Initial commit

.gitattributes

@@ -0,0 +1 @@
+*.png filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,176 @@
+# 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
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.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
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/latest/usage/project/#working-with-version-control
+.pdm.toml
+.pdm-python
+.pdm-build/
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__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 maintained 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/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+.idea/

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Fangyikang Wang 
+
+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.

README.md

@@ -0,0 +1,180 @@
+<div align="center">
+
+# 🚀🚀🚀 Improve Diffusion Image Generation Quality using Levenberg-Marquardt-Langevin
+
+We introduce **LML**, an accelerated sampler for diffusion models leveraging the second-order Hessian geometry. Our LML imlpementation is completely compatible with the **[diffusers](https://github.com/huggingface/diffusers)**.
+
+This repository is the official implementation of the **ICCV 2025** paper:
+_"Unleashing High-Quality Image Generation in Diffusion Sampling Using Second-Order Levenberg-Marquardt-Langevin"_ 
+
+
+> **Fangyikang Wang<sup>1,2</sup>, Hubery Yin<sup>2</sup>, Lei Qian<sup>1</sup>, Yinan Li<sup>1</sup>, Shaobin Zhuang<sup>3,2</sup>, Huminhao Zhu<sup>1</sup>, Yilin Zhang<sup>1</sup>, Yanlong Tang<sup>4</sup>, Chao Zhang<sup>1</sup>, Hanbin Zhao<sup>1</sup>, Hui Qian<sup>1</sup>, Chen Li<sup>2</sup>**
+>
+> <sup>1</sup>Zhejiang University <sup>2</sup>WeChat Vision, Tencent Inc <sup>3</sup>Shanghai Jiao Tong University <sup>4</sup>Tencent Lightspeed Studio
+
+[![arXiv](https://img.shields.io/badge/arXiv%20paper-2505.24222-b31b1b.svg)](https://www.arxiv.org/abs/2505.24222)&nbsp;
+[![Github](https://img.shields.io/badge/Github-LML-blue)](https://github.com/zituitui/LML-diffusion-sampler)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)&nbsp;
+
+<img src="assets/lml-sd-visual_2_new-1.png" alt="SD Results" style="width: 100%;">
+
+<img src="assets/lml-celeb-visual-1.png" alt="celeb Results" style="width: 70%;">
+
+</div>
+
+## The intuition of our LML diffusion sampler
+
+![anneal](assets/anneal_path.drawio-1.png)
+> **Schematic comparison** between our LML method and baselines. While previous works mainly focus on intriguing designs along the annealing path to improve diffusion sampling, they leave operations at specific noise levels to be performed using first-order Langevin. Our approach proposes to leverage the Levenberg-Marquardt approximated Hessian geometry to guide the Langevin update to be more accurate.
+
+
+![Some edits](assets/newton_algos.drawio-1.png)
+> The relation between optimization algorithms and MCMC sampling algorithms. We initially wanted to develop a diffusion sampler utilizing Hessian geometry, following the path of Newton-Langevin dynamics.
+However, this approach proved to be highly computationally expensive within the DM context.
+Drawing inspiration from the Levenberg-Marquardt method used in optimization, our method incorporates low-rank approximation and damping techniques. This enables us to obtain the Hessian geometry in a computationally affordable manner. Subsequently, we use this approximated Hessian geometry to guide the Langevin updates.
+
+## 👨🏻‍💻 Run the code 
+### 1) Get start
+
+* Python 3.8.12
+* CUDA 11.7
+* NVIDIA A100 40GB PCIe
+* Torch 2.0.0
+* Torchvision 0.14.0
+
+Please follow **[diffusers](https://github.com/huggingface/diffusers)** to install diffusers.
+
+### 2) Sampling
+first, please switch to the root directory.
+
+- #### CIFAR-10 sampling
+  For baseline, you can do CIAFR-10 sampling as follows, choose sampler_type within [ddim, pndm, dpm, dpm++, unipc]:
+  ```bash
+  python3 ./scripts/cifar10.py --test_num 1 --batch_size 1 --num_inference_steps 10  --save_dir YOUR/SAVE/DIR --model_id xx/xx/ddpm_ema_cifar10 --sampler_type ddim
+  ```
+  For our LML sampler, there is an additional $\lambda$ hyperparameter:
+  ```bash
+  python3 ./scripts/cifar10.py --test_num 1 --batch_size 1 --num_inference_steps 10  --save_dir YOUR/SAVE/DIR --model_id xx/xx/ddpm_ema_cifar10 --sampler_type dpm_lm --lamb 0.0008
+  ```
+
+  For the optimal choice of LML, we have:
+  |         | 5 NFEs  | 6 NFEs  | 7 NFEs  | 8 NFEs  | 9 NFEs  | 10 NFEs | 12 NFEs | 15 NFEs | 20 NFEs | 30 NFEs | 50 NFEs | 100 NFEs |
+  |---------|---------|---------|---------|---------|---------|----------|----------|----------|----------|----------|----------|-----------|
+  | optimal value of lamb     | 0.0008  | 0.0008  | 0.001   | 0.001   | 0.001   | 0.0008   | 0.001    | 0.001    | 0.0005   | 0.0003   | 0.0001   | 0.00005   |
+
+
+
+- #### CelebA-HQ sampling
+  For baseline:
+  ```bash
+  python3 ./scripts/celeba.py --test_num 1 --batch_size 1 --num_inference_steps 10  --save_dir YOUR/SAVE/DIR --model_id xx/xx/ldm-celebahq-256 --sampler_type ddim
+  ```
+
+  For our LML:
+  ```bash
+  python3 ./scripts/celeba.py --test_num 1 --batch_size 1 --num_inference_steps 10  --save_dir YOUR/SAVE/DIR --model_id xx/xx/ldm-celebahq-256 --sampler_type ddim_lm --lamb 0.005
+  ```
+
+  - #### SD-15 and SD-2b on MS-COCO sampling
+  ```bash
+  python3 ./scripts/StableDiffusion_COCO.py --test_num 30002 --num_inference_steps 10  --save_dir YOUR/SAVE/DIR --model_id xx/xx/stable-diffusion-v1-5 --sampler_type dpm_lm --lamb 0.001
+  ```
+
+  For the optimal choice of LML on MS-COCO, for NFEs of {5, 6, 7, 8, 9, 10, 12, 15}, we always choose $\lambda = 0.001$:
+  <!-- |NFEs| 5    | 6    | 7    | 8    | 9    | 10   | 12   | 15   |
+  |---------|------|------|------|------|------|------|------|------|
+  | SD-15   | -- | -- | 0.001 | - | - | -| 0.001 | 0.001 |
+  | SD-2b   | -- | - | - | - | - | - | - | - | -->
+
+
+- #### SD-15, SD-2b, SD-XL, and PixArt-$\alpha$ on T2i-compbench sampling
+  Before running the scripts, make sure to clone T2I-CompBench repository. Generated images are stored in the directory "save_dir/model/dataset_category/sampler_type/samples".
+
+  For baseline, you can do T2i-compbench sampling as follows, choose sampler_type within [ddim, pndm, dpm, dpm++, unipc] and model within [sd15, sd2_base, sdxl, pixart]:
+  ```bash
+  python3 ./scripts/StableDiffusion_PixArt_T2i_Sampling.py --dataset_category color --dataset_path PATH/TO/T2I-COMPBENCH --test_num 10 --num_inference_steps 10 --model_dir YOUR/MODEL/DIR --save_dir YOUR/SAVE/DIR --model sd15 --sampler_type ddim
+  ```
+  For our LML sampler, there is an additional $\lambda$ hyperparameter:
+  ```bash
+  python3 ./scripts/StableDiffusion_PixArt_T2i_Sampling.py --dataset_category color --dataset_path PATH/TO/T2I-COMPBENCH --test_num 10 --num_inference_steps 10 --model_dir YOUR/MODEL/DIR --save_dir YOUR/SAVE/DIR --model sd15 --sampler_type dpm_lm --lamb 0.006
+  ```
+
+- #### Use our LML diffusion sampler with ControlNet
+
+  **canny**
+  ```bash
+  python3 ./scripts/control_net_canny.py --num_inference_steps 10  --original_image_path /xxx/xxx/data/input_image_vermeer.png --controlnet_dir /xxx/xxx/sd-controlnet-canny --sd_dir /xxx/xxx/stable-diffusion-v1-5  --save_dir YOUR/SAVE/DIR  --sampler_type dpm_lm --lamb 0.001
+  ```
+
+  **depth**
+  ```bash
+  python3 ./scripts/control_net_depth.py --num_inference_steps 10  --controlnet_dir /xxx/xxx/control_v11f1p_sd15_depth --sd_dir /xxx/xxx/stable-diffusion-v1-5  --save_dir YOUR/SAVE/DIR  --sampler_type dpm_lm --lamb 0.001
+  ```
+
+  **pose**
+  ```bash
+  python3 ./scripts/control_net_canny.py --num_inference_steps 10 --controlnet_dir /xxx/xxx/sd-controlnet-openpose --sd_dir /xxx/xxx/stable-diffusion-v1-5  --save_dir YOUR/SAVE/DIR  --sampler_type dpm_lm --lamb 0.001
+  ```
+
+
+- #### LML sampling on FLUX
+  For baseline:
+  ```bash
+  python3 ./scripts/FLUX_T2i_Sampling.py --dataset_category color --dataset_path PATH/TO/T2I-COMPBENCH --test_num 10 --num_inference_steps 10 --model_id YOUR/MODEL/DIR --save_dir YOUR/SAVE/DIR --sampler_type fm_euler
+  ```
+  For our LML:
+  ```bash
+  python3 ./scripts/FLUX_T2i_Sampling.py --dataset_category color --dataset_path PATH/TO/T2I-COMPBENCH --test_num 10 --num_inference_steps 10 --model_id YOUR/MODEL/DIR --save_dir YOUR/SAVE/DIR --sampler_type lml_euler --lamb 0.01
+  ```
+
+
+### 3) Evaluation
+- #### FID evaluation on CIFAR-10
+  [Coming Soon] ⏳
+
+- #### FID evaluation on MS-COCO
+  [Coming Soon] ⏳
+
+- #### T2I-compbench evaluation 
+  Please refer to the [T2I-CompBench](https://github.com/Karine-Huang/T2I-CompBench) guide. Create a new environment and install the dependencies for T2I-CompBench evaluation.
+  For testing combinations of multiple models and samplers, we also provide a convenient one-click script. Place the script file in the corresponding directory of **T2I-CompBench** to replace the origin script. For example:
+  ```sh
+  # BLIP-VQA for Attribute Binding
+  cd T2I-CompBench
+  bash BLIPvqa_eval/test.sh
+        ||
+        ||
+        \/
+  cp evaluations/T2I-CompBench/BLIPvqa_test.sh T2I-CompBench/BLIPvqa_eval
+  cd T2I-CompBench
+  bash BLIPvqa_eval/BLIPvqa_test.sh 'save_dir'
+  ```
+  The directory structure of **'save_dir'** should satisfy the following format:
+  ```
+  {save_dir}/model/dataset_category/sampler_type/samples/
+                                                ├── a green bench and a blue bowl_000000.png
+                                                ├── a green bench and a blue bowl_000001.png
+                                                └──...
+  ```
+
+
+
+## 🪪 License
+This project is licensed under the MIT License - see the [LICENSE](LICENSE.txt) file for details.
+
+## 📝 Citation
+If our work assists your research, feel free to give us a star ⭐ or cite us using:
+```
+@article{wang2025unleashing,
+  title={Unleashing High-Quality Image Generation in Diffusion Sampling Using Second-Order Levenberg-Marquardt-Langevin},
+  author={Wang, Fangyikang and Yin, Hubery and Qian, Lei and Li, Yinan and Zhuang, Shaobin and Zhu, Huminhao and Zhang, Yilin and Tang, Yanlong and Zhang, Chao and Zhao, Hanbin and others},
+  journal={arXiv preprint arXiv:2505.24222},
+  year={2025}
+}
+```
+
+## 📩 Contact me
+Our e-mail address:
+```
+[email protected], [email protected], [email protected]
+```

evaluations/T2I-CompBench/3-in-1_test.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+
+export project_dir="3_in_1_eval/"
+cd "$project_dir" || exit
+
+save_dir=$1
+model_list=($(ls -d "$save_dir"/*/ | xargs -n1 basename))
+
+for model in "${model_list[@]}"
+do
+    # attr_list=($(ls -d "$save_dir/$model"/*/ | xargs -n1 basename))
+    attr_list=("complex")
+    for attr in "${attr_list[@]}"
+    do
+        sampler_list=($(ls -d "$save_dir/$model/$attr"/*/ | xargs -n1 basename))
+        for sampler in "${sampler_list[@]}"
+        do
+            out_dir="$save_dir/$model/$attr/$sampler"
+            # run python script
+            echo "Running for model=$model, attr=$attr, sampler=$sampler"
+            python 3_in_1.py --outpath="$out_dir"
+            
+            # check if the command was successful
+            if [ $? -ne 0 ]; then
+                failed_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            else
+                success_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            fi
+        done
+    done
+done
+
+# print count of runs
+echo "Total runs: ${#success_runs[@]} succeeded, ${#failed_runs[@]} failed."
+
+# print all failed runs
+if [ ${#failed_runs[@]} -ne 0 ]; then
+    echo "The following runs failed:"
+    for run in "${failed_runs[@]}"
+    do
+        echo "$run"
+    done
+else
+    echo "All runs completed successfully."
+fi

