mirror of
https://github.com/storytold/storyteller-ml.git
synced 2026-10-19 07:54:56 +00:00
remove old comfyUI
This commit is contained in:
salt
@@ -1,165 +0,0 @@
|
||||
# Byte-compiled / optimized / DLL files
|
||||
__pycache__/
|
||||
*.py[cod]
|
||||
*$py.class
|
||||
|
||||
# C extensions
|
||||
*.so
|
||||
|
||||
# Distribution / packaging
|
||||
.Python
|
||||
build/
|
||||
develop-eggs/
|
||||
dist/
|
||||
downloads/
|
||||
eggs/
|
||||
.eggs/
|
||||
lib/
|
||||
lib64/
|
||||
parts/
|
||||
sdist/
|
||||
var/
|
||||
wheels/
|
||||
share/python-wheels/
|
||||
*.egg-info/
|
||||
.installed.cfg
|
||||
*.egg
|
||||
MANIFEST
|
||||
|
||||
# PyInstaller
|
||||
# Usually these files are written by a python script from a template
|
||||
# before PyInstaller builds the exe, so as to inject date/other infos into it.
|
||||
*.manifest
|
||||
*.spec
|
||||
|
||||
# Installer logs
|
||||
pip-log.txt
|
||||
pip-delete-this-directory.txt
|
||||
|
||||
# Unit test / coverage reports
|
||||
htmlcov/
|
||||
.tox/
|
||||
.nox/
|
||||
.coverage
|
||||
.coverage.*
|
||||
.cache
|
||||
nosetests.xml
|
||||
coverage.xml
|
||||
*.cover
|
||||
*.py,cover
|
||||
.hypothesis/
|
||||
.pytest_cache/
|
||||
cover/
|
||||
|
||||
# Translations
|
||||
*.mo
|
||||
*.pot
|
||||
|
||||
# Django stuff:
|
||||
*.log
|
||||
local_settings.py
|
||||
db.sqlite3
|
||||
db.sqlite3-journal
|
||||
|
||||
# Flask stuff:
|
||||
instance/
|
||||
.webassets-cache
|
||||
|
||||
# Scrapy stuff:
|
||||
.scrapy
|
||||
|
||||
# Sphinx documentation
|
||||
docs/_build/
|
||||
|
||||
# PyBuilder
|
||||
.pybuilder/
|
||||
target/
|
||||
|
||||
# Jupyter Notebook
|
||||
.ipynb_checkpoints
|
||||
|
||||
# IPython
|
||||
profile_default/
|
||||
ipython_config.py
|
||||
|
||||
# pyenv
|
||||
# For a library or package, you might want to ignore these files since the code is
|
||||
# intended to run in multiple environments; otherwise, check them in:
|
||||
# .python-version
|
||||
|
||||
# pipenv
|
||||
# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
|
||||
# However, in case of collaboration, if having platform-specific dependencies or dependencies
|
||||
# having no cross-platform support, pipenv may install dependencies that don't work, or not
|
||||
# install all needed dependencies.
|
||||
#Pipfile.lock
|
||||
|
||||
# poetry
|
||||
# Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
|
||||
# This is especially recommended for binary packages to ensure reproducibility, and is more
|
||||
# commonly ignored for libraries.
|
||||
# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
|
||||
#poetry.lock
|
||||
|
||||
# PEP 582; used by e.g. github.com/David-OConnor/pyflow
|
||||
__pypackages__/
|
||||
|
||||
# Celery stuff
|
||||
celerybeat-schedule
|
||||
celerybeat.pid
|
||||
|
||||
# SageMath parsed files
|
||||
*.sage.py
|
||||
|
||||
# Environments
|
||||
.env
|
||||
.venv
|
||||
env/
|
||||
venv/
|
||||
ENV/
|
||||
env.bak/
|
||||
venv.bak/
|
||||
|
||||
# Spyder project settings
|
||||
.spyderproject
|
||||
.spyproject
|
||||
|
||||
# Rope project settings
|
||||
.ropeproject
|
||||
|
||||
# mkdocs documentation
|
||||
/site
|
||||
|
||||
# mypy
|
||||
.mypy_cache/
|
||||
.dmypy.json
|
||||
dmypy.json
|
||||
|
||||
# Pyre type checker
|
||||
.pyre/
|
||||
|
||||
# pytype static type analyzer
|
||||
.pytype/
|
||||
|
||||
# Cython debug symbols
|
||||
cython_debug/
|
||||
|
||||
# PyCharm
|
||||
# JetBrains specific template is maintainted in a separate JetBrains.gitignore that can
|
||||
# be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
|
||||
# and can be added to the global gitignore or merged into this file. For a more nuclear
|
||||
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
|
||||
#.idea/
|
||||
*.pyc
|
||||
|
||||
# Custom
|
||||
was_suite_settings.json
|
||||
styles.json
|
||||
was_suite_config.json
|
||||
workflows/
|
||||
was_history.json
|
||||
nsp_pantry.json
|
||||
cache/
|
||||
*.latent
|
||||
*.image
|
||||
*.conditioning
|
||||
-429
@@ -1,429 +0,0 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {
|
||||
"id": "view-in-github",
|
||||
"colab_type": "text"
|
||||
},
|
||||
"source": [
|
||||
"<a href=\"https://colab.research.google.com/github/WASasquatch/was-node-suite-comfyui/blob/main/ComfyUI_%2B_WAS_Node_Suite_and_ComfyUI_Manager.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {
|
||||
"id": "aaaaaaaaaa"
|
||||
},
|
||||
"source": [
|
||||
"# <font color=\"#0497e0\">**Comfy**</font>UI + <font color=\"#4684a3\">**WAS** Node Suite</font> [](https://hits.seeyoufarm.com)\n",
|
||||
"A version of ComfyUI Colab with WAS Node Suite installatoin."
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {
|
||||
"id": "bbbbbbbbbb",
|
||||
"cellView": "form"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"#@title <font color=\"orange\" size=\"5\">Environment Setup</font>\n",
|
||||
"#@markdown <font color=\"#0497e0\">Download and install ComfyUI + WAS Node Suite. You can run this cell again with the `UPDATE_COMFY_UI` or `UPDATE_WAS_NS` options selected to update.\n",
|
||||
"\n",
|
||||
"from pathlib import Path\n",
|
||||
"import time\n",
|
||||
"\n",
|
||||
"OPTIONS = {}\n",
|
||||
"\n",
|
||||
"#@markdown <font color=\"#48768c\">Store ComfyUI on Google Drive instead of Colab</font>\n",
|
||||
"USE_GOOGLE_DRIVE = True #@param {type:\"boolean\"}\n",
|
||||
"#markdown <font color=\"#48768c\">Update ComfyUI</font>\n",
|
||||
"UPDATE_COMFY_UI = True #@param {type:\"boolean\"}\n",
|
||||
"#@markdown <font color=\"#48768c\">Update WAS Node Suite</font>\n",
|
||||
"UPDATE_WAS_NS = True #@param {type:\"boolean\"}\n",
|
||||
"#@markdown <font color=\"#48768c\">Update Pillow for WAS NS:</font>\n",
|
||||
"UPDATE_PILLOW = False #@param {type:\"boolean\"}\n",
|
||||
"#@markdown <font color=\"#48768c\">ComfyUI Manager:</font>\n",
|
||||
"USE_COMFYUI_MANAGER = True #@param {type:\"boolean\"}\n",
|
||||
"UPDATE_COMFYUI_MANAGER = True #@param {type:\"boolean\"}\n",
|
||||
"\n",
|
||||
"WORKSPACE = '/content/ComfyUI'\n",
|
||||
"OPTIONS['USE_GOOGLE_DRIVE'] = USE_GOOGLE_DRIVE\n",
|
||||
"OPTIONS['UPDATE_COMFY_UI'] = UPDATE_COMFY_UI\n",
|
||||
"\n",
|
||||
"if USE_GOOGLE_DRIVE:\n",
|
||||
" !echo \"Mounting Google Drive...\"\n",
|
||||
" %cd /\n",
|
||||
" from google.colab import drive\n",
|
||||
" drive.mount('/content/drive')\n",
|
||||
" WORKSPACE = \"/content/drive/MyDrive/ComfyUI\"\n",
|
||||
" %cd /content/drive/MyDrive\n",
|
||||
"\n",
|
||||
"![ ! -d $WORKSPACE ] && echo -= Initial setup ComfyUI =- && git clone https://github.com/comfyanonymous/ComfyUI $WORKSPACE\n",
|
||||
"%cd $WORKSPACE\n",
|
||||
"\n",
|
||||
"if UPDATE_COMFY_UI:\n",
|
||||
" !echo -= Updating ComfyUI =-\n",
|
||||
" !git pull\n",
|
||||
"\n",
|
||||
"!echo -= Install dependencies =-\n",
|
||||
"!pip install xformers -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu118\n",
|
||||
"!git clone https://github.com/WASasquatch/was-node-suite-comfyui $WORKSPACE/custom_nodes/was-node-suite-comfyui\n",
|
||||
"\n",
|
||||
"if USE_COMFYUI_MANAGER:\n",
|
||||
" !git clone https://github.com/ltdrdata/ComfyUI-Manager.git $WORKSPACE/custom_nodes/ComfyUI-Manager\n",
|
||||
"\n",
|
||||
"if UPDATE_WAS_NS:\n",
|
||||
" %cd $WORKSPACE/custom_nodes/was-node-suite-comfyui\n",
|
||||
" !git pull\n",
|
||||
" %cd $WORKSPACE\n",
|
||||
"\n",
|
||||
"if UPDATE_COMFYUI_MANAGER:\n",
|
||||
" %cd $WORKSPACE/custom_nodes/ComfyUI-Manager\n",
|
||||
" !git pull\n",
|
||||
" %cd $WORKSPACE\n",
|
||||
"\n",
|
||||
"if UPDATE_PILLOW:\n",
|
||||
" !pip install --upgrade --force-reinstall pillow\n",
|
||||
" print('\\n\\033[91m\\033[1mRestarting runtime for Pillow Update. Run this cell again without `UPDATE_PILLOW` selected!\\033[0m')\n",
|
||||
" time.sleep(5)\n",
|
||||
"\n",
|
||||
" import os\n",
|
||||
" os.kill(os.getpid(), 9)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {
|
||||
"id": "cccccccccc"
|
||||
},
|
||||
"source": [
|
||||
"Download some models/checkpoints/vae or custom comfyui nodes (uncomment the commands for the ones you want)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {
|
||||
"id": "dddddddddd",
|
||||
"cellView": "form"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"#@title <font color=\"orange\" size=\"5\">Download Models</font>\n",
|
||||
"#@markdown <font color=\"#0497e0\">Download models and other resources to use in ComfyUI. Select your options and run the this cell. Can run multiple times with different options.</font>\n",
|
||||
"\n",
|
||||
"#@markdown ---\n",
|
||||
"\n",
|
||||
"# Checkpoints\n",
|
||||
"\n",
|
||||
"MODEL_OPTION = 'stable-diffusion-xl-base-1.0.safetensors' #@param['None', 'stable-diffusion-xl-base-1.0.safetensors', 'sd_xl_base_1.0_0.9vae.safetensors', 'sd_xl_base_0.9.safetensors', 'v1-5-pruned-emaonly.ckpt', 'v2-1_512-ema-pruned.safetensors', 'v2-1_768-ema-pruned.safetensors', 'AbyssOrangeMix2_hard.safetensors', 'AOM3A1_orangemixs.safetensors', 'AOM3A3_orangemixs.safetensors', 'wd-1-5-beta2-fp16.safetensors']\n",
|
||||
"VAE_OPTION = 'vae-ft-mse-840000-ema-pruned.safetensors' #@param['None', 'vae-ft-mse-840000-ema-pruned.safetensors', 'orangemix.vae.pt', 'kl-f8-anime2.ckpt']\n",
|
||||
"UPSCALE_MODEL_OPTION = 'None' #@param['None', 'RealESRGAN_x2.pth', 'RealESRGAN_x4.pth', '4x-UltraSharp', '4x_RealisticRescaler_100000_G.pth', 'BSRGAN.pth', 'BSRGANx2.pth']\n",
|
||||
"LORA_OPTION = 'None' #@param['None', 'theovercomer8sContrastFix_sd21768.safetensors', 'theovercomer8sContrastFix_sd15.safetensors']\n",
|
||||
"T2I_OPTION = 'None' #@param['None', 't2iadapter_depth_sd14v1.pth', 't2iadapter_seg_sd14v1.pth', 't2iadapter_sketch_sd14v1.pth', 't2iadapter_keypose_sd14v1.pth', 't2iadapter_openpose_sd14v1.pth', 't2iadapter_color_sd14v1.pth', 't2iadapter_canny_sd14v1.pth', '/t2iadapter_style_sd14v1.pth']\n",
|
||||
"CONTROLNET_OPTION = 'None' #@param['None', 'control_depth-fp16.safetensors', 'control_scribble-fp16.safetensors', 'control_openpose-fp16.safetensors']\n",
|
||||
"\n",
|
||||
"#@markdown ---\n",
|
||||
"\n",
|
||||
"#@markdown <font color=\"#0497e0\">**Download and instlal CLIPVision**:</font>\n",
|
||||
"DOWNLOAD_CLIPVISION = False #@param {type:\"boolean\"}\n",
|
||||
"#@markdown <font color=\"#0497e0\">**ControlNet Preprocessor Nodes** by Fannovel16:</font>\n",
|
||||
"INSTALL_CONTROLNET_NODES = False #@param {type:\"boolean\"}\n",
|
||||
"\n",
|
||||
"# SDXL\n",
|
||||
"if MODEL_OPTION == 'stable-diffusion-xl-base-1.0.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"if MODEL_OPTION == 'sd_xl_base_1.0_0.9vae.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0_0.9vae.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"if MODEL_OPTION == 'sd_xl_base_0.9.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9/resolve/main/sd_xl_base_0.9.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"# SD1.5\n",
|
||||
"if MODEL_OPTION == 'v1-5-pruned-emaonly.ckpt':\n",
|
||||
" !wget -c https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.ckpt -P $WORKSPACE/models/checkpoints/\n",
|
||||
"\n",
|
||||
"# SD2\n",
|
||||
"if MODEL_OPTION == 'v2-1_512-ema-pruned.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/stabilityai/stable-diffusion-2-1-base/resolve/main/v2-1_512-ema-pruned.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"if MODEL_OPTION == 'v2-1_768-ema-pruned.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/stabilityai/stable-diffusion-2-1/resolve/main/v2-1_768-ema-pruned.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"\n",
|
||||
"# Some SD1.5 anime style\n",
|
||||
"if MODEL_OPTION == 'AbyssOrangeMix2_hard.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/WarriorMama777/OrangeMixs/resolve/main/Models/AbyssOrangeMix2/AbyssOrangeMix2_hard.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"if MODEL_OPTION == 'AOM3A1_orangemixs.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/WarriorMama777/OrangeMixs/resolve/main/Models/AbyssOrangeMix3/AOM3A1_orangemixs.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"if MODEL_OPTION == 'AOM3A3_orangemixs.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/WarriorMama777/OrangeMixs/resolve/main/Models/AbyssOrangeMix3/AOM3A3_orangemixs.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"if MODEL_OPTION == 'anything-v3-fp16-pruned.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/Linaqruf/anything-v3.0/resolve/main/anything-v3-fp16-pruned.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"\n",
|
||||
"# Waifu Diffusion 1.5 (anime style SD2.x 768-v)\n",
|
||||
"if MODEL_OPTION == 'wd-1-5-beta2-fp16.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/waifu-diffusion/wd-1-5-beta2/resolve/main/checkpoints/wd-1-5-beta2-fp16.safetensors -P $WORKSPACE/models/checkpoints/\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"# VAE\n",
|
||||
"if VAE_OPTION == 'vae-ft-mse-840000-ema-pruned.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.safetensors -P $WORKSPACE/models/vae/\n",
|
||||
"if VAE_OPTION == 'orangemix.vae.pt':\n",
|
||||
" !wget -c https://huggingface.co/WarriorMama777/OrangeMixs/resolve/main/VAEs/orangemix.vae.pt -P $WORKSPACE/models/vae/'\n",
|
||||
"if VAE_OPTION == 'kl-f8-anime2.ckpt':\n",
|
||||
" !wget -c https://huggingface.co/hakurei/waifu-diffusion-v1-4/resolve/main/vae/kl-f8-anime2.ckpt -P $WORKSPACE/models/vae/\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"# Loras\n",
|
||||
"if LORA_OPTION == 'theovercomer8sContrastFix_sd21768.safetensors':\n",
|
||||
" !wget -c https://civitai.com/api/download/models/10350 -O $WORKSPACE/models/loras/theovercomer8sContrastFix_sd21768.safetensors #theovercomer8sContrastFix SD2.x 768-v\n",
|
||||
"if LORA_OPTION == 'theovercomer8sContrastFix_sd15.safetensors':\n",
|
||||
" !wget -c https://civitai.com/api/download/models/10638 -O $WORKSPACE/models/loras/theovercomer8sContrastFix_sd15.safetensors #theovercomer8sContrastFix SD1.x\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"# T2I-Adapter\n",
|
||||
"if T2I_OPTION == 't2iadapter_depth_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_depth_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"if T2I_OPTION == 't2iadapter_seg_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_seg_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"if T2I_OPTION == 't2iadapter_sketch_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_sketch_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"if T2I_OPTION == 't2iadapter_keypose_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_keypose_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"if T2I_OPTION == 't2iadapter_openpose_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_openpose_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"if T2I_OPTION == 't2iadapter_color_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_color_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"if T2I_OPTION == 't2iadapter_canny_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_canny_sd14v1.pth -P $WORKSPACE/models/controlnet/\n",
|
||||
"\n",
|
||||
"# T2I Styles Model\n",
|
||||
"if T2I_OPTION == '/t2iadapter_style_sd14v1.pth':\n",
|
||||
" !wget -c https://huggingface.co/TencentARC/T2I-Adapter/resolve/main/models/t2iadapter_style_sd14v1.pth -P $WORKSPACE/models/style_models/\n",
