+ for j in range(len(remade_batch_tokens)):
+ if len(remade_batch_tokens[j]) > 0:
+ tokens.append(remade_batch_tokens[j][:75])
+ multipliers.append(batch_multipliers[j][:75])
+ else:
+ tokens.append([self.id_end] * 75)
+ multipliers.append([1.0] * 75)
+
+ z1 = self.process_tokens(tokens, multipliers)
+ z = z1 if z is None else torch.cat((z, z1), axis=-2)
+
+ remade_batch_tokens = rem_tokens
+ batch_multipliers = rem_multipliers
+ i += 1
+
+ return z
+
+ def process_tokens(self, remade_batch_tokens, batch_multipliers):
+ if not opts.use_old_emphasis_implementation:
+ remade_batch_tokens = [[self.id_start] + x[:75] + [self.id_end] for x in remade_batch_tokens]
+ batch_multipliers = [[1.0] + x[:75] + [1.0] for x in batch_multipliers]
+
+ tokens = torch.asarray(remade_batch_tokens).to(devices.device)
+
+ if self.id_end != self.id_pad:
+ for batch_pos in range(len(remade_batch_tokens)):
+ index = remade_batch_tokens[batch_pos].index(self.id_end)
+ tokens[batch_pos, index+1:tokens.shape[1]] = self.id_pad
+
+ z = self.encode_with_transformers(tokens)
+
+ # restoring original mean is likely not correct, but it seems to work well to prevent artifacts that happen otherwise
+ batch_multipliers_of_same_length = [x + [1.0] * (75 - len(x)) for x in batch_multipliers]
+ batch_multipliers = torch.asarray(batch_multipliers_of_same_length).to(devices.device)
+ original_mean = z.mean()
+ z *= batch_multipliers.reshape(batch_multipliers.shape + (1,)).expand(z.shape)
+ new_mean = z.mean()
+ z *= original_mean / new_mean
+
+ return z
+
+
+class FrozenCLIPEmbedderWithCustomWords(FrozenCLIPEmbedderWithCustomWordsBase):
+ def __init__(self, wrapped, hijack):
+ super().__init__(wrapped, hijack)
+ self.tokenizer = wrapped.tokenizer
+ self.comma_token = [v for k, v in self.tokenizer.get_vocab().items() if k == ',</w>'][0]
+
+ self.token_mults = {}
+ tokens_with_parens = [(k, v) for k, v in self.tokenizer.get_vocab().items() if '(' in k or ')' in k or '[' in k or ']' in k]
+ for text, ident in tokens_with_parens:
+ mult = 1.0
+ for c in text:
+ if c == '[':
+ mult /= 1.1
+ if c == ']':
+ mult *= 1.1
+ if c == '(':
+ mult *= 1.1
+ if c == ')':
+ mult /= 1.1
+
+ if mult != 1.0:
+ self.token_mults[ident] = mult
+
+ self.id_start = self.wrapped.tokenizer.bos_token_id
+ self.id_end = self.wrapped.tokenizer.eos_token_id
+ self.id_pad = self.id_end
+
+ def tokenize(self, texts):
+ tokenized = self.wrapped.tokenizer(texts, truncation=False, add_special_tokens=False)["input_ids"]
+
+ return tokenized
+
+ def encode_with_transformers(self, tokens):
+ outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=-opts.CLIP_stop_at_last_layers)
+
+ if opts.CLIP_stop_at_last_layers > 1:
+ z = outputs.hidden_states[-opts.CLIP_stop_at_last_layers]
+ z = self.wrapped.transformer.text_model.final_layer_norm(z)
+ else:
+ z = outputs.last_hidden_state
+
+ return z
+
+ def encode_embedding_init_text(self, init_text, nvpt):
+ embedding_layer = self.wrapped.transformer.text_model.embeddings
+ ids = self.wrapped.tokenizer(init_text, max_length=nvpt, return_tensors="pt", add_special_tokens=False)["input_ids"]
+ embedded = embedding_layer.token_embedding.wrapped(ids.to(devices.device)).squeeze(0)
+
+ return embedded
diff --git a/modules/sd_hijack_inpainting.py b/modules/sd_hijack_inpainting.py
index 46714a4f..938f9a58 100644
--- a/modules/sd_hijack_inpainting.py
+++ b/modules/sd_hijack_inpainting.py
@@ -199,8 +199,8 @@ def sample_plms(self,
@torch.no_grad()
def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
- temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
- unconditional_guidance_scale=1., unconditional_conditioning=None, old_eps=None, t_next=None):
+ temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+ unconditional_guidance_scale=1., unconditional_conditioning=None, old_eps=None, t_next=None, dynamic_threshold=None):
b, *_, device = *x.shape, x.device
def get_model_output(x, t):
@@ -249,6 +249,8 @@ def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=F
pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