evaluations/T2I-CompBench/BLIPvqa_test.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+
+export project_dir="BLIPvqa_eval/"
+cd "$project_dir" || exit
+
+save_dir=$1
+model_list=($(ls -d "$save_dir"/*/ | xargs -n1 basename))
+
+for model in "${model_list[@]}"
+do
+    # attr_list=($(ls -d "$save_dir/$model"/*/ | xargs -n1 basename))
+    attr_list=("color" "shape" "texture")  # attribute list
+    for attr in "${attr_list[@]}"
+    do
+        sampler_list=($(ls -d "$save_dir/$model/$attr"/*/ | xargs -n1 basename))
+        for sampler in "${sampler_list[@]}"
+        do
+            out_dir="$save_dir/$model/$attr/$sampler"
+            # run python script
+            echo "Running for model=$model, attr=$attr, sampler=$sampler"
+            python BLIP_vqa.py --out_dir="$out_dir"
+            
+            # check if the command was successful
+            if [ $? -ne 0 ]; then
+                failed_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            else
+                success_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            fi
+        done
+    done
+done
+
+# print count of runs
+echo "Total runs: ${#success_runs[@]} succeeded, ${#failed_runs[@]} failed."
+
+# print all failed runs
+if [ ${#failed_runs[@]} -ne 0 ]; then
+    echo "The following runs failed:"
+    for run in "${failed_runs[@]}"
+    do
+        echo "$run"
+    done
+else
+    echo "All runs completed successfully."
+fi

evaluations/T2I-CompBench/CLIPScore_test.sh

@@ -0,0 +1,46 @@
+#!/bin/bash
+
+save_dir=$1
+model_list=($(ls -d "$save_dir"/*/ | xargs -n1 basename))
+
+for model in "${model_list[@]}"
+do
+    # attr_list=($(ls -d "$save_dir/$model"/*/ | xargs -n1 basename))
+    attr_list=("complex" "non_spatial")
+    for attr in "${attr_list[@]}"
+    do
+        sampler_list=($(ls -d "$save_dir/$model/$attr"/*/ | xargs -n1 basename))
+        for sampler in "${sampler_list[@]}"
+        do
+            out_dir="$save_dir/$model/$attr/$sampler"
+            # run python script
+            echo "Running for model=$model, attr=$attr, sampler=$sampler"
+            if [ "$attr" = "complex" ]; then
+                python CLIPScore_eval/CLIP_similarity.py --outpath="$out_dir" --complex=True
+            else
+                python CLIPScore_eval/CLIP_similarity.py --outpath="$out_dir"
+            fi
+            
+            # check if the command was successful
+            if [ $? -ne 0 ]; then
+                failed_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            else
+                success_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            fi
+        done
+    done
+done
+
+# print count of runs
+echo "Total runs: ${#success_runs[@]} succeeded, ${#failed_runs[@]} failed."
+
+# print all failed runs
+if [ ${#failed_runs[@]} -ne 0 ]; then
+    echo "The following runs failed:"
+    for run in "${failed_runs[@]}"
+    do
+        echo "$run"
+    done
+else
+    echo "All runs completed successfully."
+fi

evaluations/T2I-CompBench/UniDet_test.sh

@@ -0,0 +1,50 @@
+#!/bin/bash
+
+export project_dir="UniDet_eval/"
+cd "$project_dir" || exit
+
+save_dir=$1
+model_list=($(ls -d "$save_dir"/*/ | xargs -n1 basename))
+
+for model in "${model_list[@]}"
+do
+    # attr_list=($(ls -d "$save_dir/$model"/*/ | xargs -n1 basename))
+    attr_list=("spatial" "complex")
+    for attr in "${attr_list[@]}"
+    do
+        sampler_list=($(ls -d "$save_dir/$model/$attr"/*/ | xargs -n1 basename))
+        for sampler in "${sampler_list[@]}"
+        do
+            out_dir="$save_dir/$model/$attr/$sampler/"
+
+            # run python script
+            echo "Running for model=$model, attr=$attr, sampler=$sampler"
+            if [ "$attr" = "complex" ]; then
+                python 2D_spatial_eval.py --outpath="$out_dir" --complex=True
+            else
+                python 2D_spatial_eval.py --outpath="$out_dir"
+            fi
+            
+            # check if the command was successful
+            if [ $? -ne 0 ]; then
+                failed_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            else
+                success_runs+=("model=$model, attr=$attr, sampler=$sampler")
+            fi
+        done
+    done
+done
+
+# print count of runs
+echo "Total runs: ${#success_runs[@]} succeeded, ${#failed_runs[@]} failed."
+
+# print all failed runs
+if [ ${#failed_runs[@]} -ne 0 ]; then
+    echo "The following runs failed:"
+    for run in "${failed_runs[@]}"
+    do
+        echo "$run"
+    done
+else
+    echo "All runs completed successfully."
+fi

evaluations/fid_score.py

@@ -0,0 +1,417 @@
+"""Calculates the Frechet Inception Distance (FID) to evalulate GANs
+
+The FID metric calculates the distance between two distributions of images.
+Typically, we have summary statistics (mean & covariance matrix) of one
+of these distributions, while the 2nd distribution is given by a GAN.
+
+When run as a stand-alone program, it compares the distribution of
+images that are stored as PNG/JPEG at a specified location with a
+distribution given by summary statistics (in pickle format).
+
+The FID is calculated by assuming that X_1 and X_2 are the activations of
+the pool_3 layer of the inception net for generated samples and real world
+samples respectively.
+
+See --help to see further details.
+
+Code apapted from https://github.com/bioinf-jku/TTUR to use PyTorch instead
+of Tensorflow
+
+Copyright 2018 Institute of Bioinformatics, JKU Linz
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+"""
+import os
+import pathlib
+from argparse import ArgumentDefaultsHelpFormatter, ArgumentParser
+
+import numpy as np
+import torch
+import torchvision.transforms as TF
+from PIL import Image
+from scipy import linalg
+from torch.nn.functional import adaptive_avg_pool2d
+
+try:
+    from tqdm import tqdm
+except ImportError:
+    # If tqdm is not available, provide a mock version of it
+    def tqdm(x):
+        return x
+
+from inception import InceptionV3
+
+IMAGE_EXTENSIONS = {'bmp', 'jpg', 'jpeg', 'pgm', 'png', 'ppm',
+                        'tif', 'tiff', 'webp'}
+
+
+
+class ImagePathDataset(torch.utils.data.Dataset):
+    def __init__(self, files, transforms=None):
+        self.files = files
+        self.transforms = transforms
+
+    def __len__(self):
+        return len(self.files)
+
+    def __getitem__(self, i):
+        path = self.files[i]
+        img = Image.open(path).convert('RGB')
+        if self.transforms is not None:
+            img = self.transforms(img)
+        return img
+
+
+def get_activations(files, model, batch_size=50, dims=2048, device='cpu',
+                    num_workers=1):
+    """Calculates the activations of the pool_3 layer for all images.
+
+    Params:
+    -- files       : List of image files paths
+    -- model       : Instance of inception model
+    -- batch_size  : Batch size of images for the model to process at once.
+                     Make sure that the number of samples is a multiple of
+                     the batch size, otherwise some samples are ignored. This
+                     behavior is retained to match the original FID score
+                     implementation.
+    -- dims        : Dimensionality of features returned by Inception
+    -- device      : Device to run calculations
+    -- num_workers : Number of parallel dataloader workers
+
+    Returns:
+    -- A numpy array of dimension (num images, dims) that contains the
+       activations of the given tensor when feeding inception with the
+       query tensor.
+    """
+    model.eval()
+
+    if batch_size > len(files):
+        print(('Warning: batch size is bigger than the data size. '
+               'Setting batch size to data size'))
+        batch_size = len(files)
+
+    coco_transform = TF.Compose([
+     TF.Resize(512),
+     TF.CenterCrop(512),
+     TF.ToTensor()])
+
+    dataset = ImagePathDataset(files, transforms=coco_transform)
+    dataloader = torch.utils.data.DataLoader(dataset,
+                                             batch_size=batch_size,
+                                             shuffle=False,
+                                             drop_last=False,
+                                             num_workers=num_workers)
+
+    pred_arr = np.empty((len(files), dims))
+
+    start_idx = 0
+
+    for batch in tqdm(dataloader):
+        batch = batch.to(device)
+
+        with torch.no_grad():
+            pred = model(batch)[0]
+
+        # If model output is not scalar, apply global spatial average pooling.
+        # This happens if you choose a dimensionality not equal 2048.
+        if pred.size(2) != 1 or pred.size(3) != 1:
+            pred = adaptive_avg_pool2d(pred, output_size=(1, 1))
+
+        pred = pred.squeeze(3).squeeze(2).cpu().numpy()
+
+        pred_arr[start_idx:start_idx + pred.shape[0]] = pred
+
+        start_idx = start_idx + pred.shape[0]
+
+    return pred_arr
+
+
+def calculate_frechet_distance(mu1, sigma1, mu2, sigma2, eps=1e-6):
+    """Numpy implementation of the Frechet Distance.
+    The Frechet distance between two multivariate Gaussians X_1 ~ N(mu_1, C_1)
+    and X_2 ~ N(mu_2, C_2) is
+            d^2 = ||mu_1 - mu_2||^2 + Tr(C_1 + C_2 - 2*sqrt(C_1*C_2)).
+
+    Stable version by Dougal J. Sutherland.
+
+    Params:
+    -- mu1   : Numpy array containing the activations of a layer of the
+               inception net (like returned by the function 'get_predictions')
+               for generated samples.
+    -- mu2   : The sample mean over activations, precalculated on an
+               representative data set.
+    -- sigma1: The covariance matrix over activations for generated samples.
+    -- sigma2: The covariance matrix over activations, precalculated on an
+               representative data set.
+
+    Returns:
+    --   : The Frechet Distance.
+    """
+
+    mu1 = np.atleast_1d(mu1)
+    mu2 = np.atleast_1d(mu2)
+
+    sigma1 = np.atleast_2d(sigma1)
+    sigma2 = np.atleast_2d(sigma2)
+
+    assert mu1.shape == mu2.shape, \
+        'Training and test mean vectors have different lengths'
+    assert sigma1.shape == sigma2.shape, \
+        'Training and test covariances have different dimensions'
+
+    diff = mu1 - mu2
+
+    # Product might be almost singular
+    covmean, _ = linalg.sqrtm(sigma1.dot(sigma2), disp=False)
+    if not np.isfinite(covmean).all():
+        msg = ('fid calculation produces singular product; '
+               'adding %s to diagonal of cov estimates') % eps
+        print(msg)
+        offset = np.eye(sigma1.shape[0]) * eps
+        covmean = linalg.sqrtm((sigma1 + offset).dot(sigma2 + offset))
+
+    # Numerical error might give slight imaginary component
+    if np.iscomplexobj(covmean):
+        if not np.allclose(np.diagonal(covmean).imag, 0, atol=1e-3):
+            m = np.max(np.abs(covmean.imag))
+            raise ValueError('Imaginary component {}'.format(m))
+        covmean = covmean.real
+
+    tr_covmean = np.trace(covmean)
+
+    return (diff.dot(diff) + np.trace(sigma1)
+            + np.trace(sigma2) - 2 * tr_covmean)
+
+
+def calculate_activation_statistics(files, model, batch_size=50, dims=2048,
+                                    device='cpu', num_workers=1):
+    """Calculation of the statistics used by the FID.
+    Params:
+    -- files       : List of image files paths
+    -- model       : Instance of inception model
+    -- batch_size  : The images numpy array is split into batches with
+                     batch size batch_size. A reasonable batch size
+                     depends on the hardware.
+    -- dims        : Dimensionality of features returned by Inception
+    -- device      : Device to run calculations
+    -- num_workers : Number of parallel dataloader workers
+
+    Returns:
+    -- mu    : The mean over samples of the activations of the pool_3 layer of
+               the inception model.
+    -- sigma : The covariance matrix of the activations of the pool_3 layer of
+               the inception model.
+    """
+    act = get_activations(files, model, batch_size, dims, device, num_workers)
+    mu = np.mean(act, axis=0)
+    sigma = np.cov(act, rowvar=False)
+    return mu, sigma
+
+
+def compute_statistics_of_path(path, model, batch_size, dims, device,
+                               num_workers=1):
+    if path.endswith('.npz'):
+        with np.load(path) as f:
+            m, s = f['mu'][:], f['sigma'][:]
+    else:
+        path = pathlib.Path(path)
+        files = sorted([file for ext in IMAGE_EXTENSIONS
+                       for file in path.glob('**/*.{}'.format(ext))])
+        m, s = calculate_activation_statistics(files, model, batch_size,
+                                               dims, device, num_workers)
+
+    return m, s
+
+
+def calculate_fid_given_paths(paths, batch_size, device, dims, num_workers=1, output=''):
+    """Calculates the FID of two paths"""
+    for p in paths:
+        if not os.path.exists(p):
+            raise RuntimeError('Invalid path: %s' % p)
+
+    block_idx = InceptionV3.BLOCK_INDEX_BY_DIM[dims]
+
+    model = InceptionV3([block_idx]).to(device)
+
+    m1, s1 = compute_statistics_of_path(paths[0], model, batch_size,
+                                        dims, device, num_workers)
+
+    #np.savez_compressed('./1W.npz', mu=m1, sigma=s1)
+
+    #np.save('./'+output+'_m.npy', m1)
+    #np.save('./'+output+'_s.npy', s1)
+    m2, s2 = compute_statistics_of_path(paths[1], model, batch_size,
+                                        dims, device, num_workers)
+    fid_value = calculate_frechet_distance(m1, s1, m2, s2)
+
+    return fid_value
+
+
+def save_fid_stats(paths, batch_size, device, dims, num_workers=1):
+    """Calculates the FID of two paths"""
+    if not os.path.exists(paths[0]):
+        raise RuntimeError('Invalid path: %s' % paths[0])
+
+    if os.path.exists(paths[1]):
+        raise RuntimeError('Existing output file: %s' % paths[1])
+
+    block_idx = InceptionV3.BLOCK_INDEX_BY_DIM[dims]
+
+    model = InceptionV3([block_idx]).to(device)
+
+    print(f"Saving statistics for {paths[0]}")
+
+    m1, s1 = compute_statistics_of_path(paths[0], model, batch_size,
+                                        dims, device, num_workers)
+
+    np.savez_compressed(paths[1], mu=m1, sigma=s1)
+
+
+def main():
+    args = parser.parse_args()
+
+    if args.device is None:
+        device = torch.device('cuda' if (torch.cuda.is_available()) else 'cpu')
+    else:
+        device = torch.device(args.device)
+
+    if args.num_workers is None:
+        try:
+            num_cpus = len(os.sched_getaffinity(0))
+        except AttributeError:
+            # os.sched_getaffinity is not available under Windows, use
+            # os.cpu_count instead (which may not return the *available* number
+            # of CPUs).
+            num_cpus = os.cpu_count()
+
+        num_workers = min(num_cpus, 8) if num_cpus is not None else 0
+    else:
+        num_workers = args.num_workers
+
+    if args.save_stats:
+        save_fid_stats(args.path, args.batch_size, device, args.dims, num_workers)
+        return
+
+    fid_value = calculate_fid_given_paths(args.path,
+                                          args.batch_size,
+                                          device,
+                                          args.dims,
+                                          num_workers,
+                                          args.output)
+    print('FID: ', fid_value)
+
+
+def main():
+    parser = ArgumentParser(formatter_class=ArgumentDefaultsHelpFormatter)
+    parser.add_argument('--batch-size', type=int, default=50,
+                        help='Batch size to use')
+    parser.add_argument('--num-workers', type=int,
+                        help=('Number of processes to use for data loading. '
+                              'Defaults to `min(8, num_cpus)`'))
+    parser.add_argument('--device', type=str, default=None,
+                        help='Device to use. Like cuda, cuda:0 or cpu')
+    parser.add_argument('--dims', type=int, default=2048,
+                        choices=list(InceptionV3.BLOCK_INDEX_BY_DIM),
+                        help=('Dimensionality of Inception features to use. '
+                              'By default, uses pool3 features'))
+    parser.add_argument('--save-stats', action='store_true',
+                        help=('Generate an npz archive from a directory of samples. '
+                              'The first path is used as input and the second as output.'))
+    parser.add_argument('path', type=str, nargs=2,
+                        help=('Paths to the generated images or '
+                              'to .npz statistic files'))
+    parser.add_argument('--output', type=str, default='',
+                        help='Device to use. Like cuda, cuda:0 or cpu')
+    parser.add_argument('--input', type=str, default='',
+                        help='Device to use. Like cuda, cuda:0 or cpu')
+
+    IMAGE_EXTENSIONS = {'bmp', 'jpg', 'jpeg', 'pgm', 'png', 'ppm',
+                        'tif', 'tiff', 'webp'}
+    args = parser.parse_args()
+
+    if args.device is None:
+        device = torch.device('cuda' if (torch.cuda.is_available()) else 'cpu')
+    else:
+        device = torch.device(args.device)
+
+    if args.num_workers is None:
+        try:
+            num_cpus = len(os.sched_getaffinity(0))
+        except AttributeError:
+            # os.sched_getaffinity is not available under Windows, use
+            # os.cpu_count instead (which may not return the *available* number
+            # of CPUs).
+            num_cpus = os.cpu_count()
+
+        num_workers = min(num_cpus, 8) if num_cpus is not None else 0
+    else:
+        num_workers = args.num_workers
+
+    if args.save_stats:
+        save_fid_stats(args.path, args.batch_size, device, args.dims, num_workers)
+        return
+
+    fid_value = calculate_fid_given_paths(args.path,
+                                          args.batch_size,
+                                          device,
+                                          args.dims,
+                                          num_workers,
+                                          args.output)
+    print('FID: ', fid_value)
+
+
+def cal_fid():
+    parser = ArgumentParser(formatter_class=ArgumentDefaultsHelpFormatter)
+    parser.add_argument('--batch-size', type=int, default=50,
+                        help='Batch size to use')
+    parser.add_argument('--num-workers', type=int,
+                        help=('Number of processes to use for data loading. '
+                              'Defaults to `min(8, num_cpus)`'))
+    parser.add_argument('--device', type=str, default=None,
+                        help='Device to use. Like cuda, cuda:0 or cpu')
+    parser.add_argument('--dims', type=int, default=2048,
+                        choices=list(InceptionV3.BLOCK_INDEX_BY_DIM),
+                        help=('Dimensionality of Inception features to use. '
+                              'By default, uses pool3 features'))
+    parser.add_argument('--save-stats', action='store_true',
+                        help=('Generate an npz archive from a directory of samples. '
+                              'The first path is used as input and the second as output.'))
+    parser.add_argument('path', type=str, nargs=2,
+                        help=('Paths to the generated images or '
+                              'to .npz statistic files'))
+    parser.add_argument('--output', type=str, default='',
+                        help='Device to use. Like cuda, cuda:0 or cpu')
+    parser.add_argument('--input', type=str, default='',
+                        help='Device to use. Like cuda, cuda:0 or cpu')
+
+    IMAGE_EXTENSIONS = {'bmp', 'jpg', 'jpeg', 'pgm', 'png', 'ppm',
+                        'tif', 'tiff', 'webp'}
+    args = parser.parse_args()
+
+    m1 = np.load('./target_fid_test_m.npy')
+    s1 = np.load('./target_fid_test_s.npy')
+
+    m2 = np.load(args.input+'_m.npy')
+    s2 = np.load(args.input+'_s.npy')
+
+    fid_value = calculate_frechet_distance(m1, s1, m2, s2)
+
+
+    print('FID: ', fid_value)
+
+
+if __name__ == '__main__':
+    print("Alibaba")
+    main()
+    
+    #cal_fid()