|
||||
"\n",
|
||||
"# CLIPVision model (needed for styles model)\n",
|
||||
"if DOWNLOAD_CLIPVISION:\n",
|
||||
" !wget -c https://huggingface.co/openai/clip-vit-large-patch14/resolve/main/pytorch_model.bin -O $WORKSPACE/models/clip_vision/clip_vit14.bin\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"# ControlNet\n",
|
||||
"if CONTROLNET_OPTION == 'control_depth-fp16.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/webui/ControlNet-modules-safetensors/resolve/main/control_depth-fp16.safetensors -P $WORKSPACE/models/controlnet/\n",
|
||||
"if CONTROLNET_OPTION == 'control_scribble-fp16.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/webui/ControlNet-modules-safetensors/resolve/main/control_scribble-fp16.safetensors -P $WORKSPACE/models/controlnet/\n",
|
||||
"if CONTROLNET_OPTION == 'control_openpose-fp16.safetensors':\n",
|
||||
" !wget -c https://huggingface.co/webui/ControlNet-modules-safetensors/resolve/main/control_openpose-fp16.safetensors -P $WORKSPACE/models/controlnet/\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"# Controlnet Preprocessor nodes by Fannovel16\n",
|
||||
"if INSTALL_CONTROLNET_NODES:\n",
|
||||
" !cd custom_nodes && git clone https://github.com/Fannovel16/comfy_controlnet_preprocessors; cd comfy_controlnet_preprocessors && python install.py\n",
|
||||
"\n",
|
||||
"# ESRGAN upscale model\n",
|
||||
"if UPSCALE_MODEL_OPTION == 'RealESRGAN_x2.pth':\n",
|
||||
" !wget -c https://huggingface.co/sberbank-ai/Real-ESRGAN/resolve/main/RealESRGAN_x2.pth -P $WORKSPACE/models/upscale_models/\n",
|
||||
"if UPSCALE_MODEL_OPTION == 'RealESRGAN_x4.pth':\n",
|
||||
" !wget -c https://huggingface.co/sberbank-ai/Real-ESRGAN/resolve/main/RealESRGAN_x4.pth -P $WORKSPACE/models/upscale_models/\n",
|
||||
"if UPSCALE_MODEL_OPTION == '4x-UltraSharp':\n",
|
||||
" !wget -c https://huggingface.co/uwg/upscaler/resolve/main/ESRGAN/4x-UltraSharp.pth -P $WORKSPACE/models/upscale_models/\n",
|
||||
"if UPSCALE_MODEL_OPTION == '4x_RealisticRescaler_100000_G.pth':\n",
|
||||
" !wget -c https://huggingface.co/uwg/upscaler/resolve/main/ESRGAN/4x_RealisticRescaler_100000_G.pth -P $WORKSPACE/models/upscale_models/\n",
|
||||
"if UPSCALE_MODEL_OPTION == 'BSRGAN.pth':\n",
|
||||
" !wget -c https://huggingface.co/uwg/upscaler/resolve/main/ESRGAN/BSRGAN.pth -P $WORKSPACE/models/upscale_models/\n",
|
||||
"if UPSCALE_MODEL_OPTION == 'BSRGANx2.pth':\n",
|
||||
" !wget -c https://huggingface.co/uwg/upscaler/resolve/main/ESRGAN/BSRGANx2.pth -P $WORKSPACE/models/upscale_models/\n",
|
||||
"\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"source": [
|
||||
"#@title <font size=\"5\" color=\"orange\">Direct Download Models</font>\n",
|
||||
"\n",
|
||||
"import os\n",
|
||||
"\n",
|
||||
"types = {\n",
|
||||
" 'CHECKPOINTS': os.path.join(WORKSPACE, 'models/checkpoints'),\n",
|
||||
" 'CLIP': os.path.join(WORKSPACE, 'models/clip'),\n",
|
||||
" 'CLIP_VISION': os.path.join(WORKSPACE, 'models/clip_vision'),\n",
|
||||
" 'CONFIGS': os.path.join(WORKSPACE, 'models/configs'),\n",
|
||||
" 'CONTROLNET': os.path.join(WORKSPACE, 'models/controlnet'),\n",
|
||||
" 'DIFFUSERS': os.path.join(WORKSPACE, 'models/diffusers'),\n",
|
||||
" 'EMBEDDINGS': os.path.join(WORKSPACE, 'models/embeddings'),\n",
|
||||
" 'GLIGEN': os.path.join(WORKSPACE, 'models/gligen'),\n",
|
||||
" 'HYPERNETWORKS': os.path.join(WORKSPACE, 'models/hypernetworks'),\n",
|
||||
" 'LORAS': os.path.join(WORKSPACE, 'models/loras'),\n",
|
||||
" 'STYLE_MODEL': os.path.join(WORKSPACE, 'models/style_models'),\n",
|
||||
" 'UNET': os.path.join(WORKSPACE, 'models/unet'),\n",
|
||||
" 'UPSCALE_MODELS': os.path.join(WORKSPACE, 'models/upscale_models'),\n",
|
||||
" 'VAE': os.path.join(WORKSPACE, 'models/vae'),\n",
|
||||
" 'VAE_APPROX': os.path.join(WORKSPACE, 'models/vae_approx')\n",
|
||||
"}\n",
|
||||
"\n",
|
||||
"TYPE = 'CHECKPOINTS' #@param ['CHECKPOINTS', 'CLIP', 'CLIP_VISION', 'CONFIGS', 'CONTROLNET', 'DIFFUSERS', 'EMBEDDINGS', 'GLIGEN', 'HYPERNETWORKS', 'LORAS', 'STYLE_MODEL', 'UNET', 'UPSCALE_MODELS', 'VAE', 'VAE_APPROX']\n",
|
||||
"DIRECT_URL = 'https://civitai.com/api/download/models/141627' #@param {type:\"string\"}\n",
|
||||
"SAVE_AS = 'Differentia_V1.safetensors' #@param {type: 'string'}\n",
|
||||
"#@markdown <font color=\"0497e0\">Direct link to the download. The example URL is Differentia (https://civitai.com/models/129232)</font>\n",
|
||||
"\n",
|
||||
"if TYPE and DIRECT_URL:\n",
|
||||
" target = os.path.join(types[TYPE], SAVE_AS)\n",
|
||||
" !wget -c $DIRECT_URL -O $target"
|
||||
],
|
||||
"metadata": {
|
||||
"cellView": "form",
|
||||
"id": "Bo0pf4So3tCK"
|
||||
},
|
||||
"execution_count": null,
|
||||
"outputs": []
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"source": [
|
||||
"#@title <font size=\"5\" color=\"orange\">ComfyUI Cloudfare (Recommended)</font>\n",
|
||||
"#@markdown <font color=\"0497e0\">Running ComfyUI with Cloudfare is now the recommended method.</font>\n",
|
||||
"!wget https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb\n",
|
||||
"!dpkg -i cloudflared-linux-amd64.deb\n",
|
||||
"\n",
|
||||
"import subprocess\n",
|
||||
"import threading\n",
|
||||
"import time\n",
|
||||
"import socket\n",
|
||||
"import urllib.request\n",
|
||||
"\n",
|
||||
"def iframe_thread(port):\n",
|
||||
" while True:\n",
|
||||
" time.sleep(0.5)\n",
|
||||
" sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n",
|
||||
" result = sock.connect_ex(('127.0.0.1', port))\n",
|
||||
" if result == 0:\n",
|
||||
" break\n",
|
||||
" sock.close()\n",
|
||||
" print(\"\\nComfyUI finished loading, trying to launch cloudflared (if it gets stuck here cloudflared is having issues)\\n\")\n",
|
||||
"\n",
|
||||
" p = subprocess.Popen([\"cloudflared\", \"tunnel\", \"--url\", \"http://127.0.0.1:{}\".format(port)], stdout=subprocess.PIPE, stderr=subprocess.PIPE)\n",
|
||||
" for line in p.stderr:\n",
|
||||
" l = line.decode()\n",
|
||||
" if \"trycloudflare.com \" in l:\n",
|
||||
" print(\"This is the URL to access ComfyUI:\", l[l.find(\"http\"):], end='')\n",
|
||||
" #print(l, end='')\n",
|
||||
"\n",
|
||||
"threading.Thread(target=iframe_thread, daemon=True, args=(8188,)).start()\n",
|
||||
"\n",
|
||||
"!python main.py --dont-print-server"
|
||||
],
|
||||
"metadata": {
|
||||
"cellView": "form",
|
||||
"id": "StSynv5tp2nL"
|
||||
},
|
||||
"execution_count": null,
|
||||
"outputs": []
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {
|
||||
"id": "jjjjjjjjjjjjj",
|
||||
"cellView": "form"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"#@title <font size=\"5\" color=\"orange\">ComfyUI Localtunnel</font>\n",
|
||||
"#@markdown <font color=\"0497e0\">Run this cell to start ComfyUI. You'll see a link similar to `your url is: https://slow-yaks-jog-34-72-173-3.loca.lt` (example)</font>\n",
|
||||
"#@markdown <br>*If you have trouble with the red screen of death \"reminder\" not letting you generate, use the iFrame version below.*\n",
|
||||
"!npm install -g localtunnel\n",
|
||||
"\n",
|
||||
"import subprocess\n",
|
||||
"import threading\n",
|
||||
"import time\n",
|
||||
"import socket\n",
|
||||
"import urllib.request\n",
|
||||
"\n",
|
||||
"def iframe_thread(port):\n",
|
||||
" while True:\n",
|
||||
" time.sleep(0.5)\n",
|
||||
" sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n",
|
||||
" result = sock.connect_ex(('127.0.0.1', port))\n",
|
||||
" if result == 0:\n",
|
||||
" break\n",
|
||||
" sock.close()\n",
|
||||
" print(\"\\nComfyUI finished loading, trying to launch localtunnel (if it gets stuck here localtunnel is having issues)\\n\")\n",
|
||||
"\n",
|
||||
" print(\"The password/enpoint ip for localtunnel is:\", urllib.request.urlopen('https://ipv4.icanhazip.com').read().decode('utf8').strip(\"\\n\"))\n",
|
||||
" p = subprocess.Popen([\"lt\", \"--port\", \"{}\".format(port)], stdout=subprocess.PIPE)\n",
|
||||
" for line in p.stdout:\n",
|
||||
" print(line.decode(), end='')\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"threading.Thread(target=iframe_thread, daemon=True, args=(8188,)).start()\n",
|
||||
"\n",
|
||||
"!python main.py --dont-print-server"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {
|
||||
"id": "gggggggggg"
|
||||
},
|
||||
"source": [
|
||||
"### Run ComfyUI with colab iframe ***(use only in case the previous ways don't work)***\n",
|
||||
"\n",
|
||||
"You should see the ui appear in an iframe. If you get a 403 error, it's your firefox settings or an extension that's messing things up.\n",
|
||||
"\n",
|
||||
"If you want to open it in another window use the link.\n",
|
||||
"\n",
|
||||
"Note that some UI features like live image previews won't work because the colab iframe blocks websockets."
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {
|
||||
"id": "hhhhhhhhhh",
|
||||
"cellView": "form"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"#@title <font size=\"5\" color=\"orange\">ComfyUI iFrame</font>\n",
|
||||
"import threading\n",
|
||||
"import time\n",
|
||||
"import socket\n",
|
||||
"def iframe_thread(port):\n",
|
||||
" while True:\n",
|
||||
" time.sleep(0.5)\n",
|
||||
" sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)\n",
|
||||
" result = sock.connect_ex(('127.0.0.1', port))\n",
|
||||
" if result == 0:\n",
|
||||
" break\n",
|
||||
" sock.close()\n",
|
||||
" from google.colab import output\n",
|
||||
" output.serve_kernel_port_as_iframe(port, height=1024)\n",
|
||||
" print(\"to open it in a window you can open this link here:\")\n",
|
||||
" output.serve_kernel_port_as_window(port)\n",
|
||||
"\n",
|
||||
"threading.Thread(target=iframe_thread, daemon=True, args=(8188,)).start()\n",
|
||||
"\n",
|
||||
"!python main.py --dont-print-server"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"accelerator": "GPU",
|
||||
"colab": {
|
||||
"provenance": [],
|
||||
"private_outputs": true,
|
||||
"gpuType": "T4",
|
||||
"include_colab_link": true
|
||||
},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"name": "python"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 0
|
||||
}
|
||||
@@ -1,21 +0,0 @@
|
||||
MIT License
|
||||
|
||||
Copyright (c) 2023 Jordan Thompson (WASasquatch)
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in all
|
||||
copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
||||
SOFTWARE.
|
||||
@@ -1,410 +0,0 @@
|
||||
# **WAS** Node Suite [](https://colab.research.google.com/github/WASasquatch/was-node-suite-comfyui/blob/main/ComfyUI_%2B_WAS_Node_Suite_and_ComfyUI_Manager.ipynb) [](https://hits.seeyoufarm.com) [](https://paypal.me/ThompsonJordan?country.x=US&locale.x=en_US)
|
||||
|
||||
<p align="center">
|
||||
<img src="https://user-images.githubusercontent.com/1151589/228982359-4a6215cc-3ca9-4c24-8a7b-d229d7bce277.png">
|
||||
</p>
|
||||
|
||||
### A node suite for [ComfyUI](https://github.com/comfyanonymous/ComfyUI) with many new nodes, such as image processing, text processing, and more.
|
||||
|
||||
#### [Share Workflows](https://github.com/WASasquatch/was-node-suite-comfyui/wiki/Workflow-Examples) to the workflows wiki. Preferably embedded PNGs with workflows, but JSON is OK too. [You can use this tool to add a workflow to a PNG file easily](https://colab.research.google.com/drive/1hQMjNUdhMQ3rw1Wcm3_umvmOMeS_K4s8?usp=sharing).
|
||||
#### Consider [donating to the project](https://paypal.me/ThompsonJordan?country.x=US&locale.x=en_US) to help it's continued development.
|
||||
|
||||
# Important Updates
|
||||
|
||||
- **12/15/2023** WAS-NS is not under active development. I do not have the time and have other obligations. Feel free to fork and continue the project. I will approve appropriate and beneficial PRs.
|
||||
- **[Updated 10/8/2023]** BLIP is now a shipped module of WAS-NS and no longer requires the BLIP Repo
|
||||
- **[Updated 5/29/2023]** `ASCII` **is deprecated**. The new preferred method of text node output is `STRING`. This is a change from `ASCII` so that it is more clear what data is being passed.
|
||||
- The `was_suite_config.json` will automatically set `use_legacy_ascii_text` to `false`.
|
||||
- [Video Nodes](https://github.com/WASasquatch/was-node-suite-comfyui#video-nodes) - There are two new video nodes, `Write to Video` and `Create Video from Path`. These are experimental nodes.
|
||||
|
||||
# Current Nodes:
|
||||
|
||||
<details>
|
||||
<summary>$\Large\color{orange}{Expand\ Node\ List}$</summary>
|
||||
|
||||
<br/>
|
||||
|
||||
- BLIP Model Loader: Load a BLIP model to input into the BLIP Analyze node
|
||||
- BLIP Analyze Image: Get a text caption from a image, or interrogate the image with a question.
|
||||
- Model will download automatically from default URL, but you can point the download to another location/caption model in `was_suite_config`
|
||||
- Models will be stored in `ComfyUI/models/blip/checkpoints/`
|
||||
- SAM Model Loader: Load a SAM Segmentation model
|
||||
- SAM Parameters: Define your SAM parameters for segmentation of a image
|
||||
- SAM Parameters Combine: Combine SAM parameters
|
||||
- SAM Image Mask: SAM image masking
|
||||
- Image Bounds: Bounds a image
|
||||
- Inset Image Bounds: Inset a image bounds
|
||||
- Bounded Image Blend: Blend bounds image
|
||||
- Bounded Image Blend with Mask: Blend a bounds image by mask
|
||||
- Bounded Image Crop: Crop a bounds image
|
||||
- Bounded Image Crop with Mask: Crop a bounds image by mask
|
||||
- Bus Node: condense the 5 common connectors into one, keep your workspace tidy (Model, Clip, VAE, Positive Conditioning, Negative Conditioning)
|
||||
- Cache Node: Cache Latnet, Tensor Batches (Image), and Conditioning to disk to use later.
|
||||
- CLIPTextEncode (NSP): Parse noodle soups from the NSP pantry, or parse wildcards from a directory containing A1111 style wildacrds.
|
||||
- Wildcards are in the style of `__filename__`, which also includes subdirectories like `__appearance/haircolour__` (if you noodle_key is set to `__`)
|
||||
- You can set a custom wildcards path in `was_suite_config.json` file with key:
|
||||
- ` "wildcards_path": "E:\\python\\automatic\\webui3\\stable-diffusion-webui\\extensions\\sd-dynamic-prompts\\wildcards"`
|
||||
- If no path is set the wildcards dir is located at the root of WAS Node Suite as `/wildcards`
|
||||
- CLIP Input Switch: Switch between two CLIP inputs based on a boolean switch.
|
||||
- CLIP Vision Input Switch: Switch between two CLIP Vision inputs based on a boolean switch.
|
||||
- Conditioning Input Switch: Switch between two conditioning inputs.
|
||||
- Constant Number
|
||||
- Control Net Model Input Switch: Switch between two Control Net Model inputs based on a boolean switch.
|
||||
- Create Grid Image: Create a image grid from images at a destination with customizable glob pattern. Optional border size and color.
|
||||
- Create Grid Image from Batch: Create a grid image from a batch tensor of images.
|
||||
- Create Morph Image: Create a GIF/APNG animation from two images, fading between them.
|
||||
- Create Morph Image by Path: Create a GIF/APNG animation from a path to a directory containing images, with optional pattern.
|
||||
- Create Video from Path: Create video from images from a specified path.