if quantize_denoised:
pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+ if dynamic_threshold is not None:
+ pred_x0 = norm_thresholding(pred_x0, dynamic_threshold)
# direction pointing to x_t
dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
@@ -321,12 +323,16 @@ def should_hijack_inpainting(checkpoint_info):
def do_inpainting_hijack():
- ldm.models.diffusion.ddpm.get_unconditional_conditioning = get_unconditional_conditioning
+ # most of this stuff seems to no longer be needed because it is already included into SD2.0
+ # LatentInpaintDiffusion remains because SD2.0's LatentInpaintDiffusion can't be loaded without specifying a checkpoint
+ # p_sample_plms is needed because PLMS can't work with dicts as conditionings
+ # this file should be cleaned up later if weverything tuens out to work fine
+
+ # ldm.models.diffusion.ddpm.get_unconditional_conditioning = get_unconditional_conditioning
ldm.models.diffusion.ddpm.LatentInpaintDiffusion = LatentInpaintDiffusion
- ldm.models.diffusion.ddim.DDIMSampler.p_sample_ddim = p_sample_ddim
- ldm.models.diffusion.ddim.DDIMSampler.sample = sample_ddim
+ # ldm.models.diffusion.ddim.DDIMSampler.p_sample_ddim = p_sample_ddim
+ # ldm.models.diffusion.ddim.DDIMSampler.sample = sample_ddim
ldm.models.diffusion.plms.PLMSSampler.p_sample_plms = p_sample_plms
- ldm.models.diffusion.plms.PLMSSampler.sample = sample_plms
-
+ # ldm.models.diffusion.plms.PLMSSampler.sample = sample_plms
diff --git a/modules/sd_hijack_open_clip.py b/modules/sd_hijack_open_clip.py
new file mode 100644
index 00000000..f733e852
--- /dev/null
+++ b/modules/sd_hijack_open_clip.py
@@ -0,0 +1,37 @@
+import open_clip.tokenizer
+import torch
+
+from modules import sd_hijack_clip, devices
+from modules.shared import opts
+
+tokenizer = open_clip.tokenizer._tokenizer
+
+
+class FrozenOpenCLIPEmbedderWithCustomWords(sd_hijack_clip.FrozenCLIPEmbedderWithCustomWordsBase):
+ def __init__(self, wrapped, hijack):
+ super().__init__(wrapped, hijack)
+
+ self.comma_token = [v for k, v in tokenizer.encoder.items() if k == ',</w>'][0]
+ self.id_start = tokenizer.encoder["<start_of_text>"]
+ self.id_end = tokenizer.encoder["<end_of_text>"]
+ self.id_pad = 0
+
+ def tokenize(self, texts):
+ assert not opts.use_old_emphasis_implementation, 'Old emphasis implementation not supported for Open Clip'
+
+ tokenized = [tokenizer.encode(text) for text in texts]
+
+ return tokenized
+
+ def encode_with_transformers(self, tokens):
+ # set self.wrapped.layer_idx here according to opts.CLIP_stop_at_last_layers
+ z = self.wrapped.encode_with_transformer(tokens)
+
+ return z
+
+ def encode_embedding_init_text(self, init_text, nvpt):
+ ids = tokenizer.encode(init_text)
+ ids = torch.asarray([ids], device=devices.device, dtype=torch.int)
+ embedded = self.wrapped.model.token_embedding.wrapped(ids).squeeze(0)
+
+ return embedded
diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py
index 4fe67854..4edd8c60 100644
--- a/modules/sd_samplers.py
+++ b/modules/sd_samplers.py
@@ -127,7 +127,8 @@ class InterruptedException(BaseException):
class VanillaStableDiffusionSampler:
def __init__(self, constructor, sd_model):
self.sampler = constructor(sd_model)
- self.orig_p_sample_ddim = self.sampler.p_sample_ddim if hasattr(self.sampler, 'p_sample_ddim') else self.sampler.p_sample_plms
+ self.is_plms = hasattr(self.sampler, 'p_sample_plms')
+ self.orig_p_sample_ddim = self.sampler.p_sample_plms if self.is_plms else self.sampler.p_sample_ddim
self.mask = None
self.nmask = None
self.init_latent = None
@@ -218,7 +219,6 @@ class VanillaStableDiffusionSampler:
self.mask = p.mask if hasattr(p, 'mask') else None
self.nmask = p.nmask if hasattr(p, 'nmask') else None
-
def adjust_steps_if_invalid(self, p, num_steps):
if (self.config.name == 'DDIM' and p.ddim_discretize == 'uniform') or (self.config.name == 'PLMS'):
valid_step = 999 / (1000 // num_steps)
@@ -227,7 +227,6 @@ class VanillaStableDiffusionSampler:
return num_steps
-
def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning, steps=None, image_conditioning=None):