scheduler/scheduling_ddim_lm.py

@@ -0,0 +1,558 @@
+# Copyright 2025 ZJU Lab304 Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# DISCLAIMER: This code is strongly influenced by https://github.com/pesser/pytorch_diffusion
+# and https://github.com/hojonathanho/diffusion
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from diffusers.configuration_utils import ConfigMixin, register_to_config
+from diffusers.utils import BaseOutput
+from diffusers.utils.torch_utils import randn_tensor
+from diffusers.schedulers.scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin
+
+
+@dataclass
+# Copied from diffusers.schedulers.scheduling_ddpm.DDPMSchedulerOutput with DDPM->DDIM
+class DDIMLMSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's step function output.
+
+    Args:
+        prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+            Computed sample (x_{t-1}) of previous timestep. `prev_sample` should be used as next model input in the
+            denoising loop.
+        pred_original_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+            The predicted denoised sample (x_{0}) based on the model output from the current timestep.
+            `pred_original_sample` can be used to preview progress or for guidance.
+    """
+
+    prev_sample: torch.FloatTensor
+    pred_original_sample: Optional[torch.FloatTensor] = None
+
+
+# Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
+def betas_for_alpha_bar(
+    num_diffusion_timesteps,
+    max_beta=0.999,
+    alpha_transform_type="cosine",
+):
+    """
+    Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of
+    (1-beta) over time from t = [0,1].
+
+    Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up
+    to that part of the diffusion process.
+
+
+    Args:
+        num_diffusion_timesteps (`int`): the number of betas to produce.
+        max_beta (`float`): the maximum beta to use; use values lower than 1 to
+                     prevent singularities.
+        alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar.
+                     Choose from `cosine` or `exp`
+
+    Returns:
+        betas (`np.ndarray`): the betas used by the scheduler to step the model outputs
+    """
+    if alpha_transform_type == "cosine":
+
+        def alpha_bar_fn(t):
+            return math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2
+
+    elif alpha_transform_type == "exp":
+
+        def alpha_bar_fn(t):
+            return math.exp(t * -12.0)
+
+    else:
+        raise ValueError(f"Unsupported alpha_tranform_type: {alpha_transform_type}")
+
+    betas = []
+    for i in range(num_diffusion_timesteps):
+        t1 = i / num_diffusion_timesteps
+        t2 = (i + 1) / num_diffusion_timesteps
+        betas.append(min(1 - alpha_bar_fn(t2) / alpha_bar_fn(t1), max_beta))
+    return torch.tensor(betas, dtype=torch.float32)
+
+
+def rescale_zero_terminal_snr(betas):
+    """
+    Rescales betas to have zero terminal SNR Based on https://arxiv.org/pdf/2305.08891.pdf (Algorithm 1)
+
+
+    Args:
+        betas (`torch.FloatTensor`):
+            the betas that the scheduler is being initialized with.
+
+    Returns:
+        `torch.FloatTensor`: rescaled betas with zero terminal SNR
+    """
+    # Convert betas to alphas_bar_sqrt
+    alphas = 1.0 - betas
+    alphas_cumprod = torch.cumprod(alphas, dim=0)
+    alphas_bar_sqrt = alphas_cumprod.sqrt()
+
+    # Store old values.
+    alphas_bar_sqrt_0 = alphas_bar_sqrt[0].clone()
+    alphas_bar_sqrt_T = alphas_bar_sqrt[-1].clone()
+
+    # Shift so the last timestep is zero.
+    alphas_bar_sqrt -= alphas_bar_sqrt_T
+
+    # Scale so the first timestep is back to the old value.
+    alphas_bar_sqrt *= alphas_bar_sqrt_0 / (alphas_bar_sqrt_0 - alphas_bar_sqrt_T)
+
+    # Convert alphas_bar_sqrt to betas
+    alphas_bar = alphas_bar_sqrt**2  # Revert sqrt
+    alphas = alphas_bar[1:] / alphas_bar[:-1]  # Revert cumprod
+    alphas = torch.cat([alphas_bar[0:1], alphas])
+    betas = 1 - alphas
+
+    return betas
+
+
+class DDIMLMScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Denoising diffusion implicit models is a scheduler that extends the denoising procedure introduced in denoising
+    diffusion probabilistic models (DDPMs) with non-Markovian guidance.
+
+    [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
+    function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
+
+    For more details, see the original paper: https://arxiv.org/abs/2010.02502
+
+    Args:
+        num_train_timesteps (`int`): number of diffusion steps used to train the model.
+        beta_start (`float`): the starting `beta` value of inference.
+        beta_end (`float`): the final `beta` value.
+        beta_schedule (`str`):
+            the beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from
+            `linear`, `scaled_linear`, or `squaredcos_cap_v2`.
+        trained_betas (`np.ndarray`, optional):
+            option to pass an array of betas directly to the constructor to bypass `beta_start`, `beta_end` etc.
+        clip_sample (`bool`, default `True`):
+            option to clip predicted sample for numerical stability.
+        clip_sample_range (`float`, default `1.0`):
+            the maximum magnitude for sample clipping. Valid only when `clip_sample=True`.
+        set_alpha_to_one (`bool`, default `True`):
+            each diffusion step uses the value of alphas product at that step and at the previous one. For the final
+            step there is no previous alpha. When this option is `True` the previous alpha product is fixed to `1`,
+            otherwise it uses the value of alpha at step 0.
+        steps_offset (`int`, default `0`):
+            an offset added to the inference steps. You can use a combination of `offset=1` and
+            `set_alpha_to_one=False`, to make the last step use step 0 for the previous alpha product, as done in
+            stable diffusion.
+        prediction_type (`str`, default `epsilon`, optional):
+            prediction type of the scheduler function, one of `epsilon` (predicting the noise of the diffusion
+            process), `sample` (directly predicting the noisy sample`) or `v_prediction` (see section 2.4
+            https://imagen.research.google/video/paper.pdf)
+        thresholding (`bool`, default `False`):
+            whether to use the "dynamic thresholding" method (introduced by Imagen, https://arxiv.org/abs/2205.11487).
+            Note that the thresholding method is unsuitable for latent-space diffusion models (such as
+            stable-diffusion).
+        dynamic_thresholding_ratio (`float`, default `0.995`):
+            the ratio for the dynamic thresholding method. Default is `0.995`, the same as Imagen
+            (https://arxiv.org/abs/2205.11487). Valid only when `thresholding=True`.
+        sample_max_value (`float`, default `1.0`):
+            the threshold value for dynamic thresholding. Valid only when `thresholding=True`.
+        timestep_spacing (`str`, default `"leading"`):
+            The way the timesteps should be scaled. Refer to Table 2. of [Common Diffusion Noise Schedules and Sample
+            Steps are Flawed](https://arxiv.org/abs/2305.08891) for more information.
+        rescale_betas_zero_snr (`bool`, default `False`):
+            whether to rescale the betas to have zero terminal SNR (proposed by https://arxiv.org/pdf/2305.08891.pdf).
+            This can enable the model to generate very bright and dark samples instead of limiting it to samples with
+            medium brightness. Loosely related to
+            [`--offset_noise`](https://github.com/huggingface/diffusers/blob/74fd735eb073eb1d774b1ab4154a0876eb82f055/examples/dreambooth/train_dreambooth.py#L506).
+    """
+
+    _compatibles = [e.name for e in KarrasDiffusionSchedulers]
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.0001,
+        beta_end: float = 0.02,
+        beta_schedule: str = "linear",
+        trained_betas: Optional[Union[np.ndarray, List[float]]] = None,
+        clip_sample: bool = True,
+        set_alpha_to_one: bool = True,
+        steps_offset: int = 0,
+        prediction_type: str = "epsilon",
+        thresholding: bool = False,
+        dynamic_thresholding_ratio: float = 0.995,
+        clip_sample_range: float = 1.0,
+        sample_max_value: float = 1.0,
+        timestep_spacing: str = "leading",
+        rescale_betas_zero_snr: bool = False,
+        lamb: float = 1.0,
+        lm: bool = False,
+        kappa: float = 0.0,
+        freeze = 0.0,
+    ):
+        self.lamb = lamb
+        self.lm = lm
+        self.kappa = kappa
+        self.prev_noise = None
+        self.freeze = freeze
+        if trained_betas is not None:
+            self.betas = torch.tensor(trained_betas, dtype=torch.float32)
+        elif beta_schedule == "linear":
+            self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
+        elif beta_schedule == "scaled_linear":
+            # this schedule is very specific to the latent diffusion model.
+            self.betas = (
+                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
+            )
+        elif beta_schedule == "squaredcos_cap_v2":
+            # Glide cosine schedule
+            self.betas = betas_for_alpha_bar(num_train_timesteps)
+        else:
+            raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}")
+
+        # Rescale for zero SNR
+        if rescale_betas_zero_snr:
+            self.betas = rescale_zero_terminal_snr(self.betas)
+
+        self.alphas = 1.0 - self.betas
+        self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)
+
+        # At every step in ddim, we are looking into the previous alphas_cumprod
+        # For the final step, there is no previous alphas_cumprod because we are already at 0
+        # `set_alpha_to_one` decides whether we set this parameter simply to one or
+        # whether we use the final alpha of the "non-previous" one.
+        self.final_alpha_cumprod = torch.tensor(1.0) if set_alpha_to_one else self.alphas_cumprod[0]
+
+        # standard deviation of the initial noise distribution
+        self.init_noise_sigma = 1.0
+
+        # setable values
+        self.num_inference_steps = None
+        self.timesteps = torch.from_numpy(np.arange(0, num_train_timesteps)[::-1].copy().astype(np.int64))
+
+    def scale_model_input(self, sample: torch.FloatTensor, timestep: Optional[int] = None) -> torch.FloatTensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep.
+
+        Args:
+            sample (`torch.FloatTensor`): input sample
+            timestep (`int`, optional): current timestep
+
+        Returns:
+            `torch.FloatTensor`: scaled input sample
+        """
+        return sample
+
+    def _get_variance(self, timestep, prev_timestep):
+        alpha_prod_t = self.alphas_cumprod[timestep]
+        alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
+        beta_prod_t = 1 - alpha_prod_t
+        beta_prod_t_prev = 1 - alpha_prod_t_prev
+
+        variance = (beta_prod_t_prev / beta_prod_t) * (1 - alpha_prod_t / alpha_prod_t_prev)
+
+        return variance
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler._threshold_sample
+    def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
+        """
+        "Dynamic thresholding: At each sampling step we set s to a certain percentile absolute pixel value in xt0 (the
+        prediction of x_0 at timestep t), and if s > 1, then we threshold xt0 to the range [-s, s] and then divide by
+        s. Dynamic thresholding pushes saturated pixels (those near -1 and 1) inwards, thereby actively preventing
+        pixels from saturation at each step. We find that dynamic thresholding results in significantly better