|
||||
- CLIPSeg Masking: Mask a image with CLIPSeg and return a raw mask
|
||||
- CLIPSeg Masking Batch: Create a batch image (from image inputs) and batch mask with CLIPSeg
|
||||
- Dictionary to Console: Print a dictionary input to the console
|
||||
- Image Analyze
|
||||
- Black White Levels
|
||||
- RGB Levels
|
||||
- Depends on `matplotlib`, will attempt to install on first run
|
||||
- Diffusers Hub Down-Loader: Download a diffusers model from the HuggingFace Hub and load it
|
||||
- Image SSAO (Ambient Occlusion): [Expiremental Beta Node] Create Screen Space Ambient Occlusion with a image and MiDaS depth approximation (or provided depth map).
|
||||
- Image SSDO (Direct Occlusion): [Expiremental Beta Node] Create a Screen Space Direct Occlusion with a image input. Direct Occlusion presents you with direct lighting highliths, similar to how Ambient Occlusion finds the crevices and shadowy areas around objets.
|
||||
- Image Aspect Ratio: Fetch image aspect ratio in float format, common format (eg 16:9), and in if the image is portrait, landscape, or square.
|
||||
- Image Batch: Create one batch out of multiple batched tensors.
|
||||
- Image Blank: Create a blank image in any color
|
||||
- Image Blend by Mask: Blend two images by a mask
|
||||
- Image Blend: Blend two images by opacity
|
||||
- Image Blending Mode: Blend two images by various blending modes
|
||||
- Image Bloom Filter: Apply a high-pass based bloom filter
|
||||
- Image Canny Filter: Apply a canny filter to a image
|
||||
- Image Chromatic Aberration: Apply chromatic aberration lens effect to a image like in sci-fi films, movie theaters, and video games
|
||||
- Image Color Palette
|
||||
- Generate a color palette based on the input image.
|
||||
- Depends on `scikit-learn`, will attempt to install on first run.
|
||||
- Supports color range of 8-256
|
||||
- Utilizes font in `./res/` unless unavailable, then it will utilize internal better then nothing font.
|
||||
- Image Crop Face: Crop a face out of a image
|
||||
- **Limitations:**
|
||||
- Sometimes no faces are found in badly generated images, or faces at angles
|
||||
- Sometimes face crop is black, this is because the padding is too large and intersected with the image edge. Use a smaller padding size.
|
||||
- face_recognition mode sometimes finds random things as faces. It also requires a [CUDA] GPU.
|
||||
- Only detects one face. This is a design choice to make it's use easy.
|
||||
- **Notes:**
|
||||
- Detection runs in succession. If nothing is found with the selected detection cascades, it will try the next available cascades file.
|
||||
- Image Crop Location: Crop a image to specified location in top, left, right, and bottom locations relating to the pixel dimensions of the image in X and Y coordinats.
|
||||
- Image Crop Square Location: Crop a location by X/Y center, creating a square crop around that point.
|
||||
- Image Displacement Warp: Warp a image by a displacement map image by a given amplitude.
|
||||
- Image Dragan Photography Filter: Apply a Andrzej Dragan photography style to a image
|
||||
- Image Edge Detection Filter: Detect edges in a image
|
||||
- Image Film Grain: Apply film grain to a image
|
||||
- Image Filter Adjustments: Apply various image adjustments to a image
|
||||
- Image Flip: Flip a image horizontal, or vertical
|
||||
- Image Gradient Map: Apply a gradient map to a image
|
||||
- Image Generate Gradient: Generate a gradient map with desired stops and colors
|
||||
- Image High Pass Filter: Apply a high frequency pass to the image returning the details
|
||||
- Image History Loader: Load images from history based on the Load Image Batch node. Can define max history in config file. *(requires restart to show last sessions files at this time)*
|
||||
- Image Input Switch: Switch between two image inputs based on a boolean switch
|
||||
- Image Levels Adjustment: Adjust the levels of a image
|
||||
- Image Load: Load a *image* from any path on the system, or a url starting with `http`
|
||||
- Image Median Filter: Apply a median filter to a image, such as to smooth out details in surfaces
|
||||
- Image Mix RGB Channels: Mix together RGB channels into a single iamge
|
||||
- Image Monitor Effects Filter: Apply various monitor effects to a image
|
||||
- Digital Distortion
|
||||
- A digital breakup distortion effect
|
||||
- Signal Distortion
|
||||
- A analog signal distortion effect on vertical bands like a CRT monitor
|
||||
- TV Distortion
|
||||
- A TV scanline and bleed distortion effect
|
||||
- Image Nova Filter: A image that uses a sinus frequency to break apart a image into RGB frequencies
|
||||
- Image Perlin Noise: Generate perlin noise
|
||||
- Image Perlin Power Fractal: Generate a perlin power fractal
|
||||
- Image Paste Face Crop: Paste face crop back on a image at it's original location and size
|
||||
- Features a better blending funciton than GFPGAN/CodeFormer so there shouldn't be visible seams, and coupled with Diffusion Result, looks better than GFPGAN/CodeFormer.
|
||||
- Image Paste Crop: Paste a crop (such as from Image Crop Location) at it's original location and size utilizing the `crop_data` node input. This uses a different blending algorithm then Image Paste Face Crop, which may be desired in certain instances.
|
||||
- Image Power Noise: Generate power-law noise
|
||||
- frequency: The frequency parameter controls the distribution of the noise across different frequencies. In the context of Fourier analysis, higher frequencies represent fine details or high-frequency components, while lower frequencies represent coarse details or low-frequency components. Adjusting the frequency parameter can result in different textures and levels of detail in the generated noise. The specific range and meaning of the frequency parameter may vary depending on the noise type.
|
||||
- attenuation: The attenuation parameter determines the strength or intensity of the noise. It controls how much the noise values deviate from the mean or central value. Higher values of attenuation lead to more significant variations and a stronger presence of noise, while lower values result in a smoother and less noticeable noise. The specific range and interpretation of the attenuation parameter may vary depending on the noise type.
|
||||
- noise_type: The tyoe of Power-Law noise to generate (white, grey, pink, green, blue)
|
||||
- Image Paste Crop by Location: Paste a crop top a custom location. This uses the same blending algorithm as Image Paste Crop.
|
||||
- Image Pixelate: Turn a image into pixel art! Define the max number of colors, the pixelation mode, the random state, and max iterations, and max those sprites shine.
|
||||
- Image Remove Background (Alpha): Remove the background from a image by threshold and tolerance.
|
||||
- Image Remove Color: Remove a color from a image and replace it with another
|
||||
- Image Resize
|
||||
- Image Rotate: Rotate an image
|
||||
- Image Rotate Hue: Rotate the hue of a image. A hue_shift of `0.0` would represent no change, and `1.0` would represent a full circle of the hue, and also exhibit no change.
|
||||
- Image Save: A save image node with format support and path support.
|
||||
- `show_history` will show previously saved images with the WAS Save Image node. ComfyUI unfortunately resizes displayed images to the same size however, so if images are in different sizes it will force them in a different size.
|
||||
- Doesn't display images saved outside `/ComfyUI/output/`
|
||||
- You can save as `webp` if you have webp available to you system. On windows you can get that support with this [precompiled libarary](https://storage.googleapis.com/downloads.webmproject.org/releases/webp/libwebp-1.3.0-windows-x64.zip) from the [webp project](https://developers.google.com/speed/webp/download). On linux you can run `apt-get install webp`.
|
||||
- Image Seamless Texture: Create a seamless texture out of a image with optional tiling
|
||||
- Image Select Channel: Select a single channel of an RGB image
|
||||
- Image Select Color: Return the select image only on a black canvas
|
||||
- Image Shadows and Highlights: Adjust the shadows and highlights of an image
|
||||
- Image Size to Number: Get the `width` and `height` of an input image to use with **Number** nodes.
|
||||
- Image Stitch: Stitch images together on different sides with optional feathering blending between them.
|
||||
- Image Style Filter: Style a image with Pilgram instragram-like filters
|
||||
- Depends on `pilgram` module
|
||||
- Image Threshold: Return the desired threshold range of a image
|
||||
- Image Tile: Split a image up into a image batch of tiles. Can be used with Tensor Batch to Image to select a individual tile from the batch.
|
||||
- Image Transpose
|
||||
- Image fDOF Filter: Apply a fake depth of field effect to an image
|
||||
- Image to Latent Mask: Convert a image into a latent mask
|
||||
- Image to Noise: Convert a image into noise, useful for init blending or init input to theme a diffusion.
|
||||
- Images to RGB: Convert a tensor image batch to RGB if they are RGBA or some other mode.
|
||||
- Image to Seed: Convert a image to a reproducible seed
|
||||
- Image Voronoi Noise Filter
|
||||
- A custom implementation of the worley voronoi noise diagram
|
||||
- Input Switch (Disable until `*` wildcard fix)
|
||||
- KSampler (WAS): A sampler that accepts a seed as a node inputs
|
||||
- KSampler Cycle: A KSampler able to do HR pass loops, you can specify an upscale factor, and how many steps to achieve that factor. Accepts a upscale_model, as well as a 1x processor model. A secondary diffusion model can also be used.
|
||||
- Load Cache: Load cached Latent, Tensor Batch (image), and Conditioning files.
|
||||
- Load Text File
|
||||
- Now supports outputting a dictionary named after the file, or custom input.
|
||||
- The dictionary contains a list of all lines in the file.
|
||||
- Load Batch Images
|
||||
- Increment images in a folder, or fetch a single image out of a batch.
|
||||
- Will reset it's place if the path, or pattern is changed.
|
||||
- pattern is a glob that allows you to do things like `**/*` to get all files in the directory and subdirectory
|
||||
or things like `*.jpg` to select only JPEG images in the directory specified.
|
||||
- Mask to Image: Convert `MASK` to `IMAGE`
|
||||
- Mask Batch to Mask: Return a single mask from a batch of masks
|
||||
- Mask Invert: Invert a mask.
|
||||
- Mask Add: Add masks together.
|
||||
- Mask Subtract: Subtract from a mask by another.
|
||||
- Mask Dominant Region: Return the dominant region in a mask (the largest area)
|
||||
- Mask Minority Region: Return the smallest region in a mask (the smallest area)
|
||||
- Mask Crop Dominant Region: Crop mask to the dominant region with optional padding in pixels
|
||||
- Mask Crop Minority Region: Crop mask to the minority region with optional padding in pixels
|
||||
- Mask Crop Region: Crop to dominant or minority region and return `crop_data` for pasting back. Additionally outputs region location and size for other nodes like Crop Image Location.
|
||||
- Mask Arbitrary Region: Return a region that most closely matches the size input (size is not a direct representation of pixels, but approximate)
|
||||
- Mask Smooth Region: Smooth the boundaries of a mask
|
||||
- Mask Erode Region: Erode the boundaries of a mask
|
||||
- Mask Dilate Region: Dilate the boundaries of a mask
|
||||
- Mask Fill Region: Fill holes within the masks regions
|
||||
- Mask Ceiling Region": Return only white pixels within a offset range.
|
||||
- Mask Floor Region: Return the lower most pixel values as white (255)
|
||||
- Mask Threshold Region: Apply a thresholded image between a black value and white value
|
||||
- Mask Gaussian Region: Apply a Gaussian blur to the mask
|
||||
- Masks Combine Masks: Combine 2 or more masks into one mask.
|
||||
- Masks Combine Batch: Combine batched masks into one mask.
|
||||
- Model Input Switch: Switch between two model inputs based on a boolean switch
|
||||
- ComfyUI Loaders: A set of ComfyUI loaders that also output a string that contains the name of the model being loaded.
|
||||
- Latent Noise Injection: Inject latent noise into a latent image
|
||||
- Latent Size to Number: Latent sizes in tensor width/height
|
||||
- Latent Upscale by Factor: Upscale a latent image by a factor
|
||||
- Latent Input Switch: Switch between two latent inputs based on a boolean switch
|
||||
- Logic Boolean: A simple `1` or `0` output to use with logic
|
||||
- Logic Boolean Primitive: True/False boolean input, to use with native boolean nodes
|
||||
- Logic AND: Given 2 booleans, performs "AND"
|
||||
- Logic OR: Given 2 booleans, performs "OR"
|
||||
- Logic XOR: Given 2 booleans, performs "!="
|
||||
- Logic NOT: Given 1 boolean, returns the opposite
|
||||
- Lora Input Switch: Switch between two LORAs based on a boolean switch
|
||||
- MiDaS Model Loader: Load a MiDaS model as an optional input for MiDaS Depth Approximation
|
||||
- MiDaS Depth Approximation: Produce a depth approximation of a single image input
|
||||
- MiDaS Mask Image: Mask a input image using MiDaS with a desired color
|
||||
- Number Operation
|
||||
- Number to Seed
|
||||
- Number to Float
|
||||
- Number Input Switch: Switch between two number inputs based on a boolean switch
|
||||
- Number Input Condition: Compare between two inputs or against the A input
|
||||
- Number to Int
|
||||
- Number to String
|
||||
- Number to Text
|
||||
- Boolean to Text
|
||||
- Perlin Power Fractal Latent: Create a power fractal based latent image. Doesn't work with all samplers (unless you add noise).
|
||||
- Random Number
|
||||
- Random integer between min and max (inclusive), uniformly distributed random number
|
||||
- Random float between min and max (inclusive), uniformly distributed random number
|
||||
- Random number from 0 to 1 inclusive, this will be a 0 or 1 boolean if you use the 'int' output
|
||||
- Random shuffled list of integers between min and max inclusive. E.g. if min=0 and max=3, a possible outcome would be the string '3,1,2,0'
|
||||
- Save Text File: Save a text string to a file
|
||||
- Samples Passthrough (Stat System): Logs RAM, VRAM, and Disk usage to the console.
|
||||
- Seed: Return a seed
|
||||
- Tensor Batch to Image: Select a single image out of a latent batch for post processing with filters
|
||||
- Text Add Tokens: Add custom tokens to parse in filenames or other text.
|
||||
- Text Add Token by Input: Add custom token by inputs representing single **single line** name and value of the token
|
||||
- Text Compare: Compare two strings. Returns a boolean if they are the same, a score of similarity, and the similarity or difference text.
|
||||
- Text Concatenate: Merge two strings
|
||||
- Text Dictionary Update: Merge two dictionaries
|
||||
- Text Dictionary Get: Get a value from a dictionary (as a string)
|
||||
- Text Dictionary Convert: Convert text to dictionary object
|
||||
- Text Dictionary New: Create a new dictionary
|
||||
- Text Dictionary Keys: Returns the keys, as a list from a dictionary object
|
||||
- Text Dictionary To Text: Returns the dictionary as text
|
||||
- Text File History: Show previously opened text files *(requires restart to show last sessions files at this time)*
|
||||
- Text Find and Replace: Find and replace a substring in a string
|
||||
- Text Find and Replace by Dictionary: Replace substrings in a ASCII text input with a dictionary.
|
||||
- The dictionary keys are used as the key to replace, and the list of lines it contains chosen at random based on the seed.
|
||||
- Text Input Switch: Switch between two text inputs
|
||||
- Text List: Create a list of text strings
|
||||
- Text Load Line From File: Load lines from a file sequentially each *batch prompt* run, or select a line index.
|
||||
- Text Concatenate: Merge lists of strings
|
||||
- Text Contains: Checks if substring is in another string (case insensitive optional)
|
||||
- Text Multiline: Write a multiline text string
|
||||
- Text Parse A1111 Embeddings: Convert embeddings filenames in your prompts to `embedding:[filename]]` format based on your `/ComfyUI/models/embeddings/` files.
|
||||
- Text Parse Noodle Soup Prompts: Parse NSP in a text input
|
||||
- Text Parse Tokens: Parse custom tokens in text.
|
||||
- Text Random Line: Select a random line from a text input string
|
||||
- Text Random Prompt: Feeling lucky? Get a random prompt based on a search seed, such as "superhero"
|
||||
- Text String: Write a single line text string value
|
||||
- Text String Truncate: Truncate a string from the beginning or end by characters or words.
|
||||
- Text to Conditioning: Convert a text string to conditioning.
|
||||
- True Random.org Number Generator: Generate a truly random number online from atmospheric noise with [Random.org](https://random.org/)
|
||||
- [Get your API key from your account page](https://accounts.random.org/)
|
||||
- Upscale Model Input Switch: Switch between two Upscale Models inputs based on a boolean switch.
|
||||
- Write to Morph GIF: Write a new frame to an existing GIF (or create new one) with interpolation between frames.
|
||||
- Write to Video: Write a frame as you generate to a video (Best used with FFV1 for lossless images)
|
||||
- VAE Input Switch: Switch between two VAE inputs based on boolean input
|
||||
</details>
|
||||
|
||||
|
||||
<br>
|
||||
|
||||
### Extra Nodes
|
||||
|
||||
- CLIPTextEncode (BlenderNeko Advanced + NSP): Only available if you have BlenderNeko's [Advanced CLIP Text Encode](https://github.com/BlenderNeko/ComfyUI_ADV_CLIP_emb). Allows for NSP and Wildcard use with their advanced CLIPTextEncode.
|
||||
|
||||
|
||||
### Notes:
|
||||
|
||||
- **CLIPTextEncode (NSP)** and **CLIPTextEncode (BlenderNeko Advanced + NSP)**: Accept dynamic prompts in `<option1|option2|option3>` format. This will respect the nodes input seed to yield reproducible results like NSP and Wildcards.
|
||||
- **CLIPTextEncode (NSP)** and **CLIPTextEncode (BlenderNeko Advanced + NSP)**: Assign variables with `$|prompt words|$` format. You can then print this word again within the prompt with the number corresponding the order you created them. So the first prompt var would be printed with `$1` and the second with `$2` and so on.