steps, t_enc = setup_img2img_steps(p, steps)
steps = self.adjust_steps_if_invalid(p, steps)
@@ -260,9 +259,10 @@ class VanillaStableDiffusionSampler:
steps = self.adjust_steps_if_invalid(p, steps or p.steps)
# Wrap the conditioning models with additional image conditioning for inpainting model
+ # dummy_for_plms is needed because PLMS code checks the first item in the dict to have the right shape
if image_conditioning is not None:
- conditioning = {"c_concat": [image_conditioning], "c_crossattn": [conditioning]}
- unconditional_conditioning = {"c_concat": [image_conditioning], "c_crossattn": [unconditional_conditioning]}
+ conditioning = {"dummy_for_plms": np.zeros((conditioning.shape[0],)), "c_crossattn": [conditioning], "c_concat": [image_conditioning]}
+ unconditional_conditioning = {"c_crossattn": [unconditional_conditioning], "c_concat": [image_conditioning]}
samples_ddim = self.launch_sampling(steps, lambda: self.sampler.sample(S=steps, conditioning=conditioning, batch_size=int(x.shape[0]), shape=x[0].shape, verbose=False, unconditional_guidance_scale=p.cfg_scale, unconditional_conditioning=unconditional_conditioning, x_T=x, eta=self.eta)[0])
@@ -350,7 +350,9 @@ class TorchHijack:
class KDiffusionSampler:
def __init__(self, funcname, sd_model):
- self.model_wrap = k_diffusion.external.CompVisDenoiser(sd_model, quantize=shared.opts.enable_quantization)
+ denoiser = k_diffusion.external.CompVisVDenoiser if sd_model.parameterization == "v" else k_diffusion.external.CompVisDenoiser
+
+ self.model_wrap = denoiser(sd_model, quantize=shared.opts.enable_quantization)
self.funcname = funcname
self.func = getattr(k_diffusion.sampling, self.funcname)
self.extra_params = sampler_extra_params.get(funcname, [])
diff --git a/modules/shared.py b/modules/shared.py
index 3704ce23..5e8a9d27 100644
--- a/modules/shared.py
+++ b/modules/shared.py
@@ -11,17 +11,18 @@ import tqdm
import modules.artists
import modules.interrogate
import modules.memmon
-import modules.sd_models
import modules.styles
import modules.devices as devices
-from modules import sd_samplers, sd_models, localization, sd_vae, extensions, script_loading
-from modules.hypernetworks import hypernetwork
+from modules import localization, sd_vae, extensions, script_loading
from modules.paths import models_path, script_path, sd_path
+
+demo = None
+
sd_model_file = os.path.join(script_path, 'model.ckpt')
default_sd_model_file = sd_model_file
parser = argparse.ArgumentParser()
-parser.add_argument("--config", type=str, default=os.path.join(sd_path, "configs/stable-diffusion/v1-inference.yaml"), help="path to config which constructs model",)
+parser.add_argument("--config", type=str, default=os.path.join(script_path, "v1-inference.yaml"), help="path to config which constructs model",)
parser.add_argument("--ckpt", type=str, default=sd_model_file, help="path to checkpoint of stable diffusion model; if specified, this checkpoint will be added to the list of checkpoints and loaded",)
parser.add_argument("--ckpt-dir", type=str, default=None, help="Path to directory with stable diffusion checkpoints")
parser.add_argument("--gfpgan-dir", type=str, help="GFPGAN directory", default=('./src/gfpgan' if os.path.exists('./src/gfpgan') else './GFPGAN'))
@@ -55,7 +56,7 @@ parser.add_argument("--ldsr-models-path", type=str, help="Path to directory with
parser.add_argument("--clip-models-path", type=str, help="Path to directory with CLIP model file(s).", default=None)
parser.add_argument("--xformers", action='store_true', help="enable xformers for cross attention layers")
parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work")
-parser.add_argument("--deepdanbooru", action='store_true', help="enable deepdanbooru interrogator")
+parser.add_argument("--deepdanbooru", action='store_true', help="does not do anything")
parser.add_argument("--opt-split-attention", action='store_true', help="force-enables Doggettx's cross-attention layer optimization. By default, it's on for torch cuda.")