+        photorealism as well as better image-text alignment, especially when using very large guidance weights."
+
+        https://arxiv.org/abs/2205.11487
+        """
+        dtype = sample.dtype
+        batch_size, channels, height, width = sample.shape
+
+        if dtype not in (torch.float32, torch.float64):
+            sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
+
+        # Flatten sample for doing quantile calculation along each image
+        sample = sample.reshape(batch_size, channels * height * width)
+
+        abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
+
+        s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
+        s = torch.clamp(
+            s, min=1, max=self.config.sample_max_value
+        )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
+
+        s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
+        sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
+
+        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.to(dtype)
+
+        return sample
+
+    def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.device] = None):
+        """
+        Sets the discrete timesteps used for the diffusion chain. Supporting function to be run before inference.
+
+        Args:
+            num_inference_steps (`int`):
+                the number of diffusion steps used when generating samples with a pre-trained model.
+        """
+
+        if num_inference_steps > self.config.num_train_timesteps:
+            raise ValueError(
+                f"`num_inference_steps`: {num_inference_steps} cannot be larger than `self.config.train_timesteps`:"
+                f" {self.config.num_train_timesteps} as the unet model trained with this scheduler can only handle"
+                f" maximal {self.config.num_train_timesteps} timesteps."
+            )
+
+        self.num_inference_steps = num_inference_steps
+
+        # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://arxiv.org/abs/2305.08891
+        if self.config.timestep_spacing == "linspace":
+            timesteps = (
+                np.linspace(0, self.config.num_train_timesteps - 1, num_inference_steps)
+                .round()[::-1]
+                .copy()
+                .astype(np.int64)
+            )
+        elif self.config.timestep_spacing == "leading":
+            step_ratio = self.config.num_train_timesteps // self.num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(np.int64)
+            timesteps += self.config.steps_offset
+        elif self.config.timestep_spacing == "trailing":
+            step_ratio = self.config.num_train_timesteps / self.num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = np.round(np.arange(self.config.num_train_timesteps, 0, -step_ratio)).astype(np.int64)
+            timesteps -= 1
+        else:
+            raise ValueError(
+                f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'leading' or 'trailing'."
+            )
+
+        self.timesteps = torch.from_numpy(timesteps).to(device)
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: int,
+        sample: torch.FloatTensor,
+        eta: float = 0.0,
+        use_clipped_model_output: bool = False,
+        generator=None,
+        variance_noise: Optional[torch.FloatTensor] = None,
+        return_dict: bool = True,
+    ) -> Union[DDIMLMSchedulerOutput, Tuple]:
+        """
+        Predict the sample at the previous timestep by reversing the SDE. Core function to propagate the diffusion
+        process from the learned model outputs (most often the predicted noise).
+
+        Args:
+            model_output (`torch.FloatTensor`): direct output from learned diffusion model.
+            timestep (`int`): current discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                current instance of sample being created by diffusion process.
+            eta (`float`): weight of noise for added noise in diffusion step.
+            use_clipped_model_output (`bool`): if `True`, compute "corrected" `model_output` from the clipped
+                predicted original sample. Necessary because predicted original sample is clipped to [-1, 1] when
+                `self.config.clip_sample` is `True`. If no clipping has happened, "corrected" `model_output` would
+                coincide with the one provided as input and `use_clipped_model_output` will have not effect.
+            generator: random number generator.
+            variance_noise (`torch.FloatTensor`): instead of generating noise for the variance using `generator`, we
+                can directly provide the noise for the variance itself. This is useful for methods such as
+                CycleDiffusion. (https://arxiv.org/abs/2210.05559)
+            return_dict (`bool`): option for returning tuple rather than DDIMSchedulerOutput class
+
+        Returns:
+            [`~schedulers.scheduling_utils.DDIMSchedulerOutput`] or `tuple`:
+            [`~schedulers.scheduling_utils.DDIMSchedulerOutput`] if `return_dict` is True, otherwise a `tuple`. When
+            returning a tuple, the first element is the sample tensor.
+
+        """
+        if self.num_inference_steps is None:
+            raise ValueError(
+                "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
+            )
+
+        # See formulas (12) and (16) of DDIM paper https://arxiv.org/pdf/2010.02502.pdf
+        # Ideally, read DDIM paper in-detail understanding
+
+        # Notation (<variable name> -> <name in paper>
+        # - pred_noise_t -> e_theta(x_t, t)
+        # - pred_original_sample -> f_theta(x_t, t) or x_0
+        # - std_dev_t -> sigma_t
+        # - eta -> η
+        # - pred_sample_direction -> "direction pointing to x_t"
+        # - pred_prev_sample -> "x_t-1"
+
+        # 1. get previous step value (=t-1)
+        prev_timestep = timestep - self.config.num_train_timesteps // self.num_inference_steps
+
+        # 2. compute alphas, betas
+        alpha_prod_t = self.alphas_cumprod[timestep]
+        alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
+
+        beta_prod_t = 1 - alpha_prod_t
+
+        # 3. compute predicted original sample from predicted noise also called
+        # "predicted x_0" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+        if self.config.prediction_type == "epsilon":
+            pred_original_sample = (sample - beta_prod_t ** (0.5) * model_output) / alpha_prod_t ** (0.5)
+            pred_epsilon = model_output
+        elif self.config.prediction_type == "sample":
+            pred_original_sample = model_output
+            pred_epsilon = (sample - alpha_prod_t ** (0.5) * pred_original_sample) / beta_prod_t ** (0.5)
+        elif self.config.prediction_type == "v_prediction":
+            pred_original_sample = (alpha_prod_t**0.5) * sample - (beta_prod_t**0.5) * model_output
+            pred_epsilon = (alpha_prod_t**0.5) * model_output + (beta_prod_t**0.5) * sample
+        else:
+            raise ValueError(
+                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or"
+                " `v_prediction`"
+            )
+
+        # 4. Clip or threshold "predicted x_0"
+        if self.config.thresholding:
+            pred_original_sample = self._threshold_sample(pred_original_sample)
+        elif self.config.clip_sample:
+            pred_original_sample = pred_original_sample.clamp(
+                -self.config.clip_sample_range, self.config.clip_sample_range
+            )
+
+        # 5. compute variance: "sigma_t(η)" -> see formula (16)
+        # σ_t = sqrt((1 − α_t−1)/(1 − α_t)) * sqrt(1 − α_t/α_t−1)
+        variance = self._get_variance(timestep, prev_timestep)
+        std_dev_t = eta * variance ** (0.5)
+
+        if use_clipped_model_output:
+            # the pred_epsilon is always re-derived from the clipped x_0 in Glide
+            pred_epsilon = (sample - alpha_prod_t ** (0.5) * pred_original_sample) / beta_prod_t ** (0.5)
+
+
+        # 5.2 LM corection
+        noise_pred = pred_epsilon
+        noise_pred_ema = noise_pred
+        if self.prev_noise is not None:
+            noise_pred_ema = self.kappa * self.prev_noise + (1 - self.kappa) * noise_pred
+        norm_squared = (noise_pred * noise_pred).sum(dim=(1, 2, 3))
+        norm_squared = norm_squared.unsqueeze(1).unsqueeze(2).unsqueeze(3)
+        part1 = noise_pred
+
+        norm_squared_ema = (noise_pred_ema * noise_pred_ema).sum(dim=(1, 2, 3))
+        norm_squared_ema = norm_squared_ema.unsqueeze(1).unsqueeze(2).unsqueeze(3)
+
+        inner_product = torch.sum(noise_pred * noise_pred_ema, dim=(1, 2, 3))
+        mp = noise_pred_ema * inner_product.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
+        # coef_2 = (sigma_t * norm_squared)/(lamb*(lamb+norm_squared_ema))
+        part2 = mp / (self.lamb + norm_squared_ema)
+
+        inversed_pred = part1 - part2
+
+        # normalize the direction
+        norm = torch.sqrt(norm_squared)
+        norm_squared_lm = (inversed_pred * inversed_pred).sum(dim=(1, 2, 3))
+        norm_squared_lm = norm_squared_lm.unsqueeze(1).unsqueeze(2).unsqueeze(3)
+        norm_lm = torch.sqrt(norm_squared_lm)
+        inversed_pred = inversed_pred * norm / norm_lm
+
+        # 6. compute "direction pointing to x_t" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+        print(timestep)
+        if timestep < 1000 - 1000 * self.freeze and self.lm is True:
+            pred_sample_direction = (1 - alpha_prod_t_prev - std_dev_t**2) ** (0.5) * inversed_pred
+            print("do lm")
+        else:
+            pred_sample_direction = (1 - alpha_prod_t_prev - std_dev_t ** 2) ** (0.5) * pred_epsilon
+            print("freezed")
+
+        # 7. compute x_t without "random noise" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+        prev_sample = alpha_prod_t_prev ** (0.5) * pred_original_sample + pred_sample_direction
+        self.prev_noise = pred_epsilon
+        if eta > 0:
+            if variance_noise is not None and generator is not None:
+                raise ValueError(
+                    "Cannot pass both generator and variance_noise. Please make sure that either `generator` or"
+                    " `variance_noise` stays `None`."
+                )
+
+            if variance_noise is None:
+                variance_noise = randn_tensor(
+                    model_output.shape, generator=generator, device=model_output.device, dtype=model_output.dtype
+                )
+            variance = std_dev_t * variance_noise
+
+            prev_sample = prev_sample + variance
+
+        if not return_dict:
+            return (prev_sample,)
+
+        return DDIMLMSchedulerOutput(prev_sample=prev_sample, pred_original_sample=pred_original_sample)
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.FloatTensor,
+        noise: torch.FloatTensor,
+        timesteps: torch.IntTensor,
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as original_samples
+        alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype)
+        timesteps = timesteps.to(original_samples.device)
+
+        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
+        return noisy_samples
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.get_velocity
+    def get_velocity(
+        self, sample: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as sample
+        alphas_cumprod = self.alphas_cumprod.to(device=sample.device, dtype=sample.dtype)
+        timesteps = timesteps.to(sample.device)
+
+        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(sample.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(sample.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        velocity = sqrt_alpha_prod * noise - sqrt_one_minus_alpha_prod * sample
+        return velocity
+
+    def __len__(self):
+        return self.config.num_train_timesteps