|
||||
|
||||
---
|
||||
|
||||
|
||||
## Video Nodes
|
||||
|
||||
### Codecs
|
||||
You can use codecs that are available to your ffmpeg binaries by adding their fourcc ID (in one string), and appropriate container extension to the `was_suite_config.json`
|
||||
|
||||
Example [H264 Codecs](https://github.com/cisco/openh264/releases/tag/v1.8.0) (Defaults)
|
||||
```
|
||||
"ffmpeg_extra_codecs": {
|
||||
"avc1": ".mp4",
|
||||
"h264": ".mkv"
|
||||
}
|
||||
```
|
||||
|
||||
### Notes
|
||||
- For now I am only supporting **Windows** installations for video nodes.
|
||||
- I do not have access to Mac or a stand-alone linux distro. If you get them working and want to PR a patch/directions, feel free.
|
||||
- Video nodes require [FFMPEG](https://ffmpeg.org/download.html). You should download the proper FFMPEG binaries for you system and set the FFMPEG path in the config file.
|
||||
- Additionally, if you want to use H264 codec need to [download OpenH264 1.8.0](https://github.com/cisco/openh264/releases/tag/v1.8.0) and place it in the root of ComfyUI (Example: `C:\ComfyUI_windows_portable`).
|
||||
- FFV1 will complain about invalid container. You can ignore this. The resulting MKV file is readable. I have not figured out what this issue is about. Documentaion tells me to use MKV, but it's telling me it's unsupported.
|
||||
- If you know how to resolve this, I'd love a PR
|
||||
- `Write to Video` node should use a lossless video codec or when it copies frames, and reapplies compression, it will start expontentially ruining the starting frames run to run.
|
||||
|
||||
---
|
||||
|
||||
# Text Tokens
|
||||
Text tokens can be used in the Save Text File and Save Image nodes. You can also add your own custom tokens with the Text Add Tokens node.
|
||||
|
||||
The token name can be anything excluding the `:` character to define your token. It can also be simple Regular Expressions.
|
||||
|
||||
## Built-in Tokens
|
||||
- [time]
|
||||
- The current system microtime
|
||||
- [time(`format_code`)]
|
||||
- The current system time in human readable format. Utilizing [datetime](https://docs.python.org/3/library/datetime.html) formatting
|
||||
- Example: `[hostname]_[time]__[time(%Y-%m-%d__%I-%M%p)]` would output: **SKYNET-MASTER_1680897261__2023-04-07__07-54PM**
|
||||
- [hostname]
|
||||
- The hostname of the system executing ComfyUI
|
||||
- [cuda_device]
|
||||
- The cuda device from `comfy.model_management.get_cuda_device()`
|
||||
- [cuda_name]
|
||||
- The cuda name from `comfy.model_management.get_cuda_device_name()`
|
||||
|
||||
<br>
|
||||
|
||||
<details>
|
||||
<summary>$\color{orange}{Expand\ Date\ Code\ List}$</summary>
|
||||
|
||||
<br>
|
||||
|
||||
| Directive | Meaning | Example | Notes |
|
||||
| --- | --- | --- | --- |
|
||||
| %a | Weekday as locale’s abbreviated name. | Sun, Mon, …, Sat (en_US); So, Mo, …, Sa (de_DE) | (1) |
|
||||
| %A | Weekday as locale’s full name. | Sunday, Monday, …, Saturday (en_US); Sonntag, Montag, …, Samstag (de_DE) | (1) |
|
||||
| %w | Weekday as a decimal number, where 0 is Sunday and 6 is Saturday. | 0, 1, …, 6 | |
|
||||
| %d | Day of the month as a zero-padded decimal number. | 01, 02, …, 31 | (9) |
|
||||
| %b | Month as locale’s abbreviated name. | Jan, Feb, …, Dec (en_US); Jan, Feb, …, Dez (de_DE) | (1) |
|
||||
| %B | Month as locale’s full name. | January, February, …, December (en_US); Januar, Februar, …, Dezember (de_DE) | (1) |
|
||||
| %m | Month as a zero-padded decimal number. | 01, 02, …, 12 | (9) |
|
||||
| %y | Year without century as a zero-padded decimal number. | 00, 01, …, 99 | (9) |
|
||||
| %Y | Year with century as a decimal number. | 0001, 0002, …, 2013, 2014, …, 9998, 9999 | (2) |
|
||||
| %H | Hour (24-hour clock) as a zero-padded decimal number. | 00, 01, …, 23 | (9) |
|
||||
| %I | Hour (12-hour clock) as a zero-padded decimal number. | 01, 02, …, 12 | (9) |
|
||||
| %p | Locale’s equivalent of either AM or PM. | AM, PM (en_US); am, pm (de_DE) | (1), (3) |
|
||||
| %M | Minute as a zero-padded decimal number. | 00, 01, …, 59 | (9) |
|
||||
| %S | Second as a zero-padded decimal number. | 00, 01, …, 59 | (4), (9) |
|
||||
| %f | Microsecond as a decimal number, zero-padded to 6 digits. | 000000, 000001, …, 999999 | (5) |
|
||||
| %z | UTC offset in the form ±HHMM[SS[.ffffff]] (empty string if the object is naive). | (empty), +0000, -0400, +1030, +063415, -030712.345216 | (6) |
|
||||
| %Z | Time zone name (empty string if the object is naive). | (empty), UTC, GMT | (6) |
|
||||
| %j | Day of the year as a zero-padded decimal number. | 001, 002, …, 366 | (9) |
|
||||
| %U | Week number of the year (Sunday as the first day of the week) as a zero-padded decimal number. All days in a new year preceding the first Sunday are considered to be in week 0. | 00, 01, …, 53 | (7), (9) |
|
||||
| %W | Week number of the year (Monday as the first day of the week) as a zero-padded decimal number. All days in a new year preceding the first Monday are considered to be in week 0. | 00, 01, …, 53 | (7), (9) |
|
||||
| %c | Locale’s appropriate date and time representation. | Tue Aug 16 21:30:00 1988 (en_US); Di 16 Aug 21:30:00 1988 (de_DE) | (1) |
|
||||
| %x | Locale’s appropriate date representation. | 08/16/88 (None); 08/16/1988 (en_US); 16.08.1988 (de_DE) | (1) |
|
||||
| %X | Locale’s appropriate time representation. | 21:30:00 (en_US); 21:30:00 (de_DE) | (1) |
|
||||
| %% | A literal '%' character. | % | |
|
||||
|
||||
</details>
|
||||
|
||||
<br>
|
||||
|
||||
---
|
||||
|
||||
# Other Features
|
||||
|
||||
### Import AUTOMATIC1111 WebUI Styles
|
||||
When using the latest builds of WAS Node Suite a `was_suite_config.json` file will be generated (if it doesn't exist). In this file you can setup a A1111 styles import.
|
||||
|
||||
- Run ComfyUI to generate the new `/custom-nodes/was-node-suite-comfyui/was_Suite_config.json` file.
|
||||
- Open the `was_suite_config.json` file with a text editor.
|
||||
- Replace the `webui_styles` value from `None` to the path of your A1111 styles file called **styles.csv**. Be sure to use double backslashes for Windows paths.
|
||||
- Example `C:\\python\\stable-diffusion-webui\\styles.csv`
|
||||
- Restart ComfyUI
|
||||
- Select a style with the `Prompt Styles Node`.
|
||||
- The first ASCII output is your positive prompt, and the second ASCII output is your negative prompt.
|
||||
|
||||
You can set `webui_styles_persistent_update` to `true` to update the WAS Node Suite styles from WebUI every start of ComfyUI
|
||||
|
||||
# Recommended Installation:
|
||||
If you're running on Linux, or non-admin account on windows you'll want to ensure `/ComfyUI/custom_nodes`, `was-node-suite-comfyui`, and `WAS_Node_Suite.py` has write permissions.
|
||||
|
||||
There is now a **install.bat** you can run to install to portable if detected. Otherwise it will default to system and assume you followed ConfyUI's manual installation steps.
|
||||
|
||||
- Navigate to your `/ComfyUI/custom_nodes/` folder
|
||||
- Run `git clone https://github.com/WASasquatch/was-node-suite-comfyui/`
|
||||
- Navigate to your `was-node-suite-comfyui` folder
|
||||
- Portable/venv:
|
||||
- Run `path/to/ComfUI/python_embeded/python.exe -s -m pip install -r requirements.txt`
|
||||
- With system python
|
||||
- Run `pip install -r requirements.txt`
|
||||
- Start ComfyUI
|
||||
- WAS Suite should uninstall legacy nodes automatically for you.
|
||||
- Tools will be located in the WAS Suite menu.
|
||||
|
||||
## Alternate [Legacy] Installation:
|
||||
If you're running on Linux, or non-admin account on windows you'll want to ensure `/ComfyUI/custom_nodes`, and `WAS_Node_Suite.py` has write permissions.
|
||||
|
||||
- Download `WAS_Node_Suite.py`
|
||||
- Move the file to your `/ComfyUI/custom_nodes/` folder
|
||||
- WAS Node Suite will attempt install dependencies on it's own, but you may need to manually do so. The dependencies required are in the `requirements.txt` on this repo. See installation steps above.
|
||||
- If this process fails attempt the following:
|
||||
- Navigate to your `was-node-suite-comfyui` folder
|
||||
- Portable/venv:
|
||||
- Run `path/to/ComfUI/python_embeded/python.exe -s -m pip install -r requirements.txt`
|
||||
- With system python
|
||||
- Run `pip install -r requirements.txt`
|
||||
- Start, or Restart ComfyUI
|
||||
- WAS Suite should uninstall legacy nodes automatically for you.
|
||||
- Tools will be located in the WAS Suite menu.
|
||||
|
||||
This method will not install the resources required for Image Crop Face node, and you'll have to download the [./res/](https://github.com/WASasquatch/was-node-suite-comfyui/tree/main/res) folder yourself.
|
||||
|
||||
## Installing on Colab
|
||||
Create a new cell and add the following code, then run the cell. You may need to edit the path to your `custom_nodes` folder. You can also use the [colab hosted here](https://colab.research.google.com/github/WASasquatch/comfyui-colab-was-node-suite/blob/main/ComfyUI_%2B_WAS_Node_Suite.ipynb)
|
||||
|
||||
- `!git clone https://github.com/WASasquatch/was-node-suite-comfyui /content/ComfyUI/custom_nodes/was-node-suite-comfyui`
|
||||
- `!pip install -r /content/ComfyUI/custom_nodes/was-node-suite-comfyui/requirements.txt`
|
||||
- Restart Colab Runtime (don't disconnect)
|
||||
- Tools will be located in the WAS Suite menu.
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1,3 +0,0 @@
|
||||
from .WAS_Node_Suite import NODE_CLASS_MAPPINGS
|
||||
|
||||
__all__ = ['NODE_CLASS_MAPPINGS']
|
||||
BIN
Binary file not shown.
BIN
Binary file not shown.
@@ -1,16 +0,0 @@
|
||||
@echo off
|
||||
|
||||
set "requirements_txt=%~dp0\requirements.txt"
|
||||
set "python_exec=..\..\..\python_embeded\python.exe"
|
||||
|
||||
echo Installing WAS-NS ...
|
||||
|
||||
if exist "%python_exec%" (
|
||||
echo Installing with ComfyUI Portable
|
||||
"%python_exec%" -s -m pip install -r "%requirements_txt%"
|
||||
) else (
|
||||
echo Installing with system Python
|
||||
pip install -r "%requirements_txt%"
|
||||
)
|
||||
|
||||
pause
|
||||
@@ -1,20 +0,0 @@
|
||||
@echo off
|
||||
|
||||
set "requirements_txt=%~dp0\requirements.txt"
|
||||
set "python_exec=..\..\..\python_embeded\python.exe"
|
||||
|
||||
echo Installing WAS-NS ...
|
||||
|
||||
if exist "%python_exec%" (
|
||||
echo Installing with ComfyUI Portable
|
||||
for /f "delims=" %%i in (%requirements_txt%) do (
|
||||
%python_exec% -s -m pip install "%%i"
|
||||
)
|
||||
) else (
|
||||
echo Installing with system Python
|
||||
for /f "delims=" %%i in (%requirements_txt%) do (
|
||||
pip install "%%i"
|
||||
)
|
||||
)
|
||||
|
||||
pause
|
||||
-1
@@ -1 +0,0 @@
|
||||
|
||||
-21
@@ -1,21 +0,0 @@
|
||||
{
|
||||
"architectures": [
|
||||
"BertModel"
|
||||
],
|
||||
"attention_probs_dropout_prob": 0.1,
|
||||
"hidden_act": "gelu",
|
||||
"hidden_dropout_prob": 0.1,
|
||||
"hidden_size": 768,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 3072,
|
||||
"layer_norm_eps": 1e-12,
|
||||
"max_position_embeddings": 512,
|
||||
"model_type": "bert",
|
||||
"num_attention_heads": 12,
|
||||
"num_hidden_layers": 12,
|
||||
"pad_token_id": 0,
|
||||
"type_vocab_size": 2,
|
||||
"vocab_size": 30522,
|
||||
"encoder_width": 768,
|
||||
"add_cross_attention": true
|
||||
}
|
||||
-33
@@ -1,33 +0,0 @@
|
||||
image_root: '/export/share/datasets/vision/coco/images/'
|
||||
ann_root: 'annotation'
|
||||
coco_gt_root: 'annotation/coco_gt'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth'
|
||||
|
||||
# size of vit model; base or large
|
||||
vit: 'base'
|
||||
vit_grad_ckpt: False
|
||||
vit_ckpt_layer: 0
|
||||
batch_size: 32
|
||||
init_lr: 1e-5
|
||||
|
||||
# vit: 'large'
|
||||
# vit_grad_ckpt: True
|
||||
# vit_ckpt_layer: 5
|
||||
# batch_size: 16
|
||||
# init_lr: 2e-6
|
||||
|
||||
image_size: 384
|
||||
|
||||
# generation configs
|
||||
max_length: 20
|
||||
min_length: 5
|
||||
num_beams: 3
|
||||
prompt: 'a picture of '
|
||||
|
||||
# optimizer
|
||||
weight_decay: 0.05
|
||||
min_lr: 0
|
||||
max_epoch: 5
|
||||
|
||||
-21
@@ -1,21 +0,0 @@
|
||||
{
|
||||
"architectures": [
|
||||
"BertModel"
|
||||
],
|
||||
"attention_probs_dropout_prob": 0.1,
|
||||
"hidden_act": "gelu",
|
||||
"hidden_dropout_prob": 0.1,
|
||||
"hidden_size": 768,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 3072,
|
||||
"layer_norm_eps": 1e-12,
|
||||
"max_position_embeddings": 512,
|
||||
"model_type": "bert",
|
||||
"num_attention_heads": 12,
|
||||
"num_hidden_layers": 12,
|
||||
"pad_token_id": 0,
|
||||
"type_vocab_size": 2,
|
||||
"vocab_size": 30524,
|
||||
"encoder_width": 768,
|
||||
"add_cross_attention": true
|
||||
}
|
||||
-21
@@ -1,21 +0,0 @@
|
||||
image_root: '/export/share/datasets/vision/NLVR2/'
|
||||
ann_root: 'annotation'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_nlvr.pth'
|
||||
|
||||
#size of vit model; base or large
|
||||
vit: 'base'
|
||||
batch_size_train: 16
|
||||
batch_size_test: 64
|
||||
vit_grad_ckpt: False
|
||||
vit_ckpt_layer: 0
|
||||
max_epoch: 15
|
||||
|
||||
image_size: 384
|
||||
|
||||
# optimizer
|
||||
weight_decay: 0.05
|
||||
init_lr: 3e-5
|
||||
min_lr: 0
|
||||
|
||||
-15
@@ -1,15 +0,0 @@
|
||||
image_root: '/export/share/datasets/vision/nocaps/'
|
||||
ann_root: 'annotation'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth'
|
||||
|
||||
vit: 'base'
|
||||
batch_size: 32
|
||||
|
||||
image_size: 384
|
||||
|
||||
max_length: 20
|
||||
min_length: 5
|
||||
num_beams: 3
|
||||
prompt: 'a picture of '
|
||||
-27
@@ -1,27 +0,0 @@
|
||||
train_file: ['/export/share/junnan-li/VL_pretrain/annotation/coco_karpathy_train.json',
|
||||
'/export/share/junnan-li/VL_pretrain/annotation/vg_caption.json',
|
||||
]
|
||||
laion_path: ''
|
||||
|
||||
# size of vit model; base or large
|
||||
vit: 'base'
|
||||
vit_grad_ckpt: False
|
||||
vit_ckpt_layer: 0
|
||||
|
||||
image_size: 224
|
||||
batch_size: 75
|
||||
|
||||
queue_size: 57600
|
||||
alpha: 0.4
|
||||
|
||||
# optimizer
|
||||
weight_decay: 0.05
|
||||
init_lr: 3e-4
|
||||
min_lr: 1e-6
|
||||
warmup_lr: 1e-6
|
||||
lr_decay_rate: 0.9
|
||||
max_epoch: 20
|
||||
warmup_steps: 3000
|
||||
|
||||
|
||||
|
||||
-34
@@ -1,34 +0,0 @@
|
||||
image_root: '/export/share/datasets/vision/coco/images/'
|
||||
ann_root: 'annotation'
|
||||
dataset: 'coco'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_coco.pth'
|
||||
|
||||
# size of vit model; base or large
|
||||
|
||||
vit: 'base'
|
||||
batch_size_train: 32
|
||||
batch_size_test: 64
|
||||
vit_grad_ckpt: True
|
||||
vit_ckpt_layer: 4
|
||||
init_lr: 1e-5
|
||||
|
||||
# vit: 'large'
|
||||
# batch_size_train: 16
|
||||
# batch_size_test: 32
|
||||
# vit_grad_ckpt: True
|
||||
# vit_ckpt_layer: 12
|
||||
# init_lr: 5e-6
|
||||
|
||||
image_size: 384
|
||||
queue_size: 57600
|
||||
alpha: 0.4
|
||||
k_test: 256
|
||||
negative_all_rank: True
|
||||
|
||||
# optimizer
|
||||
weight_decay: 0.05
|
||||
min_lr: 0
|
||||
max_epoch: 6
|
||||
|
||||
-34
@@ -1,34 +0,0 @@
|
||||
image_root: '/export/share/datasets/vision/flickr30k/'
|
||||
ann_root: 'annotation'
|
||||
dataset: 'flickr'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_flickr.pth'
|
||||
|
||||
# size of vit model; base or large
|
||||
|
||||
vit: 'base'
|
||||
batch_size_train: 32
|
||||
batch_size_test: 64
|
||||
vit_grad_ckpt: True
|
||||
vit_ckpt_layer: 4
|
||||
init_lr: 1e-5
|
||||
|
||||
# vit: 'large'
|
||||
# batch_size_train: 16
|
||||
# batch_size_test: 32
|
||||
# vit_grad_ckpt: True
|
||||
# vit_ckpt_layer: 10
|
||||
# init_lr: 5e-6
|
||||
|
||||
image_size: 384
|
||||
queue_size: 57600
|
||||
alpha: 0.4
|
||||
k_test: 128
|
||||
negative_all_rank: False
|
||||
|
||||
# optimizer
|
||||
weight_decay: 0.05
|
||||
min_lr: 0
|
||||
max_epoch: 6
|
||||
|
||||
-12
@@ -1,12 +0,0 @@
|
||||
video_root: '/export/share/dongxuli/data/msrvtt_retrieval/videos'
|
||||
ann_root: 'annotation'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_coco.pth'
|
||||
|
||||
# size of vit model; base or large
|
||||
vit: 'base'
|
||||
batch_size: 64
|
||||
k_test: 128
|
||||
image_size: 384
|
||||
num_frm_test: 8
|
||||
-25
@@ -1,25 +0,0 @@
|
||||
vqa_root: '/export/share/datasets/vision/VQA/Images/mscoco/' #followed by train2014/
|
||||
vg_root: '/export/share/datasets/vision/visual-genome/' #followed by image/
|
||||
train_files: ['vqa_train','vqa_val','vg_qa']
|
||||
ann_root: 'annotation'
|
||||
|
||||
# set pretrained as a file path or an url
|
||||
pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_vqa_capfilt_large.pth'
|
||||
|
||||
# size of vit model; base or large
|
||||
vit: 'base'
|
||||
batch_size_train: 16
|
||||
batch_size_test: 32
|
||||
vit_grad_ckpt: False
|
||||
vit_ckpt_layer: 0
|
||||
init_lr: 2e-5
|
||||
|
||||
image_size: 480
|
||||
|
||||
k_test: 128
|
||||
inference: 'rank'
|
||||
|
||||
# optimizer
|
||||
weight_decay: 0.05
|
||||
min_lr: 0
|
||||
max_epoch: 10
|
||||
@@ -1,955 +0,0 @@
|
||||
'''
|
||||
* Copyright (c) 2022, salesforce.com, inc.