parser.add_argument("--opt-split-attention-invokeai", action='store_true', help="force-enables InvokeAI's cross-attention layer optimization. By default, it's on when cuda is unavailable.")
parser.add_argument("--opt-split-attention-v1", action='store_true', help="enable older version of split attention optimization that does not consume all the VRAM it can find")
@@ -121,10 +122,12 @@ xformers_available = False
config_filename = cmd_opts.ui_settings_file
os.makedirs(cmd_opts.hypernetwork_dir, exist_ok=True)
-hypernetworks = hypernetwork.list_hypernetworks(cmd_opts.hypernetwork_dir)
+hypernetworks = {}
loaded_hypernetwork = None
+
def reload_hypernetworks():
+ from modules.hypernetworks import hypernetwork
global hypernetworks
hypernetworks = hypernetwork.list_hypernetworks(cmd_opts.hypernetwork_dir)
@@ -206,10 +209,11 @@ class State:
if self.current_latent is None:
return
+ import modules.sd_samplers
if opts.show_progress_grid:
- self.current_image = sd_samplers.samples_to_image_grid(self.current_latent)
+ self.current_image = modules.sd_samplers.samples_to_image_grid(self.current_latent)
else:
- self.current_image = sd_samplers.sample_to_image(self.current_latent)
+ self.current_image = modules.sd_samplers.sample_to_image(self.current_latent)
self.current_image_sampling_step = self.sampling_step
@@ -248,6 +252,21 @@ def options_section(section_identifier, options_dict):
return options_dict
+def list_checkpoint_tiles():
+ import modules.sd_models
+ return modules.sd_models.checkpoint_tiles()
+
+
+def refresh_checkpoints():
+ import modules.sd_models
+ return modules.sd_models.list_models()
+
+
+def list_samplers():
+ import modules.sd_samplers
+ return modules.sd_samplers.all_samplers
+
+
hide_dirs = {"visible": not cmd_opts.hide_ui_dir_config}
options_templates = {}
@@ -276,6 +295,10 @@ options_templates.update(options_section(('saving-images', "Saving images/grids"
"use_original_name_batch": OptionInfo(False, "Use original name for output filename during batch process in extras tab"),
"save_selected_only": OptionInfo(True, "When using 'Save' button, only save a single selected image"),
"do_not_add_watermark": OptionInfo(False, "Do not add watermark to images"),
+
+ "temp_dir": OptionInfo("", "Directory for temporary images; leave empty for default"),
+ "clean_temp_dir_at_start": OptionInfo(False, "Cleanup non-default temporary directory when starting webui"),
+
}))
options_templates.update(options_section(('saving-paths', "Paths for saving"), {
@@ -332,7 +355,7 @@ options_templates.update(options_section(('training', "Training"), {
}))
options_templates.update(options_section(('sd', "Stable Diffusion"), {
- "sd_model_checkpoint": OptionInfo(None, "Stable Diffusion checkpoint", gr.Dropdown, lambda: {"choices": modules.sd_models.checkpoint_tiles()}, refresh=sd_models.list_models),
+ "sd_model_checkpoint": OptionInfo(None, "Stable Diffusion checkpoint", gr.Dropdown, lambda: {"choices": list_checkpoint_tiles()}, refresh=refresh_checkpoints),
"sd_checkpoint_cache": OptionInfo(0, "Checkpoints to cache in RAM", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}),
"sd_vae": OptionInfo("auto", "SD VAE", gr.Dropdown, lambda: {"choices": sd_vae.vae_list}, refresh=sd_vae.refresh_vae_list),
"sd_vae_as_default": OptionInfo(False, "Ignore selected VAE for stable diffusion checkpoints that have their own .vae.pt next to them"),
@@ -384,7 +407,7 @@ options_templates.update(options_section(('ui', "User interface"), {
}))
options_templates.update(options_section(('sampler-params', "Sampler parameters"), {