scheduler/scheduling_dpmsolver_multistep_lm.py

@@ -0,0 +1,841 @@
+# Copyright 2025 ZJU Lab304 Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# DISCLAIMER: This file is strongly influenced by https://github.com/LuChengTHU/dpm-solver
+
+import math
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from diffusers.configuration_utils import ConfigMixin, register_to_config
+from  diffusers.utils.torch_utils import randn_tensor
+from diffusers.schedulers.scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin, SchedulerOutput
+
+def lm_correct(prev_noise, noise_pred, lamb, kappa):
+    # print('entered lmc')
+    if prev_noise is not None:
+        noise_pred_ema = kappa * prev_noise + (1 - kappa) * noise_pred
+    else:
+        noise_pred_ema = noise_pred
+    # lm step for noise
+    norm_squared = (noise_pred * noise_pred).sum(dim=(1, 2, 3))
+    norm_squared = norm_squared.unsqueeze(1).unsqueeze(2).unsqueeze(3)
+    part1 =  noise_pred
+
+    norm_squared_ema = (noise_pred_ema * noise_pred_ema).sum(dim=(1, 2, 3))
+    norm_squared_ema = norm_squared_ema.unsqueeze(1).unsqueeze(2).unsqueeze(3)
+
+    inner_product = torch.sum(noise_pred * noise_pred_ema, dim=(1, 2, 3))
+    mp = noise_pred_ema * inner_product.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
+    part2 = mp /  (lamb + norm_squared_ema)
+
+    inversed_pred = part1 - part2
+
+    # normalize the direction
+    norm = torch.sqrt(norm_squared)
+    norm_squared_lm = (inversed_pred * inversed_pred).sum(dim=(1, 2, 3))
+    norm_squared_lm = norm_squared_lm.unsqueeze(1).unsqueeze(2).unsqueeze(3)
+    norm_lm = torch.sqrt(norm_squared_lm)
+    inversed_pred = inversed_pred * norm / norm_lm
+    return inversed_pred
+
+
+# Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
+def betas_for_alpha_bar(
+    num_diffusion_timesteps,
+    max_beta=0.999,
+    alpha_transform_type="cosine",
+):
+    """
+    Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of
+    (1-beta) over time from t = [0,1].
+
+    Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up
+    to that part of the diffusion process.
+
+
+    Args:
+        num_diffusion_timesteps (`int`): the number of betas to produce.
+        max_beta (`float`): the maximum beta to use; use values lower than 1 to
+                     prevent singularities.
+        alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar.
+                     Choose from `cosine` or `exp`
+
+    Returns:
+        betas (`np.ndarray`): the betas used by the scheduler to step the model outputs
+    """
+    if alpha_transform_type == "cosine":
+
+        def alpha_bar_fn(t):
+            return math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2
+
+    elif alpha_transform_type == "exp":
+
+        def alpha_bar_fn(t):
+            return math.exp(t * -12.0)
+
+    else:
+        raise ValueError(f"Unsupported alpha_tranform_type: {alpha_transform_type}")
+
+    betas = []
+    for i in range(num_diffusion_timesteps):
+        t1 = i / num_diffusion_timesteps
+        t2 = (i + 1) / num_diffusion_timesteps
+        betas.append(min(1 - alpha_bar_fn(t2) / alpha_bar_fn(t1), max_beta))
+    return torch.tensor(betas, dtype=torch.float32)
+
+
+class DPMSolverMultistepLMScheduler(SchedulerMixin, ConfigMixin):
+    """
+    DPM-Solver (and the improved version DPM-Solver++) is a fast dedicated high-order solver for diffusion ODEs with
+    the convergence order guarantee. Empirically, sampling by DPM-Solver with only 20 steps can generate high-quality
+    samples, and it can generate quite good samples even in only 10 steps.
+
+    For more details, see the original paper: https://arxiv.org/abs/2206.00927 and https://arxiv.org/abs/2211.01095
+
+    Currently, we support the multistep DPM-Solver for both noise prediction models and data prediction models. We
+    recommend to use `solver_order=2` for guided sampling, and `solver_order=3` for unconditional sampling.
+
+    We also support the "dynamic thresholding" method in Imagen (https://arxiv.org/abs/2205.11487). For pixel-space
+    diffusion models, you can set both `algorithm_type="dpmsolver++"` and `thresholding=True` to use the dynamic
+    thresholding. Note that the thresholding method is unsuitable for latent-space diffusion models (such as
+    stable-diffusion).
+
+    We also support the SDE variant of DPM-Solver and DPM-Solver++, which is a fast SDE solver for the reverse
+    diffusion SDE. Currently we only support the first-order and second-order solvers. We recommend using the
+    second-order `sde-dpmsolver++`.
+
+    [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
+    function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
+    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
+    [`~SchedulerMixin.from_pretrained`] functions.
+
+    Args:
+        num_train_timesteps (`int`): number of diffusion steps used to train the model.
+        beta_start (`float`): the starting `beta` value of inference.
+        beta_end (`float`): the final `beta` value.
+        beta_schedule (`str`):
+            the beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from
+            `linear`, `scaled_linear`, or `squaredcos_cap_v2`.
+        trained_betas (`np.ndarray`, optional):
+            option to pass an array of betas directly to the constructor to bypass `beta_start`, `beta_end` etc.
+        solver_order (`int`, default `2`):
+            the order of DPM-Solver; can be `1` or `2` or `3`. We recommend to use `solver_order=2` for guided
+            sampling, and `solver_order=3` for unconditional sampling.
+        prediction_type (`str`, default `epsilon`, optional):
+            prediction type of the scheduler function, one of `epsilon` (predicting the noise of the diffusion
+            process), `sample` (directly predicting the noisy sample`) or `v_prediction` (see section 2.4
+            https://imagen.research.google/video/paper.pdf)
+        thresholding (`bool`, default `False`):
+            whether to use the "dynamic thresholding" method (introduced by Imagen, https://arxiv.org/abs/2205.11487).
+            For pixel-space diffusion models, you can set both `algorithm_type=dpmsolver++` and `thresholding=True` to
+            use the dynamic thresholding. Note that the thresholding method is unsuitable for latent-space diffusion
+            models (such as stable-diffusion).
+        dynamic_thresholding_ratio (`float`, default `0.995`):
+            the ratio for the dynamic thresholding method. Default is `0.995`, the same as Imagen
+            (https://arxiv.org/abs/2205.11487).
+        sample_max_value (`float`, default `1.0`):
+            the threshold value for dynamic thresholding. Valid only when `thresholding=True` and
+            `algorithm_type="dpmsolver++`.
+        algorithm_type (`str`, default `dpmsolver++`):
+            the algorithm type for the solver. Either `dpmsolver` or `dpmsolver++` or `sde-dpmsolver` or
+            `sde-dpmsolver++`. The `dpmsolver` type implements the algorithms in https://arxiv.org/abs/2206.00927, and
+            the `dpmsolver++` type implements the algorithms in https://arxiv.org/abs/2211.01095. We recommend to use
+            `dpmsolver++` or `sde-dpmsolver++` with `solver_order=2` for guided sampling (e.g. stable-diffusion).
+        solver_type (`str`, default `midpoint`):
+            the solver type for the second-order solver. Either `midpoint` or `heun`. The solver type slightly affects
+            the sample quality, especially for small number of steps. We empirically find that `midpoint` solvers are
+            slightly better, so we recommend to use the `midpoint` type.
+        lower_order_final (`bool`, default `True`):
+            whether to use lower-order solvers in the final steps. Only valid for < 15 inference steps. We empirically
+            find this trick can stabilize the sampling of DPM-Solver for steps < 15, especially for steps <= 10.
+        use_karras_sigmas (`bool`, *optional*, defaults to `False`):
+             This parameter controls whether to use Karras sigmas (Karras et al. (2022) scheme) for step sizes in the
+             noise schedule during the sampling process. If True, the sigmas will be determined according to a sequence
+             of noise levels {σi} as defined in Equation (5) of the paper https://arxiv.org/pdf/2206.00364.pdf.
+        lambda_min_clipped (`float`, default `-inf`):
+            the clipping threshold for the minimum value of lambda(t) for numerical stability. This is critical for
+            cosine (squaredcos_cap_v2) noise schedule.
+        variance_type (`str`, *optional*):
+            Set to "learned" or "learned_range" for diffusion models that predict variance. For example, OpenAI's
+            guided-diffusion (https://github.com/openai/guided-diffusion) predicts both mean and variance of the
+            Gaussian distribution in the model's output. DPM-Solver only needs the "mean" output because it is based on
+            diffusion ODEs. whether the model's output contains the predicted Gaussian variance. For example, OpenAI's
+            guided-diffusion (https://github.com/openai/guided-diffusion) predicts both mean and variance of the
+            Gaussian distribution in the model's output. DPM-Solver only needs the "mean" output because it is based on
+            diffusion ODEs.
+        timestep_spacing (`str`, default `"linspace"`):
+            The way the timesteps should be scaled. Refer to Table 2. of [Common Diffusion Noise Schedules and Sample
+            Steps are Flawed](https://arxiv.org/abs/2305.08891) for more information.
+        steps_offset (`int`, default `0`):
+            an offset added to the inference steps. You can use a combination of `offset=1` and
+            `set_alpha_to_one=False`, to make the last step use step 0 for the previous alpha product, as done in
+            stable diffusion.
+    """
+
+    _compatibles = [e.name for e in KarrasDiffusionSchedulers]
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        beta_start: float = 0.0001,
+        beta_end: float = 0.02,
+        beta_schedule: str = "linear",
+        trained_betas: Optional[Union[np.ndarray, List[float]]] = None,
+        solver_order: int = 2,
+        prediction_type: str = "epsilon",
+        thresholding: bool = False,
+        dynamic_thresholding_ratio: float = 0.995,
+        sample_max_value: float = 1.0,
+        algorithm_type: str = "dpmsolver++",
+        solver_type: str = "midpoint",
+        lower_order_final: bool = True,
+        use_karras_sigmas: Optional[bool] = False,
+        lambda_min_clipped: float = -float("inf"),
+        variance_type: Optional[str] = None,
+        timestep_spacing: str = "linspace",
+        steps_offset: int = 0,
+        lamb:float =1.0,
+        lm:bool=False,
+        kappa: float = 0.0,
+    ):
+        if trained_betas is not None:
+            self.betas = torch.tensor(trained_betas, dtype=torch.float32)
+        elif beta_schedule == "linear":
+            self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
+        elif beta_schedule == "scaled_linear":
+            # this schedule is very specific to the latent diffusion model.
+            self.betas = (
+                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
+            )
+        elif beta_schedule == "squaredcos_cap_v2":
+            # Glide cosine schedule
+            self.betas = betas_for_alpha_bar(num_train_timesteps)
+        else:
+            raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}")
+
+        self.lamb = lamb
+        self.lm = lm
+        self.kappa = kappa
+        self.prev_noise = None
+        self.alphas = 1.0 - self.betas
+        self.alphas_cumprod = torch.cumprod(self.alphas, dim=0)
+        # Currently we only support VP-type noise schedule
+        self.alpha_t = torch.sqrt(self.alphas_cumprod)
+        self.sigma_t = torch.sqrt(1 - self.alphas_cumprod)
+        self.lambda_t = torch.log(self.alpha_t) - torch.log(self.sigma_t)
+
+        # standard deviation of the initial noise distribution
+        self.init_noise_sigma = 1.0
+
+        # settings for DPM-Solver
+        if algorithm_type not in ["dpmsolver", "dpmsolver++", "sde-dpmsolver", "sde-dpmsolver++"]:
+            if algorithm_type == "deis":
+                self.register_to_config(algorithm_type="dpmsolver++")
+            else:
+                raise NotImplementedError(f"{algorithm_type} does is not implemented for {self.__class__}")
+
+        if solver_type not in ["midpoint", "heun"]:
+            if solver_type in ["logrho", "bh1", "bh2"]:
+                self.register_to_config(solver_type="midpoint")
+            else:
+                raise NotImplementedError(f"{solver_type} does is not implemented for {self.__class__}")
+
+        # setable values
+        self.num_inference_steps = None
+        timesteps = np.linspace(0, num_train_timesteps - 1, num_train_timesteps, dtype=np.float32)[::-1].copy()
+        self.timesteps = torch.from_numpy(timesteps)
+        self.model_outputs = [None] * solver_order
+        self.lower_order_nums = 0
+
+    def set_timesteps(self, num_inference_steps: int = None, device: Union[str, torch.device] = None):
+        """
+        Sets the timesteps used for the diffusion chain. Supporting function to be run before inference.
+
+        Args:
+            num_inference_steps (`int`):