|
||||
* All rights reserved.
|
||||
* SPDX-License-Identifier: BSD-3-Clause
|
||||
* For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
|
||||
* By Junnan Li
|
||||
* Based on huggingface code base
|
||||
* https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert
|
||||
'''
|
||||
|
||||
import math
|
||||
import os
|
||||
import warnings
|
||||
from dataclasses import dataclass
|
||||
from typing import Optional, Tuple
|
||||
|
||||
import torch
|
||||
from torch import Tensor, device, dtype, nn
|
||||
import torch.utils.checkpoint
|
||||
from torch import nn
|
||||
from torch.nn import CrossEntropyLoss
|
||||
import torch.nn.functional as F
|
||||
|
||||
from transformers.activations import ACT2FN
|
||||
from transformers.file_utils import (
|
||||
ModelOutput,
|
||||
)
|
||||
from transformers.modeling_outputs import (
|
||||
BaseModelOutputWithPastAndCrossAttentions,
|
||||
BaseModelOutputWithPoolingAndCrossAttentions,
|
||||
CausalLMOutputWithCrossAttentions,
|
||||
MaskedLMOutput,
|
||||
MultipleChoiceModelOutput,
|
||||
NextSentencePredictorOutput,
|
||||
QuestionAnsweringModelOutput,
|
||||
SequenceClassifierOutput,
|
||||
TokenClassifierOutput,
|
||||
)
|
||||
from transformers.modeling_utils import (
|
||||
PreTrainedModel,
|
||||
apply_chunking_to_forward,
|
||||
find_pruneable_heads_and_indices,
|
||||
prune_linear_layer,
|
||||
)
|
||||
from transformers.utils import logging
|
||||
from transformers.models.bert.configuration_bert import BertConfig
|
||||
|
||||
|
||||
logger = logging.get_logger(__name__)
|
||||
|
||||
|
||||
class BertEmbeddings(nn.Module):
|
||||
"""Construct the embeddings from word and position embeddings."""
|
||||
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
|
||||
self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
|
||||
|
||||
# self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
|
||||
# any TensorFlow checkpoint file
|
||||
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
||||
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
||||
|
||||
# position_ids (1, len position emb) is contiguous in memory and exported when serialized
|
||||
self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
|
||||
self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
|
||||
|
||||
self.config = config
|
||||
|
||||
def forward(
|
||||
self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
|
||||
):
|
||||
if input_ids is not None:
|
||||
input_shape = input_ids.size()
|
||||
else:
|
||||
input_shape = inputs_embeds.size()[:-1]
|
||||
|
||||
seq_length = input_shape[1]
|
||||
|
||||
if position_ids is None:
|
||||
position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
|
||||
|
||||
if inputs_embeds is None:
|
||||
inputs_embeds = self.word_embeddings(input_ids)
|
||||
|
||||
embeddings = inputs_embeds
|
||||
|
||||
if self.position_embedding_type == "absolute":
|
||||
position_embeddings = self.position_embeddings(position_ids)
|
||||
embeddings += position_embeddings
|
||||
embeddings = self.LayerNorm(embeddings)
|
||||
embeddings = self.dropout(embeddings)
|
||||
return embeddings
|
||||
|
||||
|
||||
class BertSelfAttention(nn.Module):
|
||||
def __init__(self, config, is_cross_attention):
|
||||
super().__init__()
|
||||
self.config = config
|
||||
if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
|
||||
raise ValueError(
|
||||
"The hidden size (%d) is not a multiple of the number of attention "
|
||||
"heads (%d)" % (config.hidden_size, config.num_attention_heads)
|
||||
)
|
||||
|
||||
self.num_attention_heads = config.num_attention_heads
|
||||
self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
|
||||
self.all_head_size = self.num_attention_heads * self.attention_head_size
|
||||
|
||||
self.query = nn.Linear(config.hidden_size, self.all_head_size)
|
||||
if is_cross_attention:
|
||||
self.key = nn.Linear(config.encoder_width, self.all_head_size)
|
||||
self.value = nn.Linear(config.encoder_width, self.all_head_size)
|
||||
else:
|
||||
self.key = nn.Linear(config.hidden_size, self.all_head_size)
|
||||
self.value = nn.Linear(config.hidden_size, self.all_head_size)
|
||||
|
||||
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
|
||||
self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
|
||||
if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
|
||||
self.max_position_embeddings = config.max_position_embeddings
|
||||
self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
|
||||
self.save_attention = False
|
||||
|
||||
def save_attn_gradients(self, attn_gradients):
|
||||
self.attn_gradients = attn_gradients
|
||||
|
||||
def get_attn_gradients(self):
|
||||
return self.attn_gradients
|
||||
|
||||
def save_attention_map(self, attention_map):
|
||||
self.attention_map = attention_map
|
||||
|
||||
def get_attention_map(self):
|
||||
return self.attention_map
|
||||
|
||||
def transpose_for_scores(self, x):
|
||||
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
|
||||
x = x.view(*new_x_shape)
|
||||
return x.permute(0, 2, 1, 3)
|
||||
|
||||
def forward(
|
||||
self,
|
||||
hidden_states,
|
||||
attention_mask=None,
|
||||
head_mask=None,
|
||||
encoder_hidden_states=None,
|
||||
encoder_attention_mask=None,
|
||||
past_key_value=None,
|
||||
output_attentions=False,
|
||||
):
|
||||
mixed_query_layer = self.query(hidden_states)
|
||||
|
||||
# If this is instantiated as a cross-attention module, the keys
|
||||
# and values come from an encoder; the attention mask needs to be
|
||||
# such that the encoder's padding tokens are not attended to.
|
||||
is_cross_attention = encoder_hidden_states is not None
|
||||
|
||||
if is_cross_attention:
|
||||
key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
|
||||
value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
|
||||
attention_mask = encoder_attention_mask
|
||||
elif past_key_value is not None:
|
||||
key_layer = self.transpose_for_scores(self.key(hidden_states))
|
||||
value_layer = self.transpose_for_scores(self.value(hidden_states))
|
||||
key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
|
||||
value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
|
||||
else:
|
||||
key_layer = self.transpose_for_scores(self.key(hidden_states))
|
||||
value_layer = self.transpose_for_scores(self.value(hidden_states))
|
||||
|
||||
query_layer = self.transpose_for_scores(mixed_query_layer)
|
||||
|
||||
past_key_value = (key_layer, value_layer)
|
||||
|
||||
# Take the dot product between "query" and "key" to get the raw attention scores.
|
||||
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
|
||||
|
||||
if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
|
||||
seq_length = hidden_states.size()[1]
|
||||
position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
|
||||
position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
|
||||
distance = position_ids_l - position_ids_r
|
||||
positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
|
||||
positional_embedding = positional_embedding.to(dtype=query_layer.dtype) # fp16 compatibility
|
||||
|
||||
if self.position_embedding_type == "relative_key":
|
||||
relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
|
||||
attention_scores = attention_scores + relative_position_scores
|
||||
elif self.position_embedding_type == "relative_key_query":
|
||||
relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
|
||||
relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
|
||||
attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
|
||||
|
||||
attention_scores = attention_scores / math.sqrt(self.attention_head_size)
|
||||
if attention_mask is not None:
|
||||
# Apply the attention mask is (precomputed for all layers in BertModel forward() function)
|
||||
attention_scores = attention_scores + attention_mask
|
||||
|
||||
# Normalize the attention scores to probabilities.
|
||||
attention_probs = nn.Softmax(dim=-1)(attention_scores)
|
||||
|
||||
if is_cross_attention and self.save_attention:
|
||||
self.save_attention_map(attention_probs)
|
||||
attention_probs.register_hook(self.save_attn_gradients)
|
||||
|
||||
# This is actually dropping out entire tokens to attend to, which might
|
||||
# seem a bit unusual, but is taken from the original Transformer paper.
|
||||
attention_probs_dropped = self.dropout(attention_probs)
|
||||
|
||||
# Mask heads if we want to
|
||||
if head_mask is not None:
|
||||
attention_probs_dropped = attention_probs_dropped * head_mask
|
||||
|
||||
context_layer = torch.matmul(attention_probs_dropped, value_layer)
|
||||
|
||||
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
|
||||
new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
|
||||
context_layer = context_layer.view(*new_context_layer_shape)
|
||||
|
||||
outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
|
||||
|
||||
outputs = outputs + (past_key_value,)
|
||||
return outputs
|
||||
|
||||
|
||||
class BertSelfOutput(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
||||
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
||||
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
||||
|
||||
def forward(self, hidden_states, input_tensor):
|
||||
hidden_states = self.dense(hidden_states)
|
||||
hidden_states = self.dropout(hidden_states)
|
||||
hidden_states = self.LayerNorm(hidden_states + input_tensor)
|
||||
return hidden_states
|
||||
|
||||
|
||||
class BertAttention(nn.Module):
|
||||
def __init__(self, config, is_cross_attention=False):
|
||||
super().__init__()
|
||||
self.self = BertSelfAttention(config, is_cross_attention)
|
||||
self.output = BertSelfOutput(config)
|
||||
self.pruned_heads = set()
|
||||
|
||||
def prune_heads(self, heads):
|
||||
if len(heads) == 0:
|
||||
return
|
||||
heads, index = find_pruneable_heads_and_indices(
|
||||
heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
|
||||
)
|
||||
|
||||
# Prune linear layers
|
||||
self.self.query = prune_linear_layer(self.self.query, index)
|
||||
self.self.key = prune_linear_layer(self.self.key, index)
|
||||
self.self.value = prune_linear_layer(self.self.value, index)
|
||||
self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
|
||||
|
||||
# Update hyper params and store pruned heads
|
||||
self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
|
||||
self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
|
||||
self.pruned_heads = self.pruned_heads.union(heads)
|
||||
|
||||
def forward(
|
||||
self,
|
||||
hidden_states,
|
||||
attention_mask=None,
|
||||
head_mask=None,
|
||||
encoder_hidden_states=None,
|
||||
encoder_attention_mask=None,
|
||||
past_key_value=None,
|
||||
output_attentions=False,
|
||||
):
|
||||
self_outputs = self.self(
|
||||
hidden_states,
|
||||
attention_mask,
|
||||
head_mask,
|
||||
encoder_hidden_states,
|
||||
encoder_attention_mask,
|
||||
past_key_value,
|
||||
output_attentions,
|
||||
)
|
||||
attention_output = self.output(self_outputs[0], hidden_states)
|
||||
outputs = (attention_output,) + self_outputs[1:] # add attentions if we output them
|
||||
return outputs
|
||||
|
||||
|
||||
class BertIntermediate(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
|
||||
if isinstance(config.hidden_act, str):
|
||||
self.intermediate_act_fn = ACT2FN[config.hidden_act]
|
||||
else:
|
||||
self.intermediate_act_fn = config.hidden_act
|
||||
|
||||
def forward(self, hidden_states):
|
||||
hidden_states = self.dense(hidden_states)
|
||||
hidden_states = self.intermediate_act_fn(hidden_states)
|
||||
return hidden_states
|
||||
|
||||
|
||||
class BertOutput(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
|
||||
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
||||
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
||||
|
||||
def forward(self, hidden_states, input_tensor):
|
||||
hidden_states = self.dense(hidden_states)
|
||||
hidden_states = self.dropout(hidden_states)
|
||||
hidden_states = self.LayerNorm(hidden_states + input_tensor)
|
||||
return hidden_states
|
||||
|
||||
|
||||
class BertLayer(nn.Module):
|
||||
def __init__(self, config, layer_num):
|
||||
super().__init__()
|
||||
self.config = config
|
||||
self.chunk_size_feed_forward = config.chunk_size_feed_forward
|
||||
self.seq_len_dim = 1
|
||||
self.attention = BertAttention(config)
|
||||
self.layer_num = layer_num
|
||||
if self.config.add_cross_attention:
|
||||
self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention)
|
||||
self.intermediate = BertIntermediate(config)
|
||||
self.output = BertOutput(config)
|
||||
|
||||
def forward(
|
||||
self,
|
||||
hidden_states,
|
||||
attention_mask=None,
|
||||
head_mask=None,
|
||||
encoder_hidden_states=None,
|
||||
encoder_attention_mask=None,
|
||||
past_key_value=None,
|
||||
output_attentions=False,
|
||||
mode=None,
|
||||
):
|
||||
# decoder uni-directional self-attention cached key/values tuple is at positions 1,2
|
||||
self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
|
||||
self_attention_outputs = self.attention(
|
||||
hidden_states,
|
||||
attention_mask,
|
||||
head_mask,
|
||||
output_attentions=output_attentions,
|
||||
past_key_value=self_attn_past_key_value,
|
||||
)
|
||||
attention_output = self_attention_outputs[0]
|
||||
|
||||
outputs = self_attention_outputs[1:-1]
|
||||
present_key_value = self_attention_outputs[-1]
|
||||
|
||||
if mode=='multimodal':
|
||||
assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
|
||||
|
||||
cross_attention_outputs = self.crossattention(
|
||||
attention_output,
|
||||
attention_mask,
|
||||
head_mask,
|
||||
encoder_hidden_states,
|
||||
encoder_attention_mask,
|
||||
output_attentions=output_attentions,
|
||||
)
|
||||
attention_output = cross_attention_outputs[0]
|
||||
outputs = outputs + cross_attention_outputs[1:-1] # add cross attentions if we output attention weights
|
||||
layer_output = apply_chunking_to_forward(
|
||||
self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
|
||||
)
|
||||
outputs = (layer_output,) + outputs
|
||||
|
||||
outputs = outputs + (present_key_value,)
|
||||
|
||||
return outputs
|
||||
|
||||
def feed_forward_chunk(self, attention_output):
|
||||
intermediate_output = self.intermediate(attention_output)
|
||||
layer_output = self.output(intermediate_output, attention_output)
|
||||
return layer_output
|
||||
|
||||
|
||||
class BertEncoder(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.config = config
|
||||
self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
|
||||
self.gradient_checkpointing = False
|
||||
|
||||
def forward(
|
||||
self,
|
||||
hidden_states,
|
||||
attention_mask=None,
|
||||
head_mask=None,
|
||||
encoder_hidden_states=None,
|
||||
encoder_attention_mask=None,
|
||||
past_key_values=None,
|
||||
use_cache=None,
|
||||
output_attentions=False,
|
||||
output_hidden_states=False,
|
||||
return_dict=True,
|
||||
mode='multimodal',
|
||||
):
|
||||
all_hidden_states = () if output_hidden_states else None
|
||||
all_self_attentions = () if output_attentions else None
|
||||
all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
|
||||
|
||||
next_decoder_cache = () if use_cache else None
|
||||
|
||||
for i in range(self.config.num_hidden_layers):
|
||||
layer_module = self.layer[i]
|
||||
if output_hidden_states:
|
||||
all_hidden_states = all_hidden_states + (hidden_states,)
|
||||
|
||||
layer_head_mask = head_mask[i] if head_mask is not None else None
|
||||
past_key_value = past_key_values[i] if past_key_values is not None else None
|
||||
|
||||
if self.gradient_checkpointing and self.training:
|
||||
|
||||
if use_cache:
|
||||
logger.warn(
|
||||
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
|
||||
)
|
||||
use_cache = False
|
||||
|
||||
def create_custom_forward(module):
|
||||
def custom_forward(*inputs):
|
||||
return module(*inputs, past_key_value, output_attentions)
|
||||
|
||||
return custom_forward
|
||||
|
||||
layer_outputs = torch.utils.checkpoint.checkpoint(
|
||||
create_custom_forward(layer_module),
|
||||
hidden_states,
|
||||
attention_mask,
|
||||
layer_head_mask,
|
||||
encoder_hidden_states,
|
||||
encoder_attention_mask,
|
||||
mode=mode,
|
||||
)
|
||||
else:
|
||||
layer_outputs = layer_module(
|
||||
hidden_states,
|
||||
attention_mask,
|
||||
layer_head_mask,
|
||||
encoder_hidden_states,
|
||||
encoder_attention_mask,
|
||||
past_key_value,
|
||||
output_attentions,
|
||||
mode=mode,
|
||||
)
|
||||
|
||||
hidden_states = layer_outputs[0]
|
||||
if use_cache:
|
||||
next_decoder_cache += (layer_outputs[-1],)
|
||||
if output_attentions:
|
||||
all_self_attentions = all_self_attentions + (layer_outputs[1],)
|
||||
|
||||
if output_hidden_states:
|
||||
all_hidden_states = all_hidden_states + (hidden_states,)
|
||||
|
||||
if not return_dict:
|
||||
return tuple(
|
||||
v
|
||||
for v in [
|
||||
hidden_states,
|
||||
next_decoder_cache,
|
||||
all_hidden_states,
|
||||
all_self_attentions,
|
||||
all_cross_attentions,
|
||||
]
|
||||
if v is not None
|
||||
)
|
||||
return BaseModelOutputWithPastAndCrossAttentions(
|
||||
last_hidden_state=hidden_states,
|
||||
past_key_values=next_decoder_cache,
|
||||
hidden_states=all_hidden_states,
|
||||
attentions=all_self_attentions,
|
||||
cross_attentions=all_cross_attentions,
|
||||
)
|
||||
|
||||
|
||||
class BertPooler(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
||||
self.activation = nn.Tanh()
|
||||
|
||||
def forward(self, hidden_states):