- "hide_samplers": OptionInfo([], "Hide samplers in user interface (requires restart)", gr.CheckboxGroup, lambda: {"choices": [x.name for x in sd_samplers.all_samplers]}),
+ "hide_samplers": OptionInfo([], "Hide samplers in user interface (requires restart)", gr.CheckboxGroup, lambda: {"choices": [x.name for x in list_samplers()]}),
"eta_ddim": OptionInfo(0.0, "eta (noise multiplier) for DDIM", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
"eta_ancestral": OptionInfo(1.0, "eta (noise multiplier) for ancestral samplers", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
"ddim_discretize": OptionInfo('uniform', "img2img DDIM discretize", gr.Radio, {"choices": ['uniform', 'quad']}),
diff --git a/modules/textual_inversion/preprocess.py b/modules/textual_inversion/preprocess.py
index 488aa5b5..56b9b2eb 100644
--- a/modules/textual_inversion/preprocess.py
+++ b/modules/textual_inversion/preprocess.py
@@ -6,12 +6,10 @@ import sys
import tqdm
import time
-from modules import shared, images
+from modules import shared, images, deepbooru
from modules.paths import models_path
from modules.shared import opts, cmd_opts
from modules.textual_inversion import autocrop
-if cmd_opts.deepdanbooru:
- import modules.deepbooru as deepbooru
def preprocess(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2, process_focal_crop=False, process_focal_crop_face_weight=0.9, process_focal_crop_entropy_weight=0.3, process_focal_crop_edges_weight=0.5, process_focal_crop_debug=False):
@@ -20,9 +18,7 @@ def preprocess(process_src, process_dst, process_width, process_height, preproce
shared.interrogator.load()
if process_caption_deepbooru:
- db_opts = deepbooru.create_deepbooru_opts()
- db_opts[deepbooru.OPT_INCLUDE_RANKS] = False
- deepbooru.create_deepbooru_process(opts.interrogate_deepbooru_score_threshold, db_opts)
+ deepbooru.model.start()
preprocess_work(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_flip, process_split, process_caption, process_caption_deepbooru, split_threshold, overlap_ratio, process_focal_crop, process_focal_crop_face_weight, process_focal_crop_entropy_weight, process_focal_crop_edges_weight, process_focal_crop_debug)
@@ -32,7 +28,7 @@ def preprocess(process_src, process_dst, process_width, process_height, preproce
shared.interrogator.send_blip_to_ram()
if process_caption_deepbooru:
- deepbooru.release_process()
+ deepbooru.model.stop()
def listfiles(dirname):
@@ -58,7 +54,7 @@ def save_pic_with_caption(image, index, params: PreprocessParams, existing_capti
if params.process_caption_deepbooru:
if len(caption) > 0:
caption += ", "
- caption += deepbooru.get_tags_from_process(image)
+ caption += deepbooru.model.tag_multi(image)
filename_part = params.src
filename_part = os.path.splitext(filename_part)[0]
diff --git a/modules/textual_inversion/textual_inversion.py b/modules/textual_inversion/textual_inversion.py
index b9b1394f..4eb75cb5 100644
--- a/modules/textual_inversion/textual_inversion.py
+++ b/modules/textual_inversion/textual_inversion.py
@@ -64,7 +64,8 @@ class EmbeddingDatabase:
self.word_embeddings[embedding.name] = embedding
- ids = model.cond_stage_model.tokenizer([embedding.name], add_special_tokens=False)['input_ids'][0]
+ # TODO changing between clip and open clip changes tokenization, which will cause embeddings to stop working
+ ids = model.cond_stage_model.tokenize([embedding.name])[0]
first_id = ids[0]
if first_id not in self.ids_lookup:
@@ -155,13 +156,11 @@ class EmbeddingDatabase:
def create_embedding(name, num_vectors_per_token, overwrite_old, init_text='*'):
cond_model = shared.sd_model.cond_stage_model