+                the number of diffusion steps used when generating samples with a pre-trained model.
+            device (`str` or `torch.device`, optional):
+                the device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        """
+        # Clipping the minimum of all lambda(t) for numerical stability.
+        # This is critical for cosine (squaredcos_cap_v2) noise schedule.
+        clipped_idx = torch.searchsorted(torch.flip(self.lambda_t, [0]), self.config.lambda_min_clipped)
+        last_timestep = ((self.config.num_train_timesteps - clipped_idx).numpy()).item()
+
+        # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://arxiv.org/abs/2305.08891
+        if self.config.timestep_spacing == "linspace":
+            timesteps = (
+                np.linspace(0, last_timestep - 1, num_inference_steps + 1).round()[::-1][:-1].copy().astype(np.int64)
+            )
+        elif self.config.timestep_spacing == "leading":
+            step_ratio = last_timestep // (num_inference_steps + 1)
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = (np.arange(0, num_inference_steps + 1) * step_ratio).round()[::-1][:-1].copy().astype(np.int64)
+            timesteps += self.config.steps_offset
+        elif self.config.timestep_spacing == "trailing":
+            step_ratio = self.config.num_train_timesteps / num_inference_steps
+            # creates integer timesteps by multiplying by ratio
+            # casting to int to avoid issues when num_inference_step is power of 3
+            timesteps = np.arange(last_timestep, 0, -step_ratio).round().copy().astype(np.int64)
+            timesteps -= 1
+        else:
+            raise ValueError(
+                f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'."
+            )
+
+        sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5)
+        if self.config.use_karras_sigmas:
+            log_sigmas = np.log(sigmas)
+            sigmas = self._convert_to_karras(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+            timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas]).round()
+            timesteps = np.flip(timesteps).copy().astype(np.int64)
+
+        self.sigmas = torch.from_numpy(sigmas)
+
+        # when num_inference_steps == num_train_timesteps, we can end up with
+        # duplicates in timesteps.
+        _, unique_indices = np.unique(timesteps, return_index=True)
+        timesteps = timesteps[np.sort(unique_indices)]
+
+        self.timesteps = torch.from_numpy(timesteps).to(device)
+
+        self.num_inference_steps = len(timesteps)
+
+        self.model_outputs = [
+            None,
+        ] * self.config.solver_order
+        self.lower_order_nums = 0
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler._threshold_sample
+    def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor:
+        """
+        "Dynamic thresholding: At each sampling step we set s to a certain percentile absolute pixel value in xt0 (the
+        prediction of x_0 at timestep t), and if s > 1, then we threshold xt0 to the range [-s, s] and then divide by
+        s. Dynamic thresholding pushes saturated pixels (those near -1 and 1) inwards, thereby actively preventing
+        pixels from saturation at each step. We find that dynamic thresholding results in significantly better
+        photorealism as well as better image-text alignment, especially when using very large guidance weights."
+
+        https://arxiv.org/abs/2205.11487
+        """
+        dtype = sample.dtype
+        batch_size, channels, height, width = sample.shape
+
+        if dtype not in (torch.float32, torch.float64):
+            sample = sample.float()  # upcast for quantile calculation, and clamp not implemented for cpu half
+
+        # Flatten sample for doing quantile calculation along each image
+        sample = sample.reshape(batch_size, channels * height * width)
+
+        abs_sample = sample.abs()  # "a certain percentile absolute pixel value"
+
+        s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1)
+        s = torch.clamp(
+            s, min=1, max=self.config.sample_max_value
+        )  # When clamped to min=1, equivalent to standard clipping to [-1, 1]
+
+        s = s.unsqueeze(1)  # (batch_size, 1) because clamp will broadcast along dim=0
+        sample = torch.clamp(sample, -s, s) / s  # "we threshold xt0 to the range [-s, s] and then divide by s"
+
+        sample = sample.reshape(batch_size, channels, height, width)
+        sample = sample.to(dtype)
+
+        return sample
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._sigma_to_t
+    def _sigma_to_t(self, sigma, log_sigmas):
+        # get log sigma
+        log_sigma = np.log(sigma)
+
+        # get distribution
+        dists = log_sigma - log_sigmas[:, np.newaxis]
+
+        # get sigmas range
+        low_idx = np.cumsum((dists >= 0), axis=0).argmax(axis=0).clip(max=log_sigmas.shape[0] - 2)
+        high_idx = low_idx + 1
+
+        low = log_sigmas[low_idx]
+        high = log_sigmas[high_idx]
+
+        # interpolate sigmas
+        w = (low - log_sigma) / (low - high)
+        w = np.clip(w, 0, 1)
+
+        # transform interpolation to time range
+        t = (1 - w) * low_idx + w * high_idx
+        t = t.reshape(sigma.shape)
+        return t
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_karras
+    def _convert_to_karras(self, in_sigmas: torch.FloatTensor, num_inference_steps) -> torch.FloatTensor:
+        """Constructs the noise schedule of Karras et al. (2022)."""
+
+        sigma_min: float = in_sigmas[-1].item()
+        sigma_max: float = in_sigmas[0].item()
+
+        rho = 7.0  # 7.0 is the value used in the paper
+        ramp = np.linspace(0, 1, num_inference_steps)
+        min_inv_rho = sigma_min ** (1 / rho)
+        max_inv_rho = sigma_max ** (1 / rho)
+        sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+        return sigmas
+
+    def convert_model_output(
+        self, model_output: torch.FloatTensor, timestep: int, sample: torch.FloatTensor
+    ) -> torch.FloatTensor:
+        """
+        Convert the model output to the corresponding type that the algorithm (DPM-Solver / DPM-Solver++) needs.
+
+        DPM-Solver is designed to discretize an integral of the noise prediction model, and DPM-Solver++ is designed to
+        discretize an integral of the data prediction model. So we need to first convert the model output to the
+        corresponding type to match the algorithm.
+
+        Note that the algorithm type and the model type is decoupled. That is to say, we can use either DPM-Solver or
+        DPM-Solver++ for both noise prediction model and data prediction model.
+
+        Args:
+            model_output (`torch.FloatTensor`): direct output from learned diffusion model.
+            timestep (`int`): current discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                current instance of sample being created by diffusion process.
+
+        Returns:
+            `torch.FloatTensor`: the converted model output.
+        """
+
+        # DPM-Solver++ needs to solve an integral of the data prediction model.
+        if self.config.algorithm_type in ["dpmsolver++", "sde-dpmsolver++"]:
+            if self.config.prediction_type == "epsilon":
+                # DPM-Solver and DPM-Solver++ only need the "mean" output.
+                if self.config.variance_type in ["learned", "learned_range"]:
+                    model_output = model_output[:, :3]
+                alpha_t, sigma_t = self.alpha_t[timestep], self.sigma_t[timestep]
+                x0_pred = (sample - sigma_t * model_output) / alpha_t
+            elif self.config.prediction_type == "sample":
+                x0_pred = model_output
+            elif self.config.prediction_type == "v_prediction":
+                alpha_t, sigma_t = self.alpha_t[timestep], self.sigma_t[timestep]
+                x0_pred = alpha_t * sample - sigma_t * model_output
+            else:
+                raise ValueError(
+                    f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or"
+                    " `v_prediction` for the DPMSolverMultistepScheduler."
+                )
+
+            if self.config.thresholding:
+                x0_pred = self._threshold_sample(x0_pred)
+
+            return x0_pred
+
+        # DPM-Solver needs to solve an integral of the noise prediction model.
+        elif self.config.algorithm_type in ["dpmsolver", "sde-dpmsolver"]:
+            if self.config.prediction_type == "epsilon":
+                # DPM-Solver and DPM-Solver++ only need the "mean" output.
+                if self.config.variance_type in ["learned", "learned_range"]:
+                    epsilon = model_output[:, :3]
+                else:
+                    epsilon = model_output
+            elif self.config.prediction_type == "sample":
+                alpha_t, sigma_t = self.alpha_t[timestep], self.sigma_t[timestep]
+                epsilon = (sample - alpha_t * model_output) / sigma_t
+            elif self.config.prediction_type == "v_prediction":
+                alpha_t, sigma_t = self.alpha_t[timestep], self.sigma_t[timestep]
+                epsilon = alpha_t * model_output + sigma_t * sample
+            else:
+                raise ValueError(
+                    f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or"
+                    " `v_prediction` for the DPMSolverMultistepScheduler."
+                )
+
+            if self.config.thresholding:
+                alpha_t, sigma_t = self.alpha_t[timestep], self.sigma_t[timestep]
+                x0_pred = (sample - sigma_t * epsilon) / alpha_t
+                x0_pred = self._threshold_sample(x0_pred)
+                epsilon = (sample - alpha_t * x0_pred) / sigma_t
+
+            return epsilon
+
+    def dpm_solver_first_order_update(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: int,
+        prev_timestep: int,
+        sample: torch.FloatTensor,
+        noise: Optional[torch.FloatTensor] = None,
+        lamb: float = 1.0,
+        lm=True,
+    ) -> torch.FloatTensor:
+        """
+        One step for the first-order DPM-Solver (equivalent to DDIM).
+
+        See https://arxiv.org/abs/2206.00927 for the detailed derivation.
+
+        Args:
+            model_output (`torch.FloatTensor`): direct output from learned diffusion model.
+            timestep (`int`): current discrete timestep in the diffusion chain.
+            prev_timestep (`int`): previous discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                current instance of sample being created by diffusion process.
+
+        Returns:
+            `torch.FloatTensor`: the sample tensor at the previous timestep.
+        """
+        lambda_t, lambda_s = self.lambda_t[prev_timestep], self.lambda_t[timestep]
+        alpha_t, alpha_s = self.alpha_t[prev_timestep], self.alpha_t[timestep]
+        sigma_t, sigma_s = self.sigma_t[prev_timestep], self.sigma_t[timestep]
+        h = lambda_t - lambda_s
+        if self.config.algorithm_type == "dpmsolver++":
+            # x_t = (sigma_t / sigma_s) * sample - (alpha_t * (torch.exp(-h) - 1.0)) * model_output
+            noise = - (alpha_t * (torch.exp(-h) - 1.0)) * model_output
+            if lm is True:
+                x_t = (sigma_t / sigma_s) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+            else:
+                x_t = (sigma_t / sigma_s) * sample + noise
+            self.prev_noise = noise
+        elif self.config.algorithm_type == "dpmsolver":
+            # x_t = (alpha_t / alpha_s) * sample - (sigma_t * (torch.exp(h) - 1.0)) * model_output
+            noise = - (sigma_t * (torch.exp(h) - 1.0)) * model_output
+            if lm is True:
+                x_t = (alpha_t / alpha_s) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+            else:
+                x_t = (alpha_t / alpha_s) * sample + noise
+            self.prev_noise = noise
+        elif self.config.algorithm_type == "sde-dpmsolver++":
+            assert noise is not None
+            x_t = (
+                (sigma_t / sigma_s * torch.exp(-h)) * sample
+                + (alpha_t * (1 - torch.exp(-2.0 * h))) * model_output
+                + sigma_t * torch.sqrt(1.0 - torch.exp(-2 * h)) * noise
+            )
+        elif self.config.algorithm_type == "sde-dpmsolver":
+            assert noise is not None
+            x_t = (
+                (alpha_t / alpha_s) * sample
+                - 2.0 * (sigma_t * (torch.exp(h) - 1.0)) * model_output
+                + sigma_t * torch.sqrt(torch.exp(2 * h) - 1.0) * noise
+            )
+        return x_t
+
+    def multistep_dpm_solver_second_order_update(
+        self,
+        model_output_list: List[torch.FloatTensor],
+        timestep_list: List[int],
+        prev_timestep: int,
+        sample: torch.FloatTensor,
+        noise: Optional[torch.FloatTensor] = None,
+        lamb: float = 1.0,
+        lm=True,
+    ) -> torch.FloatTensor:
+        """
+        One step for the second-order multistep DPM-Solver.
+
+        Args:
+            model_output_list (`List[torch.FloatTensor]`):
+                direct outputs from learned diffusion model at current and latter timesteps.
+            timestep (`int`): current and latter discrete timestep in the diffusion chain.