|
||||
# We "pool" the model by simply taking the hidden state corresponding
|
||||
# to the first token.
|
||||
first_token_tensor = hidden_states[:, 0]
|
||||
pooled_output = self.dense(first_token_tensor)
|
||||
pooled_output = self.activation(pooled_output)
|
||||
return pooled_output
|
||||
|
||||
|
||||
class BertPredictionHeadTransform(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
||||
if isinstance(config.hidden_act, str):
|
||||
self.transform_act_fn = ACT2FN[config.hidden_act]
|
||||
else:
|
||||
self.transform_act_fn = config.hidden_act
|
||||
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
||||
|
||||
def forward(self, hidden_states):
|
||||
hidden_states = self.dense(hidden_states)
|
||||
hidden_states = self.transform_act_fn(hidden_states)
|
||||
hidden_states = self.LayerNorm(hidden_states)
|
||||
return hidden_states
|
||||
|
||||
|
||||
class BertLMPredictionHead(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.transform = BertPredictionHeadTransform(config)
|
||||
|
||||
# The output weights are the same as the input embeddings, but there is
|
||||
# an output-only bias for each token.
|
||||
self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
|
||||
|
||||
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
|
||||
|
||||
# Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
|
||||
self.decoder.bias = self.bias
|
||||
|
||||
def forward(self, hidden_states):
|
||||
hidden_states = self.transform(hidden_states)
|
||||
hidden_states = self.decoder(hidden_states)
|
||||
return hidden_states
|
||||
|
||||
|
||||
class BertOnlyMLMHead(nn.Module):
|
||||
def __init__(self, config):
|
||||
super().__init__()
|
||||
self.predictions = BertLMPredictionHead(config)
|
||||
|
||||
def forward(self, sequence_output):
|
||||
prediction_scores = self.predictions(sequence_output)
|
||||
return prediction_scores
|
||||
|
||||
|
||||
class BertPreTrainedModel(PreTrainedModel):
|
||||
"""
|
||||
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
|
||||
models.
|
||||
"""
|
||||
|
||||
config_class = BertConfig
|
||||
base_model_prefix = "bert"
|
||||
_keys_to_ignore_on_load_missing = [r"position_ids"]
|
||||
|
||||
def _init_weights(self, module):
|
||||
""" Initialize the weights """
|
||||
if isinstance(module, (nn.Linear, nn.Embedding)):
|
||||
# Slightly different from the TF version which uses truncated_normal for initialization
|
||||
# cf https://github.com/pytorch/pytorch/pull/5617
|
||||
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
|
||||
elif isinstance(module, nn.LayerNorm):
|
||||
module.bias.data.zero_()
|
||||
module.weight.data.fill_(1.0)
|
||||
if isinstance(module, nn.Linear) and module.bias is not None:
|
||||
module.bias.data.zero_()
|
||||
|
||||
|
||||
class BertModel(BertPreTrainedModel):
|
||||
"""
|
||||
The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
|
||||
cross-attention is added between the self-attention layers, following the architecture described in `Attention is
|
||||
all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
|
||||
Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
|
||||
argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
|
||||
input to the forward pass.
|
||||
"""
|
||||
|
||||
def __init__(self, config, add_pooling_layer=True):
|
||||
super().__init__(config)
|
||||
self.config = config
|
||||
|
||||
self.embeddings = BertEmbeddings(config)
|
||||
|
||||
self.encoder = BertEncoder(config)
|
||||
|
||||
self.pooler = BertPooler(config) if add_pooling_layer else None
|
||||
|
||||
self.init_weights()
|
||||
|
||||
|
||||
def get_input_embeddings(self):
|
||||
return self.embeddings.word_embeddings
|
||||
|
||||
def set_input_embeddings(self, value):
|
||||
self.embeddings.word_embeddings = value
|
||||
|
||||
def _prune_heads(self, heads_to_prune):
|
||||
"""
|
||||
Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
|
||||
class PreTrainedModel
|
||||
"""
|
||||
for layer, heads in heads_to_prune.items():
|
||||
self.encoder.layer[layer].attention.prune_heads(heads)
|
||||
|
||||
|
||||
def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
|
||||
"""
|
||||
Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
|
||||
|
||||
Arguments:
|
||||
attention_mask (:obj:`torch.Tensor`):
|
||||
Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
|
||||
input_shape (:obj:`Tuple[int]`):
|
||||
The shape of the input to the model.
|
||||
device: (:obj:`torch.device`):
|
||||
The device of the input to the model.
|
||||
|
||||
Returns:
|
||||
:obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
|
||||
"""
|
||||
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
|
||||
# ourselves in which case we just need to make it broadcastable to all heads.
|
||||
if attention_mask.dim() == 3:
|
||||
extended_attention_mask = attention_mask[:, None, :, :]
|
||||
elif attention_mask.dim() == 2:
|
||||
# Provided a padding mask of dimensions [batch_size, seq_length]
|
||||
# - if the model is a decoder, apply a causal mask in addition to the padding mask
|
||||
# - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
|
||||
if is_decoder:
|
||||
batch_size, seq_length = input_shape
|
||||
|
||||
seq_ids = torch.arange(seq_length, device=device)
|
||||
causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
|
||||
# in case past_key_values are used we need to add a prefix ones mask to the causal mask
|
||||
# causal and attention masks must have same type with pytorch version < 1.3
|
||||
causal_mask = causal_mask.to(attention_mask.dtype)
|
||||
|
||||
if causal_mask.shape[1] < attention_mask.shape[1]:
|
||||
prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
|
||||
causal_mask = torch.cat(
|
||||
[
|
||||
torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
|
||||
causal_mask,
|
||||
],
|
||||
axis=-1,
|
||||
)
|
||||
|
||||
extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
|
||||
else:
|
||||
extended_attention_mask = attention_mask[:, None, None, :]
|
||||
else:
|
||||
raise ValueError(
|
||||
"Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
|
||||
input_shape, attention_mask.shape
|
||||
)
|
||||
)
|
||||
|
||||
# Since attention_mask is 1.0 for positions we want to attend and 0.0 for
|
||||
# masked positions, this operation will create a tensor which is 0.0 for
|
||||
# positions we want to attend and -10000.0 for masked positions.
|
||||
# Since we are adding it to the raw scores before the softmax, this is
|
||||
# effectively the same as removing these entirely.
|
||||
extended_attention_mask = extended_attention_mask.to(dtype=self.dtype) # fp16 compatibility
|
||||
extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
|
||||
return extended_attention_mask
|
||||
|
||||
def forward(
|
||||
self,
|
||||
input_ids=None,
|
||||
attention_mask=None,
|
||||
position_ids=None,
|
||||
head_mask=None,
|
||||
inputs_embeds=None,
|
||||
encoder_embeds=None,
|
||||
encoder_hidden_states=None,
|
||||
encoder_attention_mask=None,
|
||||
past_key_values=None,
|
||||
use_cache=None,
|
||||
output_attentions=None,
|
||||
output_hidden_states=None,
|
||||
return_dict=None,
|
||||
is_decoder=False,
|
||||
mode='multimodal',
|
||||
):
|
||||
r"""
|
||||
encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
|
||||
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
|
||||
the model is configured as a decoder.
|
||||
encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
|
||||
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
|
||||
the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
|
||||
- 1 for tokens that are **not masked**,
|
||||
- 0 for tokens that are **masked**.
|
||||
past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
|
||||
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
|
||||
If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
|
||||
(those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
|
||||
instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
|
||||
use_cache (:obj:`bool`, `optional`):
|
||||
If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
|
||||
decoding (see :obj:`past_key_values`).
|
||||
"""
|
||||
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
||||
output_hidden_states = (
|
||||
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
||||
)
|
||||
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
||||
|
||||
if is_decoder:
|
||||
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
||||
else:
|
||||
use_cache = False
|
||||
|
||||
if input_ids is not None and inputs_embeds is not None:
|
||||
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
|
||||
elif input_ids is not None:
|
||||
input_shape = input_ids.size()
|
||||
batch_size, seq_length = input_shape
|
||||
device = input_ids.device
|
||||
elif inputs_embeds is not None:
|
||||
input_shape = inputs_embeds.size()[:-1]
|
||||
batch_size, seq_length = input_shape
|
||||
device = inputs_embeds.device
|
||||
elif encoder_embeds is not None:
|
||||
input_shape = encoder_embeds.size()[:-1]
|
||||
batch_size, seq_length = input_shape
|
||||
device = encoder_embeds.device
|
||||
else:
|
||||
raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
|
||||
|
||||
# past_key_values_length
|
||||
past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
|
||||
|
||||
if attention_mask is None:
|
||||
attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
|
||||
|
||||
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
|
||||
# ourselves in which case we just need to make it broadcastable to all heads.
|
||||
extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape,
|
||||
device, is_decoder)
|
||||
|
||||
# If a 2D or 3D attention mask is provided for the cross-attention
|
||||
# we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
|
||||
if encoder_hidden_states is not None:
|
||||
if type(encoder_hidden_states) == list:
|
||||
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
|
||||
else:
|
||||
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
|
||||
encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
|
||||
|
||||
if type(encoder_attention_mask) == list:
|
||||
encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
|
||||
elif encoder_attention_mask is None:
|
||||
encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
|
||||
encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
|
||||
else:
|
||||
encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
|
||||
else:
|
||||
encoder_extended_attention_mask = None
|
||||
|
||||
# Prepare head mask if needed
|
||||
# 1.0 in head_mask indicate we keep the head
|
||||
# attention_probs has shape bsz x n_heads x N x N
|
||||
# input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
|
||||
# and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
|
||||
head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
|
||||
|
||||
if encoder_embeds is None:
|
||||
embedding_output = self.embeddings(
|
||||
input_ids=input_ids,
|
||||
position_ids=position_ids,
|
||||
inputs_embeds=inputs_embeds,
|
||||
past_key_values_length=past_key_values_length,
|
||||
)
|
||||
else:
|
||||
embedding_output = encoder_embeds
|
||||
|
||||
encoder_outputs = self.encoder(
|
||||
embedding_output,
|
||||
attention_mask=extended_attention_mask,
|
||||
head_mask=head_mask,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
encoder_attention_mask=encoder_extended_attention_mask,
|
||||
past_key_values=past_key_values,
|
||||
use_cache=use_cache,
|
||||
output_attentions=output_attentions,
|
||||
output_hidden_states=output_hidden_states,
|
||||
return_dict=return_dict,
|
||||
mode=mode,
|
||||
)
|
||||
sequence_output = encoder_outputs[0]
|
||||
pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
|
||||
|
||||
if not return_dict:
|
||||
return (sequence_output, pooled_output) + encoder_outputs[1:]
|
||||
|
||||
return BaseModelOutputWithPoolingAndCrossAttentions(
|
||||
last_hidden_state=sequence_output,
|
||||
pooler_output=pooled_output,
|
||||
past_key_values=encoder_outputs.past_key_values,
|
||||
hidden_states=encoder_outputs.hidden_states,
|
||||
attentions=encoder_outputs.attentions,
|
||||
cross_attentions=encoder_outputs.cross_attentions,
|
||||
)
|
||||
|
||||
|
||||
|
||||
class BertLMHeadModel(BertPreTrainedModel):
|
||||
|
||||
_keys_to_ignore_on_load_unexpected = [r"pooler"]
|
||||
_keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
|
||||
|
||||
def __init__(self, config):
|
||||
super().__init__(config)
|
||||
|
||||
self.bert = BertModel(config, add_pooling_layer=False)
|
||||
self.cls = BertOnlyMLMHead(config)
|
||||
|
||||
self.init_weights()
|
||||
|
||||
def get_output_embeddings(self):
|
||||
return self.cls.predictions.decoder
|
||||
|
||||
def set_output_embeddings(self, new_embeddings):
|
||||
self.cls.predictions.decoder = new_embeddings
|
||||
|
||||
def forward(
|
||||
self,
|
||||
input_ids=None,
|
||||
attention_mask=None,
|
||||
position_ids=None,
|
||||
head_mask=None,
|
||||
inputs_embeds=None,
|
||||
encoder_hidden_states=None,
|
||||
encoder_attention_mask=None,
|
||||
labels=None,
|
||||
past_key_values=None,
|
||||
use_cache=None,
|
||||
output_attentions=None,
|
||||
output_hidden_states=None,
|
||||
return_dict=None,
|
||||
return_logits=False,
|
||||
is_decoder=True,
|
||||
reduction='mean',
|
||||
mode='multimodal',
|
||||
):
|
||||
r"""
|
||||
encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
|
||||
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
|
||||
the model is configured as a decoder.
|
||||
encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
|
||||
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
|
||||
the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
|
||||
- 1 for tokens that are **not masked**,
|
||||
- 0 for tokens that are **masked**.
|
||||
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
|
||||
Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
|
||||
``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are
|
||||
ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``
|
||||
past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
|
||||
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
|
||||
If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
|
||||
(those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
|
||||
instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
|
||||
use_cache (:obj:`bool`, `optional`):
|
||||
If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
|
||||
decoding (see :obj:`past_key_values`).
|
||||
Returns:
|
||||
Example::
|
||||
>>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig
|
||||
>>> import torch
|
||||
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
|
||||
>>> config = BertConfig.from_pretrained("bert-base-cased")
|
||||
>>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)
|
||||
>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
|
||||
>>> outputs = model(**inputs)
|
||||
>>> prediction_logits = outputs.logits
|
||||
"""
|
||||
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
||||
if labels is not None:
|
||||
use_cache = False
|
||||
|
||||
outputs = self.bert(
|
||||
input_ids,
|
||||
attention_mask=attention_mask,
|
||||
position_ids=position_ids,
|
||||
head_mask=head_mask,
|
||||
inputs_embeds=inputs_embeds,
|
||||
encoder_hidden_states=encoder_hidden_states,
|
||||
encoder_attention_mask=encoder_attention_mask,
|
||||
past_key_values=past_key_values,
|
||||
use_cache=use_cache,
|
||||
output_attentions=output_attentions,
|
||||
output_hidden_states=output_hidden_states,
|
||||
return_dict=return_dict,
|
||||
is_decoder=is_decoder,
|
||||
mode=mode,
|
||||
)
|
||||
|
||||
sequence_output = outputs[0]
|
||||
prediction_scores = self.cls(sequence_output)
|
||||
|
||||
if return_logits:
|
||||
return prediction_scores[:, :-1, :].contiguous()
|
||||
|
||||
lm_loss = None
|
||||
if labels is not None:
|
||||
# we are doing next-token prediction; shift prediction scores and input ids by one
|
||||
shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
|
||||
labels = labels[:, 1:].contiguous()
|
||||
loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1)
|
||||
lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
|
||||
if reduction=='none':
|
||||
lm_loss = lm_loss.view(prediction_scores.size(0),-1).sum(1)
|
||||
|
||||
if not return_dict:
|
||||
output = (prediction_scores,) + outputs[2:]
|
||||
return ((lm_loss,) + output) if lm_loss is not None else output
|
||||
|
||||
return CausalLMOutputWithCrossAttentions(
|
||||
loss=lm_loss,
|
||||
logits=prediction_scores,
|
||||
past_key_values=outputs.past_key_values,
|
||||
hidden_states=outputs.hidden_states,
|
||||
attentions=outputs.attentions,
|
||||
cross_attentions=outputs.cross_attentions,
|
||||
)
|
||||
|
||||
def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, **model_kwargs):
|
||||
input_shape = input_ids.shape
|
||||
# if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
|
||||
if attention_mask is None:
|
||||
attention_mask = input_ids.new_ones(input_shape)
|
||||
|
||||
# cut decoder_input_ids if past is used
|
||||
if past is not None:
|
||||
input_ids = input_ids[:, -1:]
|
||||
|
||||
return {
|
||||
"input_ids": input_ids,
|
||||
"attention_mask": attention_mask,
|
||||
"past_key_values": past,
|
||||
"encoder_hidden_states": model_kwargs.get("encoder_hidden_states", None),
|
||||
"encoder_attention_mask": model_kwargs.get("encoder_attention_mask", None),
|
||||
"is_decoder": True,
|
||||
}
|
||||
|
||||
def _reorder_cache(self, past, beam_idx):
|
||||
reordered_past = ()
|
||||
for layer_past in past:
|
||||
reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
|
||||
return reordered_past
|
||||
-423
@@ -1,423 +0,0 @@
|
||||
'''
|
||||
* Copyright (c) 2022, salesforce.com, inc.