- embedding_layer = cond_model.wrapped.transformer.text_model.embeddings
with devices.autocast():
cond_model([""]) # will send cond model to GPU if lowvram/medvram is active
- ids = cond_model.tokenizer(init_text, max_length=num_vectors_per_token, return_tensors="pt", add_special_tokens=False)["input_ids"]
- embedded = embedding_layer.token_embedding.wrapped(ids.to(devices.device)).squeeze(0)
+ embedded = cond_model.encode_embedding_init_text(init_text, num_vectors_per_token)
vec = torch.zeros((num_vectors_per_token, embedded.shape[1]), device=devices.device)
for i in range(num_vectors_per_token):
diff --git a/modules/ui.py b/modules/ui.py
index 9d2a1cbf..20e248a0 100644
--- a/modules/ui.py
+++ b/modules/ui.py
@@ -19,14 +19,11 @@ import numpy as np
from PIL import Image, PngImagePlugin
-from modules import sd_hijack, sd_models, localization, script_callbacks, ui_extensions
+from modules import sd_hijack, sd_models, localization, script_callbacks, ui_extensions, deepbooru
from modules.paths import script_path
from modules.shared import opts, cmd_opts, restricted_opts
-if cmd_opts.deepdanbooru:
- from modules.deepbooru import get_deepbooru_tags
-
import modules.codeformer_model
import modules.generation_parameters_copypaste as parameters_copypaste
import modules.gfpgan_model
@@ -160,22 +157,6 @@ def save_files(js_data, images, do_make_zip, index):
return gr.File.update(value=fullfns, visible=True), '', '', plaintext_to_html(f"Saved: {filenames[0]}")
-def save_pil_to_file(pil_image, dir=None):
- use_metadata = False
- metadata = PngImagePlugin.PngInfo()
- for key, value in pil_image.info.items():
- if isinstance(key, str) and isinstance(value, str):
- metadata.add_text(key, value)
- use_metadata = True
-
- file_obj = tempfile.NamedTemporaryFile(delete=False, suffix=".png", dir=dir)
- pil_image.save(file_obj, pnginfo=(metadata if use_metadata else None))
- return file_obj
-
-
-# override save to file function so that it also writes PNG info
-gr.processing_utils.save_pil_to_file = save_pil_to_file
-
def wrap_gradio_call(func, extra_outputs=None, add_stats=False):
def f(*args, extra_outputs_array=extra_outputs, **kwargs):
@@ -352,7 +333,7 @@ def interrogate(image):
def interrogate_deepbooru(image):
- prompt = get_deepbooru_tags(image)
+ prompt = deepbooru.model.tag(image)
return gr_show(True) if prompt is None else prompt
@@ -481,9 +462,7 @@ def create_toprow(is_img2img):
if is_img2img:
with gr.Column(scale=1, elem_id="interrogate_col"):
button_interrogate = gr.Button('Interrogate\nCLIP', elem_id="interrogate")
-
- if cmd_opts.deepdanbooru:
- button_deepbooru = gr.Button('Interrogate\nDeepBooru', elem_id="deepbooru")
+ button_deepbooru = gr.Button('Interrogate\nDeepBooru', elem_id="deepbooru")
with gr.Column(scale=1):
with gr.Row():
@@ -1007,11 +986,10 @@ def create_ui(wrap_gradio_gpu_call):
outputs=[img2img_prompt],
)
- if cmd_opts.deepdanbooru:
- img2img_deepbooru.click(
- fn=interrogate_deepbooru,
- inputs=[init_img],
- outputs=[img2img_prompt],
+ img2img_deepbooru.click(
+ fn=interrogate_deepbooru,
+ inputs=[init_img],
+ outputs=[img2img_prompt],
)
@@ -1216,7 +1194,7 @@ def create_ui(wrap_gradio_gpu_call):
with gr.Tab(label="Create hypernetwork"):
new_hypernetwork_name = gr.Textbox(label="Name")
- new_hypernetwork_sizes = gr.CheckboxGroup(label="Modules", value=["768", "320", "640", "1280"], choices=["768", "320", "640", "1280"])
+ new_hypernetwork_sizes = gr.CheckboxGroup(label="Modules", value=["768", "320", "640", "1280"], choices=["768", "1024", "320", "640", "1280"])