+            prev_timestep (`int`): previous discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                current instance of sample being created by diffusion process.
+
+        Returns:
+            `torch.FloatTensor`: the sample tensor at the previous timestep.
+        """
+        t, s0, s1 = prev_timestep, timestep_list[-1], timestep_list[-2]
+        m0, m1 = model_output_list[-1], model_output_list[-2]
+        lambda_t, lambda_s0, lambda_s1 = self.lambda_t[t], self.lambda_t[s0], self.lambda_t[s1]
+        alpha_t, alpha_s0 = self.alpha_t[t], self.alpha_t[s0]
+        sigma_t, sigma_s0 = self.sigma_t[t], self.sigma_t[s0]
+        h, h_0 = lambda_t - lambda_s0, lambda_s0 - lambda_s1
+        r0 = h_0 / h
+        D0, D1 = m0, (1.0 / r0) * (m0 - m1)
+        if self.config.algorithm_type == "dpmsolver++":
+            # See https://arxiv.org/abs/2211.01095 for detailed derivations
+            if self.config.solver_type == "midpoint":
+                # x_t = (
+                #     (sigma_t / sigma_s0) * sample
+                #     - (alpha_t * (torch.exp(-h) - 1.0)) * D0
+                #     - 0.5 * (alpha_t * (torch.exp(-h) - 1.0)) * D1
+                # )
+                noise = - (alpha_t * (torch.exp(-h) - 1.0)) * D0 - 0.5 * (alpha_t * (torch.exp(-h) - 1.0)) * D1
+                if lm is True:
+                    x_t = (sigma_t / sigma_s0) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+                else:
+                    x_t = (sigma_t / sigma_s0) * sample + noise
+                self.prev_noise = noise
+            elif self.config.solver_type == "heun":
+                # x_t = (
+                #     (sigma_t / sigma_s0) * sample
+                #     - (alpha_t * (torch.exp(-h) - 1.0)) * D0
+                #     + (alpha_t * ((torch.exp(-h) - 1.0) / h + 1.0)) * D1
+                # )
+                noise = - (alpha_t * (torch.exp(-h) - 1.0)) * D0 + (alpha_t * ((torch.exp(-h) - 1.0) / h + 1.0)) * D1
+                if lm is True:
+                    x_t = (sigma_t / sigma_s0) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+                else:
+                    x_t = (sigma_t / sigma_s0) * sample + noise
+                self.prev_noise = noise
+        elif self.config.algorithm_type == "dpmsolver":
+            # See https://arxiv.org/abs/2206.00927 for detailed derivations
+            if self.config.solver_type == "midpoint":
+                # x_t = (
+                #     (alpha_t / alpha_s0) * sample
+                #     - (sigma_t * (torch.exp(h) - 1.0)) * D0
+                #     - 0.5 * (sigma_t * (torch.exp(h) - 1.0)) * D1
+                # )
+                noise = - (sigma_t * (torch.exp(h) - 1.0)) * D0 - 0.5 * (sigma_t * (torch.exp(h) - 1.0)) * D1
+                if lm is True:
+                    x_t = (alpha_t / alpha_s0) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+                else:
+                    x_t = (alpha_t / alpha_s0) * sample + noise
+                self.prev_noise = noise
+            elif self.config.solver_type == "heun":
+                # x_t = (
+                #     (alpha_t / alpha_s0) * sample
+                #     - (sigma_t * (torch.exp(h) - 1.0)) * D0
+                #     - (sigma_t * ((torch.exp(h) - 1.0) / h - 1.0)) * D1
+                # )
+                noise = - (sigma_t * (torch.exp(h) - 1.0)) * D0 - (sigma_t * ((torch.exp(h) - 1.0) / h - 1.0)) * D1
+                if lm is True:
+                    x_t = (alpha_t / alpha_s0) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+                else:
+                    x_t = (alpha_t / alpha_s0) * sample + noise
+                self.prev_noise = noise
+        elif self.config.algorithm_type == "sde-dpmsolver++":
+            assert noise is not None
+            if self.config.solver_type == "midpoint":
+                x_t = (
+                    (sigma_t / sigma_s0 * torch.exp(-h)) * sample
+                    + (alpha_t * (1 - torch.exp(-2.0 * h))) * D0
+                    + 0.5 * (alpha_t * (1 - torch.exp(-2.0 * h))) * D1
+                    + sigma_t * torch.sqrt(1.0 - torch.exp(-2 * h)) * noise
+                )
+            elif self.config.solver_type == "heun":
+                x_t = (
+                    (sigma_t / sigma_s0 * torch.exp(-h)) * sample
+                    + (alpha_t * (1 - torch.exp(-2.0 * h))) * D0
+                    + (alpha_t * ((1.0 - torch.exp(-2.0 * h)) / (-2.0 * h) + 1.0)) * D1
+                    + sigma_t * torch.sqrt(1.0 - torch.exp(-2 * h)) * noise
+                )
+        elif self.config.algorithm_type == "sde-dpmsolver":
+            assert noise is not None
+            if self.config.solver_type == "midpoint":
+                x_t = (
+                    (alpha_t / alpha_s0) * sample
+                    - 2.0 * (sigma_t * (torch.exp(h) - 1.0)) * D0
+                    - (sigma_t * (torch.exp(h) - 1.0)) * D1
+                    + sigma_t * torch.sqrt(torch.exp(2 * h) - 1.0) * noise
+                )
+            elif self.config.solver_type == "heun":
+                x_t = (
+                    (alpha_t / alpha_s0) * sample
+                    - 2.0 * (sigma_t * (torch.exp(h) - 1.0)) * D0
+                    - 2.0 * (sigma_t * ((torch.exp(h) - 1.0) / h - 1.0)) * D1
+                    + sigma_t * torch.sqrt(torch.exp(2 * h) - 1.0) * noise
+                )
+        return x_t
+
+    def multistep_dpm_solver_third_order_update(
+        self,
+        model_output_list: List[torch.FloatTensor],
+        timestep_list: List[int],
+        prev_timestep: int,
+        sample: torch.FloatTensor,
+        lamb:float = 1.0,
+        lm = True,
+    ) -> torch.FloatTensor:
+        """
+        One step for the third-order multistep DPM-Solver.
+
+        Args:
+            model_output_list (`List[torch.FloatTensor]`):
+                direct outputs from learned diffusion model at current and latter timesteps.
+            timestep (`int`): current and latter discrete timestep in the diffusion chain.
+            prev_timestep (`int`): previous discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                current instance of sample being created by diffusion process.
+
+        Returns:
+            `torch.FloatTensor`: the sample tensor at the previous timestep.
+        """
+        t, s0, s1, s2 = prev_timestep, timestep_list[-1], timestep_list[-2], timestep_list[-3]
+        m0, m1, m2 = model_output_list[-1], model_output_list[-2], model_output_list[-3]
+        lambda_t, lambda_s0, lambda_s1, lambda_s2 = (
+            self.lambda_t[t],
+            self.lambda_t[s0],
+            self.lambda_t[s1],
+            self.lambda_t[s2],
+        )
+        alpha_t, alpha_s0 = self.alpha_t[t], self.alpha_t[s0]
+        sigma_t, sigma_s0 = self.sigma_t[t], self.sigma_t[s0]
+        h, h_0, h_1 = lambda_t - lambda_s0, lambda_s0 - lambda_s1, lambda_s1 - lambda_s2
+        r0, r1 = h_0 / h, h_1 / h
+        D0 = m0
+        D1_0, D1_1 = (1.0 / r0) * (m0 - m1), (1.0 / r1) * (m1 - m2)
+        D1 = D1_0 + (r0 / (r0 + r1)) * (D1_0 - D1_1)
+        D2 = (1.0 / (r0 + r1)) * (D1_0 - D1_1)
+        if self.config.algorithm_type == "dpmsolver++":
+            # See https://arxiv.org/abs/2206.00927 for detailed derivations
+            # x_t = (
+            #     (sigma_t / sigma_s0) * sample
+            #     - (alpha_t * (torch.exp(-h) - 1.0)) * D0
+            #     + (alpha_t * ((torch.exp(-h) - 1.0) / h + 1.0)) * D1
+            #     - (alpha_t * ((torch.exp(-h) - 1.0 + h) / h**2 - 0.5)) * D2
+            # )
+            noise = - (alpha_t * (torch.exp(-h) - 1.0)) * D0 + (alpha_t * ((torch.exp(-h) - 1.0) / h + 1.0)) * D1 - (alpha_t * ((torch.exp(-h) - 1.0 + h) / h**2 - 0.5)) * D2
+            if lm is True:
+                x_t = (sigma_t / sigma_s0) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+            else:
+                x_t = (sigma_t / sigma_s0) * sample + noise
+            self.prev_noise = noise
+        elif self.config.algorithm_type == "dpmsolver":
+            # See https://arxiv.org/abs/2206.00927 for detailed derivations
+            # x_t = (
+            #     (alpha_t / alpha_s0) * sample
+            #     - (sigma_t * (torch.exp(h) - 1.0)) * D0
+            #     - (sigma_t * ((torch.exp(h) - 1.0) / h - 1.0)) * D1
+            #     - (sigma_t * ((torch.exp(h) - 1.0 - h) / h**2 - 0.5)) * D2
+            # )
+            noise = - (sigma_t * (torch.exp(h) - 1.0)) * D0 - (sigma_t * ((torch.exp(h) - 1.0) / h - 1.0)) * D1 - (sigma_t * ((torch.exp(h) - 1.0 - h) / h**2 - 0.5)) * D2
+            if lm is True:
+                x_t = (alpha_t / alpha_s0) * sample + lm_correct(prev_noise=self.prev_noise, noise_pred = noise, lamb = self.lamb, kappa=self.kappa)
+            else:
+                x_t = (alpha_t / alpha_s0) * sample + noise
+            self.prev_noise = noise
+        return x_t
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: int,
+        sample: torch.FloatTensor,
+        generator=None,
+        return_dict: bool = True,
+    ) -> Union[SchedulerOutput, Tuple]:
+        """
+        Step function propagating the sample with the multistep DPM-Solver.
+
+        Args:
+            model_output (`torch.FloatTensor`): direct output from learned diffusion model.
+            timestep (`int`): current discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                current instance of sample being created by diffusion process.
+            return_dict (`bool`): option for returning tuple rather than SchedulerOutput class
+
+        Returns:
+            [`~scheduling_utils.SchedulerOutput`] or `tuple`: [`~scheduling_utils.SchedulerOutput`] if `return_dict` is
+            True, otherwise a `tuple`. When returning a tuple, the first element is the sample tensor.
+
+        """
+        lamb = self.lamb
+        lm = self.lm
+        kappa = self.kappa
+        if self.num_inference_steps is None:
+            raise ValueError(
+                "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
+            )
+
+        if isinstance(timestep, torch.Tensor):
+            timestep = timestep.to(self.timesteps.device)
+        step_index = (self.timesteps == timestep).nonzero()
+        if len(step_index) == 0:
+            step_index = len(self.timesteps) - 1
+        else:
+            step_index = step_index.item()
+        prev_timestep = 0 if step_index == len(self.timesteps) - 1 else self.timesteps[step_index + 1]
+        lower_order_final = (
+            (step_index == len(self.timesteps) - 1) and self.config.lower_order_final and len(self.timesteps) < 15
+        )
+        lower_order_second = (
+            (step_index == len(self.timesteps) - 2) and self.config.lower_order_final and len(self.timesteps) < 15
+        )
+
+        model_output = self.convert_model_output(model_output, timestep, sample)
+        for i in range(self.config.solver_order - 1):
+            self.model_outputs[i] = self.model_outputs[i + 1]
+        self.model_outputs[-1] = model_output
+
+        if self.config.algorithm_type in ["sde-dpmsolver", "sde-dpmsolver++"]:
+            noise = randn_tensor(
+                model_output.shape, generator=generator, device=model_output.device, dtype=model_output.dtype
+            )
+        else:
+            noise = None
+
+        if self.config.solver_order == 1 or self.lower_order_nums < 1 or lower_order_final:
+            prev_sample = self.dpm_solver_first_order_update(
+                model_output, timestep, prev_timestep, sample, noise=noise, lm=lm
+            )
+        elif self.config.solver_order == 2 or self.lower_order_nums < 2 or lower_order_second:
+            timestep_list = [self.timesteps[step_index - 1], timestep]
+            prev_sample = self.multistep_dpm_solver_second_order_update(
+                self.model_outputs, timestep_list, prev_timestep, sample, noise=noise, lm=lm
+            )
+        else:
+            timestep_list = [self.timesteps[step_index - 2], self.timesteps[step_index - 1], timestep]
+            prev_sample = self.multistep_dpm_solver_third_order_update(
+                self.model_outputs, timestep_list, prev_timestep, sample, lamb=lamb, lm=lm
+            )
+
+        if self.lower_order_nums < self.config.solver_order:
+            self.lower_order_nums += 1
+
+        if not return_dict:
+            return (prev_sample,)
+
+        return SchedulerOutput(prev_sample=prev_sample)
+
+    def scale_model_input(self, sample: torch.FloatTensor, *args, **kwargs) -> torch.FloatTensor:
+        """
+        Ensures interchangeability with schedulers that need to scale the denoising model input depending on the
+        current timestep.
+
+        Args:
+            sample (`torch.FloatTensor`): input sample
+
+        Returns:
+            `torch.FloatTensor`: scaled input sample
+        """
+        return sample
+
+    # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.add_noise
+    def add_noise(
+        self,
+        original_samples: torch.FloatTensor,
+        noise: torch.FloatTensor,
+        timesteps: torch.IntTensor,
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as original_samples
+        alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype)
+        timesteps = timesteps.to(original_samples.device)
+
+        sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
+        return noisy_samples
+
+    def __len__(self):
+        return self.config.num_train_timesteps