|
||||
* All rights reserved.
|
||||
* SPDX-License-Identifier: BSD-3-Clause
|
||||
* For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
|
||||
* By Junnan Li
|
||||
'''
|
||||
import warnings
|
||||
warnings.filterwarnings("ignore")
|
||||
|
||||
from .blip_vit import VisionTransformer, interpolate_pos_embed
|
||||
from .blip_med import BertConfig, BertModel, BertLMHeadModel
|
||||
from transformers import BertTokenizer
|
||||
|
||||
import torch
|
||||
from torch import nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
import os
|
||||
from urllib.parse import urlparse
|
||||
from timm.models.hub import download_cached_file
|
||||
import numpy as np
|
||||
|
||||
from pathlib import Path
|
||||
LOCAL_PATH = os.path.dirname(os.path.abspath(__file__))
|
||||
|
||||
# BLIP
|
||||
|
||||
class BLIP_Base(nn.Module):
|
||||
def __init__(self,
|
||||
med_config = Path(LOCAL_PATH, 'blip_configs/med_config.json'),
|
||||
image_size = 224,
|
||||
vit = 'base',
|
||||
vit_grad_ckpt = False,
|
||||
vit_ckpt_layer = 0,
|
||||
):
|
||||
"""
|
||||
Args:
|
||||
med_config (str): path for the mixture of encoder-decoder model's configuration file
|
||||
image_size (int): input image size
|
||||
vit (str): model size of vision transformer
|
||||
"""
|
||||
super().__init__()
|
||||
|
||||
self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
|
||||
self.tokenizer = init_tokenizer()
|
||||
med_config = BertConfig.from_json_file(med_config)
|
||||
med_config.encoder_width = vision_width
|
||||
self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)
|
||||
|
||||
|
||||
def forward(self, image, caption, mode):
|
||||
|
||||
assert mode in ['image', 'text', 'multimodal'], "mode parameter must be image, text, or multimodal"
|
||||
text = self.tokenizer(caption, return_tensors="pt").to(image.device)
|
||||
|
||||
if mode=='image':
|
||||
# return image features
|
||||
image_embeds = self.visual_encoder(image)
|
||||
return image_embeds
|
||||
|
||||
elif mode=='text':
|
||||
# return text features
|
||||
text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,
|
||||
return_dict = True, mode = 'text')
|
||||
return text_output.last_hidden_state
|
||||
|
||||
elif mode=='multimodal':
|
||||
# return multimodel features
|
||||
image_embeds = self.visual_encoder(image)
|
||||
image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
|
||||
|
||||
text.input_ids[:,0] = self.tokenizer.enc_token_id
|
||||
output = self.text_encoder(text.input_ids,
|
||||
attention_mask = text.attention_mask,
|
||||
encoder_hidden_states = image_embeds,
|
||||
encoder_attention_mask = image_atts,
|
||||
return_dict = True,
|
||||
)
|
||||
return output.last_hidden_state
|
||||
|
||||
|
||||
|
||||
class BLIP_Decoder(nn.Module):
|
||||
def __init__(self,
|
||||
med_config = Path(LOCAL_PATH, 'blip_configs/med_config.json'),
|
||||
image_size = 384,
|
||||
vit = 'base',
|
||||
vit_grad_ckpt = False,
|
||||
vit_ckpt_layer = 0,
|
||||
prompt = 'a picture of ',
|
||||
):
|
||||
"""
|
||||
Args:
|
||||
med_config (str): path for the mixture of encoder-decoder model's configuration file
|
||||
image_size (int): input image size
|
||||
vit (str): model size of vision transformer
|
||||
"""
|
||||
super().__init__()
|
||||
|
||||
self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
|
||||
self.tokenizer = init_tokenizer()
|
||||
med_config = BertConfig.from_json_file(med_config)
|
||||
med_config.encoder_width = vision_width
|
||||
self.text_decoder = BertLMHeadModel(config=med_config)
|
||||
|
||||
self.prompt = prompt
|
||||
self.prompt_length = len(self.tokenizer(self.prompt).input_ids)-1
|
||||
|
||||
|
||||
def forward(self, image, caption):
|
||||
|
||||
image_embeds = self.visual_encoder(image)
|
||||
image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
|
||||
|
||||
text = self.tokenizer(caption, padding='longest', truncation=True, max_length=40, return_tensors="pt").to(image.device)
|
||||
|
||||
text.input_ids[:,0] = self.tokenizer.bos_token_id
|
||||
|
||||
decoder_targets = text.input_ids.masked_fill(text.input_ids == self.tokenizer.pad_token_id, -100)
|
||||
decoder_targets[:,:self.prompt_length] = -100
|
||||
|
||||
decoder_output = self.text_decoder(text.input_ids,
|
||||
attention_mask = text.attention_mask,
|
||||
encoder_hidden_states = image_embeds,
|
||||
encoder_attention_mask = image_atts,
|
||||
labels = decoder_targets,
|
||||
return_dict = True,
|
||||
)
|
||||
loss_lm = decoder_output.loss
|
||||
|
||||
return loss_lm
|
||||
|
||||
def generate(self, image, sample=False, num_beams=3, max_length=30, min_length=10, top_p=0.9, repetition_penalty=1.0):
|
||||
image_embeds = self.visual_encoder(image)
|
||||
|
||||
if not sample:
|
||||
image_embeds = image_embeds.repeat_interleave(num_beams,dim=0)
|
||||
|
||||
image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
|
||||
model_kwargs = {"encoder_hidden_states": image_embeds, "encoder_attention_mask":image_atts}
|
||||
|
||||
prompt = [self.prompt] * image.size(0)
|
||||
input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(image.device)
|
||||
input_ids[:,0] = self.tokenizer.bos_token_id
|
||||
input_ids = input_ids[:, :-1]
|
||||
|
||||
if sample:
|
||||
#nucleus sampling
|
||||
outputs = self.text_decoder.generate(input_ids=input_ids,
|
||||
max_length=max_length,
|
||||
min_length=min_length,
|
||||
do_sample=True,
|
||||
top_p=top_p,
|
||||
num_return_sequences=1,
|
||||
eos_token_id=self.tokenizer.sep_token_id,
|
||||
pad_token_id=self.tokenizer.pad_token_id,
|
||||
repetition_penalty=1.1,
|
||||
**model_kwargs)
|
||||
else:
|
||||
#beam search
|
||||
outputs = self.text_decoder.generate(input_ids=input_ids,
|
||||
max_length=max_length,
|
||||
min_length=min_length,
|
||||
num_beams=num_beams,
|
||||
eos_token_id=self.tokenizer.sep_token_id,
|
||||
pad_token_id=self.tokenizer.pad_token_id,
|
||||
repetition_penalty=repetition_penalty,
|
||||
**model_kwargs)
|
||||
|
||||
captions = []
|
||||
for output in outputs:
|
||||
caption = self.tokenizer.decode(output, skip_special_tokens=True)
|
||||
captions.append(caption[len(self.prompt):])
|
||||
return captions
|
||||
|
||||
|
||||
def blip_decoder(pretrained='',**kwargs):
|
||||
model = BLIP_Decoder(**kwargs)
|
||||
if pretrained:
|
||||
model,msg = load_checkpoint(model,pretrained)
|
||||
assert(len(msg.missing_keys)==0)
|
||||
return model
|
||||
|
||||
def blip_feature_extractor(pretrained='',**kwargs):
|
||||
model = BLIP_Base(**kwargs)
|
||||
if pretrained:
|
||||
model,msg = load_checkpoint(model,pretrained)
|
||||
assert(len(msg.missing_keys)==0)
|
||||
return model
|
||||
|
||||
def init_tokenizer():
|
||||
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
|
||||
tokenizer.add_special_tokens({'bos_token':'[DEC]'})
|
||||
tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']})
|
||||
tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]
|
||||
return tokenizer
|
||||
|
||||
|
||||
def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0):
|
||||
|
||||
assert vit in ['base', 'large'], "vit parameter must be base or large"
|
||||
if vit=='base':
|
||||
vision_width = 768
|
||||
visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12,
|
||||
num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
|
||||
drop_path_rate=0 or drop_path_rate
|
||||
)
|
||||
elif vit=='large':
|
||||
vision_width = 1024
|
||||
visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24,
|
||||
num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
|
||||
drop_path_rate=0.1 or drop_path_rate
|
||||
)
|
||||
return visual_encoder, vision_width
|
||||
|
||||
def is_url(url_or_filename):
|
||||
parsed = urlparse(url_or_filename)
|
||||
return parsed.scheme in ("http", "https")
|
||||
|
||||
def load_checkpoint(model,url_or_filename):
|
||||
if is_url(url_or_filename):
|
||||
cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
|
||||
checkpoint = torch.load(cached_file, map_location='cpu')
|
||||
elif os.path.isfile(url_or_filename):
|
||||
checkpoint = torch.load(url_or_filename, map_location='cpu')
|
||||
else:
|
||||
raise RuntimeError('checkpoint url or path is invalid')
|
||||
|
||||
state_dict = checkpoint['model']
|
||||
|
||||
state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder)
|
||||
if 'visual_encoder_m.pos_embed' in model.state_dict().keys():
|
||||
state_dict['visual_encoder_m.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder_m.pos_embed'],
|
||||
model.visual_encoder_m)
|
||||
for key in model.state_dict().keys():
|
||||
if key in state_dict.keys():
|
||||
if state_dict[key].shape!=model.state_dict()[key].shape:
|
||||
del state_dict[key]
|
||||
|
||||
msg = model.load_state_dict(state_dict,strict=False)
|
||||
print('load checkpoint from %s'%url_or_filename)
|
||||
return model,msg
|
||||
|
||||
# BLIP VQA
|
||||
|
||||
class BLIP_VQA(nn.Module):
|
||||
def __init__(self,
|
||||
med_config = Path(LOCAL_PATH, 'blip_configs/med_config.json'),
|
||||
image_size = 480,
|
||||
vit = 'base',
|
||||
vit_grad_ckpt = False,
|
||||
vit_ckpt_layer = 0,
|
||||
):
|
||||
"""
|
||||
Args:
|
||||
med_config (str): path for the mixture of encoder-decoder model's configuration file
|
||||
image_size (int): input image size
|
||||
vit (str): model size of vision transformer
|
||||
"""
|
||||
super().__init__()
|
||||
|
||||
self.visual_encoder, vision_width = create_vit(vit, image_size, vit_grad_ckpt, vit_ckpt_layer, drop_path_rate=0.1)
|
||||
self.tokenizer = init_tokenizer()
|
||||
|
||||
encoder_config = BertConfig.from_json_file(med_config)
|
||||
encoder_config.encoder_width = vision_width
|
||||
self.text_encoder = BertModel(config=encoder_config, add_pooling_layer=False)
|
||||
|
||||
decoder_config = BertConfig.from_json_file(med_config)
|
||||
self.text_decoder = BertLMHeadModel(config=decoder_config)
|
||||
|
||||
|
||||
def forward(self, image, question, answer=None, n=None, weights=None, train=True, inference='rank', k_test=128):
|
||||
|
||||
image_embeds = self.visual_encoder(image)
|
||||
image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
|
||||
|
||||
question = self.tokenizer(question, padding='longest', truncation=True, max_length=35,
|
||||
return_tensors="pt").to(image.device)
|
||||
question.input_ids[:,0] = self.tokenizer.enc_token_id
|
||||
|
||||
if train:
|
||||
'''
|
||||
n: number of answers for each question
|
||||
weights: weight for each answer
|
||||
'''
|
||||
answer = self.tokenizer(answer, padding='longest', return_tensors="pt").to(image.device)
|
||||
answer.input_ids[:,0] = self.tokenizer.bos_token_id
|
||||
answer_targets = answer.input_ids.masked_fill(answer.input_ids == self.tokenizer.pad_token_id, -100)
|
||||
|
||||
question_output = self.text_encoder(question.input_ids,
|
||||
attention_mask = question.attention_mask,
|
||||
encoder_hidden_states = image_embeds,
|
||||
encoder_attention_mask = image_atts,
|
||||
return_dict = True)
|
||||
|
||||
question_states = []
|
||||
question_atts = []
|
||||
for b, n in enumerate(n):
|
||||
question_states += [question_output.last_hidden_state[b]]*n
|
||||
question_atts += [question.attention_mask[b]]*n
|
||||
question_states = torch.stack(question_states,0)
|
||||
question_atts = torch.stack(question_atts,0)
|
||||
|
||||
answer_output = self.text_decoder(answer.input_ids,
|
||||
attention_mask = answer.attention_mask,
|
||||
encoder_hidden_states = question_states,
|
||||
encoder_attention_mask = question_atts,
|
||||
labels = answer_targets,
|
||||
return_dict = True,
|
||||
reduction = 'none',
|
||||
)
|
||||
|
||||
loss = weights * answer_output.loss
|
||||
loss = loss.sum()/image.size(0)
|
||||
|
||||
return loss
|
||||
|
||||
|
||||
else:
|
||||
question_output = self.text_encoder(question.input_ids,
|
||||
attention_mask = question.attention_mask,
|
||||
encoder_hidden_states = image_embeds,
|
||||
encoder_attention_mask = image_atts,
|
||||
return_dict = True)
|
||||
|
||||
if inference=='generate':
|
||||
num_beams = 3
|
||||
question_states = question_output.last_hidden_state.repeat_interleave(num_beams,dim=0)
|
||||
question_atts = torch.ones(question_states.size()[:-1],dtype=torch.long).to(question_states.device)
|
||||
model_kwargs = {"encoder_hidden_states": question_states, "encoder_attention_mask":question_atts}
|
||||
|
||||
bos_ids = torch.full((image.size(0),1),fill_value=self.tokenizer.bos_token_id,device=image.device)
|
||||
|
||||
outputs = self.text_decoder.generate(input_ids=bos_ids,
|
||||
max_length=10,
|
||||
min_length=1,
|
||||
num_beams=num_beams,
|
||||
eos_token_id=self.tokenizer.sep_token_id,
|
||||
pad_token_id=self.tokenizer.pad_token_id,
|
||||
**model_kwargs)
|
||||
|
||||
answers = []
|
||||
for output in outputs:
|
||||
answer = self.tokenizer.decode(output, skip_special_tokens=True)
|
||||
answers.append(answer)
|
||||
return answers
|
||||
|
||||
elif inference=='rank':
|
||||
max_ids = self.rank_answer(question_output.last_hidden_state, question.attention_mask,
|
||||
answer.input_ids, answer.attention_mask, k_test)
|
||||
return max_ids
|
||||
|
||||
|
||||
|
||||
def rank_answer(self, question_states, question_atts, answer_ids, answer_atts, k):
|
||||
|
||||
num_ques = question_states.size(0)
|
||||
start_ids = answer_ids[0,0].repeat(num_ques,1) # bos token
|
||||
|
||||
start_output = self.text_decoder(start_ids,
|
||||
encoder_hidden_states = question_states,
|
||||
encoder_attention_mask = question_atts,
|
||||
return_dict = True,
|
||||
reduction = 'none')
|
||||
logits = start_output.logits[:,0,:] # first token's logit
|
||||
|
||||
# topk_probs: top-k probability
|
||||
# topk_ids: [num_question, k]
|
||||
answer_first_token = answer_ids[:,1]
|
||||
prob_first_token = F.softmax(logits,dim=1).index_select(dim=1, index=answer_first_token)
|
||||
topk_probs, topk_ids = prob_first_token.topk(k,dim=1)
|
||||
|
||||
# answer input: [num_question*k, answer_len]
|
||||
input_ids = []
|
||||
input_atts = []
|
||||
for b, topk_id in enumerate(topk_ids):
|
||||
input_ids.append(answer_ids.index_select(dim=0, index=topk_id))
|
||||
input_atts.append(answer_atts.index_select(dim=0, index=topk_id))
|
||||
input_ids = torch.cat(input_ids,dim=0)
|
||||
input_atts = torch.cat(input_atts,dim=0)
|
||||
|
||||
targets_ids = input_ids.masked_fill(input_ids == self.tokenizer.pad_token_id, -100)
|
||||
|
||||
# repeat encoder's output for top-k answers
|
||||
question_states = tile(question_states, 0, k)
|
||||
question_atts = tile(question_atts, 0, k)
|
||||
|
||||
output = self.text_decoder(input_ids,
|
||||
attention_mask = input_atts,
|
||||
encoder_hidden_states = question_states,
|
||||
encoder_attention_mask = question_atts,
|
||||
labels = targets_ids,
|
||||
return_dict = True,
|
||||
reduction = 'none')
|
||||
|
||||
log_probs_sum = -output.loss
|
||||
log_probs_sum = log_probs_sum.view(num_ques,k)
|
||||
|
||||
max_topk_ids = log_probs_sum.argmax(dim=1)
|
||||
max_ids = topk_ids[max_topk_ids>=0,max_topk_ids]
|
||||
|
||||
return max_ids
|
||||
|
||||
|
||||
def blip_vqa(pretrained='',**kwargs):
|
||||
model = BLIP_VQA(**kwargs)
|
||||
if pretrained:
|
||||
model,msg = load_checkpoint(model,pretrained)
|
||||
# assert(len(msg.missing_keys)==0)
|
||||
return model
|
||||
|
||||
|
||||
def tile(x, dim, n_tile):
|
||||
init_dim = x.size(dim)
|
||||
repeat_idx = [1] * x.dim()
|
||||
repeat_idx[dim] = n_tile
|
||||
x = x.repeat(*(repeat_idx))
|
||||
order_index = torch.LongTensor(np.concatenate([init_dim * np.arange(n_tile) + i for i in range(init_dim)]))
|
||||
return torch.index_select(x, dim, order_index.to(x.device))
|
||||
|
||||
|
||||
-12
@@ -1,12 +0,0 @@
|
||||
Copyright (c) 2022, Salesforce.com, Inc.