new_hypernetwork_layer_structure = gr.Textbox("1, 2, 1", label="Enter hypernetwork layer structure", placeholder="1st and last digit must be 1. ex:'1, 2, 1'")
new_hypernetwork_activation_func = gr.Dropdown(value="linear", label="Select activation function of hypernetwork. Recommended : Swish / Linear(none)", choices=modules.hypernetworks.ui.keys)
new_hypernetwork_initialization_option = gr.Dropdown(value = "Normal", label="Select Layer weights initialization. Recommended: Kaiming for relu-like, Xavier for sigmoid-like, Normal otherwise", choices=["Normal", "KaimingUniform", "KaimingNormal", "XavierUniform", "XavierNormal"])
diff --git a/modules/ui_tempdir.py b/modules/ui_tempdir.py
new file mode 100644
index 00000000..9c6d3a9d
--- /dev/null
+++ b/modules/ui_tempdir.py
@@ -0,0 +1,62 @@
+import os
+import tempfile
+from collections import namedtuple
+
+import gradio as gr
+
+from PIL import PngImagePlugin
+
+from modules import shared
+
+
+Savedfile = namedtuple("Savedfile", ["name"])
+
+
+def save_pil_to_file(pil_image, dir=None):
+ already_saved_as = getattr(pil_image, 'already_saved_as', None)
+ if already_saved_as:
+ shared.demo.temp_dirs = shared.demo.temp_dirs | {os.path.abspath(os.path.dirname(already_saved_as))}
+ file_obj = Savedfile(already_saved_as)
+ return file_obj
+
+ if shared.opts.temp_dir != "":
+ dir = shared.opts.temp_dir
+
+ use_metadata = False
+ metadata = PngImagePlugin.PngInfo()
+ for key, value in pil_image.info.items():
+ if isinstance(key, str) and isinstance(value, str):
+ metadata.add_text(key, value)
+ use_metadata = True
+
+ file_obj = tempfile.NamedTemporaryFile(delete=False, suffix=".png", dir=dir)
+ pil_image.save(file_obj, pnginfo=(metadata if use_metadata else None))
+ return file_obj
+
+
+# override save to file function so that it also writes PNG info
+gr.processing_utils.save_pil_to_file = save_pil_to_file
+
+
+def on_tmpdir_changed():
+ if shared.opts.temp_dir == "" or shared.demo is None:
+ return
+
+ os.makedirs(shared.opts.temp_dir, exist_ok=True)
+
+ shared.demo.temp_dirs = shared.demo.temp_dirs | {os.path.abspath(shared.opts.temp_dir)}
+
+
+def cleanup_tmpdr():
+ temp_dir = shared.opts.temp_dir
+ if temp_dir == "" or not os.path.isdir(temp_dir):
+ return
+
+ for root, dirs, files in os.walk(temp_dir, topdown=False):
+ for name in files:
+ _, extension = os.path.splitext(name)
+ if extension != ".png":
+ continue
+
+ filename = os.path.join(root, name)
+ os.remove(filename)
diff --git a/requirements.txt b/requirements.txt
index 762db4f3..e4e5ec64 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -28,3 +28,4 @@ kornia
lark
inflection
GitPython
+torchsde
diff --git a/requirements_versions.txt b/requirements_versions.txt
index 662ca684..8d557fe3 100644
--- a/requirements_versions.txt
+++ b/requirements_versions.txt
@@ -25,3 +25,4 @@ kornia==0.6.7
lark==1.1.2
inflection==0.5.1
GitPython==3.1.27
+torchsde==0.2.5
diff --git a/v1-inference.yaml b/v1-inference.yaml
new file mode 100644
index 00000000..d4effe56
--- /dev/null
+++ b/v1-inference.yaml
@@ -0,0 +1,70 @@
+model:
+ base_learning_rate: 1.0e-04
+ target: ldm.models.diffusion.ddpm.LatentDiffusion
+ params:
+ linear_start: 0.00085
+ linear_end: 0.0120
+ num_timesteps_cond: 1
+ log_every_t: 200
+ timesteps: 1000
+ first_stage_key: "jpg"
+ cond_stage_key: "txt"
+ image_size: 64
+ channels: 4
+ cond_stage_trainable: false # Note: different from the one we trained before
+ conditioning_key: crossattn
+ monitor: val/loss_simple_ema
+ scale_factor: 0.18215
+ use_ema: False
+
+ scheduler_config: # 10000 warmup steps
+ target: ldm.lr_scheduler.LambdaLinearScheduler
+ params:
+ warm_up_steps: [ 10000 ]
+ cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+ f_start: [ 1.e-6 ]
+ f_max: [ 1. ]
+ f_min: [ 1. ]
+
+ unet_config:
+ target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+ params:
+ image_size: 32 # unused
+ in_channels: 4
+ out_channels: 4
+ model_channels: 320
+ attention_resolutions: [ 4, 2, 1 ]
+ num_res_blocks: 2
+ channel_mult: [ 1, 2, 4, 4 ]
+ num_heads: 8
+ use_spatial_transformer: True
+ transformer_depth: 1
+ context_dim: 768
+ use_checkpoint: True
+ legacy: False
+
+ first_stage_config:
+ target: ldm.models.autoencoder.AutoencoderKL
+ params:
+ embed_dim: 4
+ monitor: val/rec_loss
+ ddconfig:
+ double_z: true
+ z_channels: 4
+ resolution: 256
+ in_channels: 3
+ out_ch: 3
+ ch: 128
+ ch_mult:
+ - 1
+ - 2
+ - 4
+ - 4
+ num_res_blocks: 2
+ attn_resolutions: []
+ dropout: 0.0
+ lossconfig:
+ target: torch.nn.Identity
+
+ cond_stage_config:
+ target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
diff --git a/webui.py b/webui.py
index c5e5fe75..6b79dc55 100644
--- a/webui.py
+++ b/webui.py
@@ -10,7 +10,7 @@ from fastapi.middleware.gzip import GZipMiddleware
from modules.paths import script_path
-from modules import devices, sd_samplers, upscaler, extensions, localization
+from modules import shared, devices, sd_samplers, upscaler, extensions, localization, ui_tempdir
import modules.codeformer_model as codeformer
import modules.extras
import modules.face_restoration
@@ -23,7 +23,6 @@ import modules.scripts
import modules.sd_hijack
import modules.sd_models
import modules.sd_vae
-import modules.shared as shared
import modules.txt2img
import modules.script_callbacks
@@ -32,12 +31,14 @@ from modules import modelloader
from modules.shared import cmd_opts
import modules.hypernetworks.hypernetwork
+
queue_lock = threading.Lock()
if cmd_opts.server_name:
server_name = cmd_opts.server_name
else:
server_name = "0.0.0.0" if cmd_opts.listen else None
+
def wrap_queued_call(func):
def f(*args, **kwargs):
with queue_lock:
@@ -86,8 +87,9 @@ def initialize():
shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: modules.sd_models.reload_model_weights()))
shared.opts.onchange("sd_vae", wrap_queued_call(lambda: modules.sd_vae.reload_vae_weights()), call=False)
shared.opts.onchange("sd_vae_as_default", wrap_queued_call(lambda: modules.sd_vae.reload_vae_weights()), call=False)
- shared.opts.onchange("sd_hypernetwork", wrap_queued_call(lambda: modules.hypernetworks.hypernetwork.load_hypernetwork(shared.opts.sd_hypernetwork)))
+ shared.opts.onchange("sd_hypernetwork", wrap_queued_call(lambda: shared.reload_hypernetworks()))
shared.opts.onchange("sd_hypernetwork_strength", modules.hypernetworks.hypernetwork.apply_strength)
+ shared.opts.onchange("temp_dir", ui_tempdir.on_tmpdir_changed)
if cmd_opts.tls_keyfile is not None and cmd_opts.tls_keyfile is not None:
@@ -150,9 +152,12 @@ def webui():
initialize()
while 1:
- demo = modules.ui.create_ui(wrap_gradio_gpu_call=wrap_gradio_gpu_call)
+ if shared.opts.clean_temp_dir_at_start:
+ ui_tempdir.cleanup_tmpdr()
+
+ shared.demo = modules.ui.create_ui(wrap_gradio_gpu_call=wrap_gradio_gpu_call)
- app, local_url, share_url = demo.launch(
+ app, local_url, share_url = shared.demo.launch(
share=cmd_opts.share,
server_name=server_name,
server_port=cmd_opts.port,
@@ -179,9 +184,9 @@ def webui():
if launch_api:
create_api(app)
- modules.script_callbacks.app_started_callback(demo, app)
+ modules.script_callbacks.app_started_callback(shared.demo, app)
- wait_on_server(demo)
+ wait_on_server(shared.demo)
sd_samplers.set_samplers()
generated by cgit v1.2.1 (git 2.18.0) at 2025-05-21 05:12:19 +0000