|
||||
All rights reserved.
|
||||
|
||||
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
|
||||
|
||||
* Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
|
||||
|
||||
* Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
|
||||
|
||||
* Neither the name of Salesforce.com nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
|
||||
|
||||
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
@@ -1,305 +0,0 @@
|
||||
'''
|
||||
* Copyright (c) 2022, salesforce.com, inc.
|
||||
* All rights reserved.
|
||||
* SPDX-License-Identifier: BSD-3-Clause
|
||||
* For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
|
||||
* By Junnan Li
|
||||
* Based on timm code base
|
||||
* https://github.com/rwightman/pytorch-image-models/tree/master/timm
|
||||
'''
|
||||
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.nn.functional as F
|
||||
from functools import partial
|
||||
|
||||
from timm.models.vision_transformer import _cfg, PatchEmbed
|
||||
from timm.models.registry import register_model
|
||||
from timm.models.layers import trunc_normal_, DropPath
|
||||
from timm.models.helpers import named_apply, adapt_input_conv
|
||||
|
||||
from fairscale.nn.checkpoint.checkpoint_activations import checkpoint_wrapper
|
||||
|
||||
class Mlp(nn.Module):
|
||||
""" MLP as used in Vision Transformer, MLP-Mixer and related networks
|
||||
"""
|
||||
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
|
||||
super().__init__()
|
||||
out_features = out_features or in_features
|
||||
hidden_features = hidden_features or in_features
|
||||
self.fc1 = nn.Linear(in_features, hidden_features)
|
||||
self.act = act_layer()
|
||||
self.fc2 = nn.Linear(hidden_features, out_features)
|
||||
self.drop = nn.Dropout(drop)
|
||||
|
||||
def forward(self, x):
|
||||
x = self.fc1(x)
|
||||
x = self.act(x)
|
||||
x = self.drop(x)
|
||||
x = self.fc2(x)
|
||||
x = self.drop(x)
|
||||
return x
|
||||
|
||||
|
||||
class Attention(nn.Module):
|
||||
def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
|
||||
super().__init__()
|
||||
self.num_heads = num_heads
|
||||
head_dim = dim // num_heads
|
||||
# NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
|
||||
self.scale = qk_scale or head_dim ** -0.5
|
||||
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
|
||||
self.attn_drop = nn.Dropout(attn_drop)
|
||||
self.proj = nn.Linear(dim, dim)
|
||||
self.proj_drop = nn.Dropout(proj_drop)
|
||||
self.attn_gradients = None
|
||||
self.attention_map = None
|
||||
|
||||
def save_attn_gradients(self, attn_gradients):
|
||||
self.attn_gradients = attn_gradients
|
||||
|
||||
def get_attn_gradients(self):
|
||||
return self.attn_gradients
|
||||
|
||||
def save_attention_map(self, attention_map):
|
||||
self.attention_map = attention_map
|
||||
|
||||
def get_attention_map(self):
|
||||
return self.attention_map
|
||||
|
||||
def forward(self, x, register_hook=False):
|
||||
B, N, C = x.shape
|
||||
qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
|
||||
q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)
|
||||
|
||||
attn = (q @ k.transpose(-2, -1)) * self.scale
|
||||
attn = attn.softmax(dim=-1)
|
||||
attn = self.attn_drop(attn)
|
||||
|
||||
if register_hook:
|
||||
self.save_attention_map(attn)
|
||||
attn.register_hook(self.save_attn_gradients)
|
||||
|
||||
x = (attn @ v).transpose(1, 2).reshape(B, N, C)
|
||||
x = self.proj(x)
|
||||
x = self.proj_drop(x)
|
||||
return x
|
||||
|
||||
|
||||
class Block(nn.Module):
|
||||
|
||||
def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
|
||||
drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, use_grad_checkpointing=False):
|
||||
super().__init__()
|
||||
self.norm1 = norm_layer(dim)
|
||||
self.attn = Attention(
|
||||
dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
|
||||
# NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
|
||||
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
|
||||
self.norm2 = norm_layer(dim)
|
||||
mlp_hidden_dim = int(dim * mlp_ratio)
|
||||
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
|
||||
|
||||
if use_grad_checkpointing:
|
||||
self.attn = checkpoint_wrapper(self.attn)
|
||||
self.mlp = checkpoint_wrapper(self.mlp)
|
||||
|
||||
def forward(self, x, register_hook=False):
|
||||
x = x + self.drop_path(self.attn(self.norm1(x), register_hook=register_hook))
|
||||
x = x + self.drop_path(self.mlp(self.norm2(x)))
|
||||
return x
|
||||
|
||||
|
||||
class VisionTransformer(nn.Module):
|
||||
""" Vision Transformer
|
||||
A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale` -
|
||||
https://arxiv.org/abs/2010.11929
|
||||
"""
|
||||
def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
|
||||
num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
|
||||
drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=None,
|
||||
use_grad_checkpointing=False, ckpt_layer=0):
|
||||
"""
|
||||
Args:
|
||||
img_size (int, tuple): input image size
|
||||
patch_size (int, tuple): patch size
|
||||
in_chans (int): number of input channels
|
||||
num_classes (int): number of classes for classification head
|
||||
embed_dim (int): embedding dimension
|
||||
depth (int): depth of transformer
|
||||
num_heads (int): number of attention heads
|
||||
mlp_ratio (int): ratio of mlp hidden dim to embedding dim
|
||||
qkv_bias (bool): enable bias for qkv if True
|
||||
qk_scale (float): override default qk scale of head_dim ** -0.5 if set
|
||||
representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
|
||||
drop_rate (float): dropout rate
|
||||
attn_drop_rate (float): attention dropout rate
|
||||
drop_path_rate (float): stochastic depth rate
|
||||
norm_layer: (nn.Module): normalization layer
|
||||
"""
|
||||
super().__init__()
|
||||
self.num_features = self.embed_dim = embed_dim # num_features for consistency with other models
|
||||
norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
|
||||
|
||||
self.patch_embed = PatchEmbed(
|
||||
img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
|
||||
|
||||
num_patches = self.patch_embed.num_patches
|
||||
|
||||
self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
|
||||
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
|
||||
self.pos_drop = nn.Dropout(p=drop_rate)
|
||||
|
||||
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)] # stochastic depth decay rule
|
||||
self.blocks = nn.ModuleList([
|
||||
Block(
|
||||
dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
|
||||
drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
|
||||
use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
|
||||
)
|
||||
for i in range(depth)])
|
||||
self.norm = norm_layer(embed_dim)
|
||||
|
||||
trunc_normal_(self.pos_embed, std=.02)
|
||||
trunc_normal_(self.cls_token, std=.02)
|
||||
self.apply(self._init_weights)
|
||||
|
||||
def _init_weights(self, m):
|
||||
if isinstance(m, nn.Linear):
|
||||
trunc_normal_(m.weight, std=.02)
|
||||
if isinstance(m, nn.Linear) and m.bias is not None:
|
||||
nn.init.constant_(m.bias, 0)
|
||||
elif isinstance(m, nn.LayerNorm):
|
||||
nn.init.constant_(m.bias, 0)
|
||||
nn.init.constant_(m.weight, 1.0)
|
||||
|
||||
@torch.jit.ignore
|
||||
def no_weight_decay(self):
|
||||
return {'pos_embed', 'cls_token'}
|
||||
|
||||
def forward(self, x, register_blk=-1):
|
||||
B = x.shape[0]
|
||||
x = self.patch_embed(x)
|
||||
|
||||
cls_tokens = self.cls_token.expand(B, -1, -1) # stole cls_tokens impl from Phil Wang, thanks
|
||||
x = torch.cat((cls_tokens, x), dim=1)
|
||||
|
||||
x = x + self.pos_embed[:,:x.size(1),:]
|
||||
x = self.pos_drop(x)
|
||||
|
||||
for i,blk in enumerate(self.blocks):
|
||||
x = blk(x, register_blk==i)
|
||||
x = self.norm(x)
|
||||
|
||||
return x
|
||||
|
||||
@torch.jit.ignore()
|
||||
def load_pretrained(self, checkpoint_path, prefix=''):
|
||||
_load_weights(self, checkpoint_path, prefix)
|
||||
|
||||
|
||||
@torch.no_grad()
|
||||
def _load_weights(model: VisionTransformer, checkpoint_path: str, prefix: str = ''):
|
||||
""" Load weights from .npz checkpoints for official Google Brain Flax implementation
|
||||
"""
|
||||
import numpy as np
|
||||
|
||||
def _n2p(w, t=True):
|
||||
if w.ndim == 4 and w.shape[0] == w.shape[1] == w.shape[2] == 1:
|
||||
w = w.flatten()
|
||||
if t:
|
||||
if w.ndim == 4:
|
||||
w = w.transpose([3, 2, 0, 1])
|
||||
elif w.ndim == 3:
|
||||
w = w.transpose([2, 0, 1])
|
||||
elif w.ndim == 2:
|
||||
w = w.transpose([1, 0])
|
||||
return torch.from_numpy(w)
|
||||
|
||||
w = np.load(checkpoint_path)
|
||||
if not prefix and 'opt/target/embedding/kernel' in w:
|
||||
prefix = 'opt/target/'
|
||||
|
||||
if hasattr(model.patch_embed, 'backbone'):
|
||||
# hybrid
|
||||
backbone = model.patch_embed.backbone
|
||||
stem_only = not hasattr(backbone, 'stem')
|
||||
stem = backbone if stem_only else backbone.stem
|
||||
stem.conv.weight.copy_(adapt_input_conv(stem.conv.weight.shape[1], _n2p(w[f'{prefix}conv_root/kernel'])))
|
||||
stem.norm.weight.copy_(_n2p(w[f'{prefix}gn_root/scale']))
|
||||
stem.norm.bias.copy_(_n2p(w[f'{prefix}gn_root/bias']))
|
||||
if not stem_only:
|
||||
for i, stage in enumerate(backbone.stages):
|
||||
for j, block in enumerate(stage.blocks):
|
||||
bp = f'{prefix}block{i + 1}/unit{j + 1}/'
|
||||
for r in range(3):
|
||||
getattr(block, f'conv{r + 1}').weight.copy_(_n2p(w[f'{bp}conv{r + 1}/kernel']))
|
||||
getattr(block, f'norm{r + 1}').weight.copy_(_n2p(w[f'{bp}gn{r + 1}/scale']))
|
||||
getattr(block, f'norm{r + 1}').bias.copy_(_n2p(w[f'{bp}gn{r + 1}/bias']))
|
||||
if block.downsample is not None:
|
||||
block.downsample.conv.weight.copy_(_n2p(w[f'{bp}conv_proj/kernel']))
|
||||
block.downsample.norm.weight.copy_(_n2p(w[f'{bp}gn_proj/scale']))
|
||||
block.downsample.norm.bias.copy_(_n2p(w[f'{bp}gn_proj/bias']))
|
||||
embed_conv_w = _n2p(w[f'{prefix}embedding/kernel'])
|
||||
else:
|
||||
embed_conv_w = adapt_input_conv(
|
||||
model.patch_embed.proj.weight.shape[1], _n2p(w[f'{prefix}embedding/kernel']))
|
||||
model.patch_embed.proj.weight.copy_(embed_conv_w)
|
||||
model.patch_embed.proj.bias.copy_(_n2p(w[f'{prefix}embedding/bias']))
|
||||
model.cls_token.copy_(_n2p(w[f'{prefix}cls'], t=False))
|
||||
pos_embed_w = _n2p(w[f'{prefix}Transformer/posembed_input/pos_embedding'], t=False)
|
||||
if pos_embed_w.shape != model.pos_embed.shape:
|
||||
pos_embed_w = resize_pos_embed( # resize pos embedding when different size from pretrained weights
|
||||
pos_embed_w, model.pos_embed, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
|
||||
model.pos_embed.copy_(pos_embed_w)
|
||||
model.norm.weight.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/scale']))
|
||||
model.norm.bias.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/bias']))
|
||||
# if isinstance(model.head, nn.Linear) and model.head.bias.shape[0] == w[f'{prefix}head/bias'].shape[-1]:
|
||||
# model.head.weight.copy_(_n2p(w[f'{prefix}head/kernel']))
|
||||
# model.head.bias.copy_(_n2p(w[f'{prefix}head/bias']))
|
||||
# if isinstance(getattr(model.pre_logits, 'fc', None), nn.Linear) and f'{prefix}pre_logits/bias' in w:
|
||||
# model.pre_logits.fc.weight.copy_(_n2p(w[f'{prefix}pre_logits/kernel']))
|
||||
# model.pre_logits.fc.bias.copy_(_n2p(w[f'{prefix}pre_logits/bias']))
|
||||
for i, block in enumerate(model.blocks.children()):
|
||||
block_prefix = f'{prefix}Transformer/encoderblock_{i}/'
|
||||
mha_prefix = block_prefix + 'MultiHeadDotProductAttention_1/'
|
||||
block.norm1.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/scale']))
|
||||
block.norm1.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/bias']))
|
||||
block.attn.qkv.weight.copy_(torch.cat([
|
||||
_n2p(w[f'{mha_prefix}{n}/kernel'], t=False).flatten(1).T for n in ('query', 'key', 'value')]))
|
||||
block.attn.qkv.bias.copy_(torch.cat([
|
||||
_n2p(w[f'{mha_prefix}{n}/bias'], t=False).reshape(-1) for n in ('query', 'key', 'value')]))
|
||||
block.attn.proj.weight.copy_(_n2p(w[f'{mha_prefix}out/kernel']).flatten(1))
|
||||
block.attn.proj.bias.copy_(_n2p(w[f'{mha_prefix}out/bias']))
|
||||
for r in range(2):
|
||||
getattr(block.mlp, f'fc{r + 1}').weight.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/kernel']))
|
||||
getattr(block.mlp, f'fc{r + 1}').bias.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/bias']))
|
||||
block.norm2.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/scale']))
|
||||
block.norm2.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/bias']))
|
||||
|
||||
|
||||
def interpolate_pos_embed(pos_embed_checkpoint, visual_encoder):
|
||||
# interpolate position embedding
|
||||
embedding_size = pos_embed_checkpoint.shape[-1]
|
||||
num_patches = visual_encoder.patch_embed.num_patches
|
||||
num_extra_tokens = visual_encoder.pos_embed.shape[-2] - num_patches
|
||||
# height (== width) for the checkpoint position embedding
|
||||
orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
|
||||
# height (== width) for the new position embedding
|
||||
new_size = int(num_patches ** 0.5)
|
||||
|
||||
if orig_size!=new_size:
|
||||
# class_token and dist_token are kept unchanged
|
||||
extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
|
||||
# only the position tokens are interpolated
|
||||
pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
|
||||
pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
|
||||
pos_tokens = torch.nn.functional.interpolate(
|
||||
pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
|
||||
pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
|
||||
new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
|
||||
print('reshape position embedding from %d to %d'%(orig_size ** 2,new_size ** 2))
|
||||
|
||||
return new_pos_embed
|
||||
else:
|
||||
return pos_embed_checkpoint
|
||||
@@ -1,20 +0,0 @@
|
||||
cmake
|
||||
fairscale>=0.4.4
|
||||
git+https://github.com/WASasquatch/img2texture.git
|
||||
git+https://github.com/WASasquatch/cstr
|
||||
gitpython
|
||||
imageio
|
||||
joblib
|
||||
matplotlib
|
||||
numba
|
||||
numpy>=1.18.5, <1.25.0
|
||||
opencv-python-headless[ffmpeg]<=4.7.0.72
|
||||
pilgram
|
||||
git+https://github.com/WASasquatch/ffmpy.git
|
||||
rembg
|
||||
scikit-image==0.20.0
|
||||
scikit-learn
|
||||
scipy
|
||||
timm>=0.4.12
|
||||
tqdm
|
||||
transformers==4.26.1
|
||||
Binary file not shown.
-12213
File diff suppressed because it is too large
Load Diff
-24350
File diff suppressed because it is too large
Load Diff
-20719
File diff suppressed because it is too large
Load Diff
-103493
File diff suppressed because it is too large
Load Diff
-35712
File diff suppressed because it is too large
Load Diff
-29690
File diff suppressed because it is too large
Load Diff
-28134
File diff suppressed because it is too large
Load Diff
-6693
File diff suppressed because it is too large
Load Diff
@@ -1,23 +0,0 @@
|
||||
{
|
||||
"run_requirements": true,
|
||||
"suppress_uncomfy_warnings": true,
|
||||
"show_startup_junk": true,
|
||||
"show_inspiration_quote": true,
|
||||
"text_nodes_type": "STRING",
|
||||
"webui_styles": null,
|
||||
"webui_styles_persistent_update": true,
|
||||
"blip_model_url": "https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_capfilt_large.pth",
|
||||
"blip_model_vqa_url": "https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_vqa_capfilt_large.pth",
|
||||
"sam_model_vith_url": "https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth",
|
||||
"sam_model_vitl_url": "https://dl.fbaipublicfiles.com/segment_anything/sam_vit_l_0b3195.pth",
|
||||
"sam_model_vitb_url": "https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth",
|
||||
"history_display_limit": 36,
|
||||
"use_legacy_ascii_text": false,
|
||||
"ffmpeg_bin_path": "/path/to/ffmpeg",
|
||||
"ffmpeg_extra_codecs": {
|
||||
"avc1": ".mp4",
|
||||
"h264": ".mkv"
|
||||
},
|
||||
"wildcards_path": "/home/salt/clone/storyteller/storyteller-ml/workflows/comfy/ComfyUI/custom_nodes/was-node-suite-comfyui/wildcards",
|
||||
"wildcard_api": true
|
||||
}
|
||||
Reference in New Issue
Block a user