scheduler/scheduling_flow_match_euler_discrete_lm.py

@@ -0,0 +1,598 @@
+# Copyright 2024 Stability AI, Katherine Crowson and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+
+from diffusers.configuration_utils import ConfigMixin, register_to_config
+from diffusers.utils import BaseOutput, is_scipy_available, logging
+from diffusers.schedulers.scheduling_utils import SchedulerMixin
+
+
+if is_scipy_available():
+    import scipy.stats
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+def lm_correct(prev_noise, noise_pred, lamb, kappa):
+    noise_pred = noise_pred.to(torch.float32)
+    if prev_noise is not None:
+        noise_pred_ema = kappa * prev_noise + (1 - kappa) * noise_pred
+    else:
+        noise_pred_ema = noise_pred
+
+    # lm step for noise
+    norm_squared = (noise_pred * noise_pred).sum(dim=(1, 2))
+
+    norm_squared = norm_squared.unsqueeze(1).unsqueeze(2)
+    part1 =  noise_pred
+
+    norm_squared_ema = (noise_pred_ema * noise_pred_ema).sum(dim=(1, 2))
+    norm_squared_ema = norm_squared_ema.unsqueeze(1).unsqueeze(2)
+    inner_product = torch.sum(noise_pred * noise_pred_ema, dim=(1, 2))
+    mp = noise_pred_ema * inner_product.unsqueeze(-1).unsqueeze(-1)
+    part2 = mp /  (lamb + norm_squared_ema)
+
+    inversed_pred = part1 - part2
+
+    # normalize the direction
+    norm = torch.sqrt(norm_squared)
+    norm_squared_lm = (inversed_pred * inversed_pred).sum(dim=(1, 2))
+    norm_squared_lm = norm_squared_lm.unsqueeze(1).unsqueeze(2)
+    norm_lm = torch.sqrt(norm_squared_lm)
+    inversed_pred = inversed_pred * norm / norm_lm
+
+    return inversed_pred
+
+@dataclass
+class FlowMatchEulerDiscreteSchedulerOutput(BaseOutput):
+    """
+    Output class for the scheduler's `step` function output.
+
+    Args:
+        prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` for images):
+            Computed sample `(x_{t-1})` of previous timestep. `prev_sample` should be used as next model input in the
+            denoising loop.
+    """
+
+    prev_sample: torch.FloatTensor
+
+
+class FlowMatchEulerDiscreteLMScheduler(SchedulerMixin, ConfigMixin):
+    """
+    Euler scheduler.
+
+    This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic
+    methods the library implements for all schedulers such as loading and saving.
+
+    Args:
+        num_train_timesteps (`int`, defaults to 1000):
+            The number of diffusion steps to train the model.
+        shift (`float`, defaults to 1.0):
+            The shift value for the timestep schedule.
+        use_dynamic_shifting (`bool`, defaults to False):
+            Whether to apply timestep shifting on-the-fly based on the image resolution.
+        base_shift (`float`, defaults to 0.5):
+            Value to stabilize image generation. Increasing `base_shift` reduces variation and image is more consistent
+            with desired output.
+        max_shift (`float`, defaults to 1.15):
+            Value change allowed to latent vectors. Increasing `max_shift` encourages more variation and image may be
+            more exaggerated or stylized.
+        base_image_seq_len (`int`, defaults to 256):
+            The base image sequence length.
+        max_image_seq_len (`int`, defaults to 4096):
+            The maximum image sequence length.
+        invert_sigmas (`bool`, defaults to False):
+            Whether to invert the sigmas.
+        shift_terminal (`float`, defaults to None):
+            The end value of the shifted timestep schedule.
+        use_karras_sigmas (`bool`, defaults to False):
+            Whether to use Karras sigmas for step sizes in the noise schedule during sampling.
+        use_exponential_sigmas (`bool`, defaults to False):
+            Whether to use exponential sigmas for step sizes in the noise schedule during sampling.
+        use_beta_sigmas (`bool`, defaults to False):
+            Whether to use beta sigmas for step sizes in the noise schedule during sampling.
+        time_shift_type (`str`, defaults to "exponential"):
+            The type of dynamic resolution-dependent timestep shifting to apply. Either "exponential" or "linear".
+        stochastic_sampling (`bool`, defaults to False):
+            Whether to use stochastic sampling.
+    """
+
+    _compatibles = []
+    order = 1
+
+    @register_to_config
+    def __init__(
+        self,
+        num_train_timesteps: int = 1000,
+        shift: float = 1.0,
+        use_dynamic_shifting: bool = False,
+        base_shift: Optional[float] = 0.5,
+        max_shift: Optional[float] = 1.15,
+        base_image_seq_len: Optional[int] = 256,
+        max_image_seq_len: Optional[int] = 4096,
+        invert_sigmas: bool = False,
+        shift_terminal: Optional[float] = None,
+        use_karras_sigmas: Optional[bool] = False,
+        use_exponential_sigmas: Optional[bool] = False,
+        use_beta_sigmas: Optional[bool] = False,
+        time_shift_type: str = "exponential",
+        stochastic_sampling: bool = False,
+        lamb: float = 1.0,
+        lm: bool = True,
+        kappa: float = 0.0,
+    ):
+        if self.config.use_beta_sigmas and not is_scipy_available():
+            raise ImportError("Make sure to install scipy if you want to use beta sigmas.")
+        if sum([self.config.use_beta_sigmas, self.config.use_exponential_sigmas, self.config.use_karras_sigmas]) > 1:
+            raise ValueError(
+                "Only one of `config.use_beta_sigmas`, `config.use_exponential_sigmas`, `config.use_karras_sigmas` can be used."
+            )
+        if time_shift_type not in {"exponential", "linear"}:
+            raise ValueError("`time_shift_type` must either be 'exponential' or 'linear'.")
+
+        timesteps = np.linspace(1, num_train_timesteps, num_train_timesteps, dtype=np.float32)[::-1].copy()
+        timesteps = torch.from_numpy(timesteps).to(dtype=torch.float32)
+
+        sigmas = timesteps / num_train_timesteps
+        if not use_dynamic_shifting:
+            # when use_dynamic_shifting is True, we apply the timestep shifting on the fly based on the image resolution
+            sigmas = shift * sigmas / (1 + (shift - 1) * sigmas)
+
+        self.timesteps = sigmas * num_train_timesteps
+        self.lamb = lamb
+        self.lm = lm
+        self.kappa = kappa
+        self.prev_noise = None
+        self._step_index = None
+        self._begin_index = None
+
+        self._shift = shift
+
+        self.sigmas = sigmas.to("cpu")  # to avoid too much CPU/GPU communication
+        self.sigma_min = self.sigmas[-1].item()
+        self.sigma_max = self.sigmas[0].item()
+
+    @property
+    def shift(self):
+        """
+        The value used for shifting.
+        """
+        return self._shift
+
+    @property
+    def step_index(self):
+        """
+        The index counter for current timestep. It will increase 1 after each scheduler step.
+        """
+        return self._step_index
+
+    @property
+    def begin_index(self):
+        """
+        The index for the first timestep. It should be set from pipeline with `set_begin_index` method.
+        """
+        return self._begin_index
+
+    # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index
+    def set_begin_index(self, begin_index: int = 0):
+        """
+        Sets the begin index for the scheduler. This function should be run from pipeline before the inference.
+
+        Args:
+            begin_index (`int`):
+                The begin index for the scheduler.
+        """
+        self._begin_index = begin_index
+
+    def set_shift(self, shift: float):
+        self._shift = shift
+
+    def scale_noise(
+        self,
+        sample: torch.FloatTensor,
+        timestep: Union[float, torch.FloatTensor],
+        noise: Optional[torch.FloatTensor] = None,
+    ) -> torch.FloatTensor:
+        """
+        Forward process in flow-matching
+
+        Args:
+            sample (`torch.FloatTensor`):
+                The input sample.
+            timestep (`int`, *optional*):
+                The current timestep in the diffusion chain.
+
+        Returns:
+            `torch.FloatTensor`:
+                A scaled input sample.
+        """
+        # Make sure sigmas and timesteps have the same device and dtype as original_samples
+        sigmas = self.sigmas.to(device=sample.device, dtype=sample.dtype)
+
+        if sample.device.type == "mps" and torch.is_floating_point(timestep):
+            # mps does not support float64
+            schedule_timesteps = self.timesteps.to(sample.device, dtype=torch.float32)
+            timestep = timestep.to(sample.device, dtype=torch.float32)
+        else:
+            schedule_timesteps = self.timesteps.to(sample.device)
+            timestep = timestep.to(sample.device)
+
+        # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index
+        if self.begin_index is None:
+            step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timestep]
+        elif self.step_index is not None:
+            # add_noise is called after first denoising step (for inpainting)
+            step_indices = [self.step_index] * timestep.shape[0]
+        else:
+            # add noise is called before first denoising step to create initial latent(img2img)
+            step_indices = [self.begin_index] * timestep.shape[0]
+
+        sigma = sigmas[step_indices].flatten()
+        while len(sigma.shape) < len(sample.shape):
+            sigma = sigma.unsqueeze(-1)
+
+        sample = sigma * noise + (1.0 - sigma) * sample
+
+        return sample
+
+    def _sigma_to_t(self, sigma):
+        return sigma * self.config.num_train_timesteps
+
+    def time_shift(self, mu: float, sigma: float, t: torch.Tensor):
+        if self.config.time_shift_type == "exponential":
+            return self._time_shift_exponential(mu, sigma, t)
+        elif self.config.time_shift_type == "linear":
+            return self._time_shift_linear(mu, sigma, t)
+
+    def stretch_shift_to_terminal(self, t: torch.Tensor) -> torch.Tensor:
+        r"""
+        Stretches and shifts the timestep schedule to ensure it terminates at the configured `shift_terminal` config
+        value.
+
+        Reference:
+        https://github.com/Lightricks/LTX-Video/blob/a01a171f8fe3d99dce2728d60a73fecf4d4238ae/ltx_video/schedulers/rf.py#L51
+
+        Args:
+            t (`torch.Tensor`):
+                A tensor of timesteps to be stretched and shifted.
+
+        Returns:
+            `torch.Tensor`:
+                A tensor of adjusted timesteps such that the final value equals `self.config.shift_terminal`.
+        """
+        one_minus_z = 1 - t
+        scale_factor = one_minus_z[-1] / (1 - self.config.shift_terminal)
+        stretched_t = 1 - (one_minus_z / scale_factor)
+        return stretched_t
+
+    def set_timesteps(
+        self,
+        num_inference_steps: Optional[int] = None,
+        device: Union[str, torch.device] = None,
+        sigmas: Optional[List[float]] = None,
+        mu: Optional[float] = None,
+        timesteps: Optional[List[float]] = None,
+    ):
+        """
+        Sets the discrete timesteps used for the diffusion chain (to be run before inference).
+
+        Args:
+            num_inference_steps (`int`, *optional*):
+                The number of diffusion steps used when generating samples with a pre-trained model.
+            device (`str` or `torch.device`, *optional*):
+                The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+            sigmas (`List[float]`, *optional*):
+                Custom values for sigmas to be used for each diffusion step. If `None`, the sigmas are computed
+                automatically.
+            mu (`float`, *optional*):
+                Determines the amount of shifting applied to sigmas when performing resolution-dependent timestep
+                shifting.
+            timesteps (`List[float]`, *optional*):
+                Custom values for timesteps to be used for each diffusion step. If `None`, the timesteps are computed
+                automatically.
+        """
+        if self.config.use_dynamic_shifting and mu is None:
+            raise ValueError("`mu` must be passed when `use_dynamic_shifting` is set to be `True`")
+
+        if sigmas is not None and timesteps is not None:
+            if len(sigmas) != len(timesteps):
+                raise ValueError("`sigmas` and `timesteps` should have the same length")
+
+        if num_inference_steps is not None:
+            if (sigmas is not None and len(sigmas) != num_inference_steps) or (
+                timesteps is not None and len(timesteps) != num_inference_steps
+            ):
+                raise ValueError(
+                    "`sigmas` and `timesteps` should have the same length as num_inference_steps, if `num_inference_steps` is provided"
+                )
+        else:
+            num_inference_steps = len(sigmas) if sigmas is not None else len(timesteps)
+
+        self.num_inference_steps = num_inference_steps
+
+        # 1. Prepare default sigmas
+        is_timesteps_provided = timesteps is not None
+
+        if is_timesteps_provided:
+            timesteps = np.array(timesteps).astype(np.float32)
+
+        if sigmas is None:
+            if timesteps is None:
+                timesteps = np.linspace(
+                    self._sigma_to_t(self.sigma_max), self._sigma_to_t(self.sigma_min), num_inference_steps
+                )
+            sigmas = timesteps / self.config.num_train_timesteps
+        else:
+            sigmas = np.array(sigmas).astype(np.float32)
+            num_inference_steps = len(sigmas)
+
+        # 2. Perform timestep shifting. Either no shifting is applied, or resolution-dependent shifting of
+        #    "exponential" or "linear" type is applied
+        if self.config.use_dynamic_shifting:
+            sigmas = self.time_shift(mu, 1.0, sigmas)
+        else:
+            sigmas = self.shift * sigmas / (1 + (self.shift - 1) * sigmas)
+
+        # 3. If required, stretch the sigmas schedule to terminate at the configured `shift_terminal` value
+        if self.config.shift_terminal:
+            sigmas = self.stretch_shift_to_terminal(sigmas)
+
+        # 4. If required, convert sigmas to one of karras, exponential, or beta sigma schedules
+        if self.config.use_karras_sigmas:
+            sigmas = self._convert_to_karras(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+        elif self.config.use_exponential_sigmas:
+            sigmas = self._convert_to_exponential(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+        elif self.config.use_beta_sigmas:
+            sigmas = self._convert_to_beta(in_sigmas=sigmas, num_inference_steps=num_inference_steps)
+
+        # 5. Convert sigmas and timesteps to tensors and move to specified device
+        sigmas = torch.from_numpy(sigmas).to(dtype=torch.float32, device=device)
+        if not is_timesteps_provided:
+            timesteps = sigmas * self.config.num_train_timesteps
+        else:
+            timesteps = torch.from_numpy(timesteps).to(dtype=torch.float32, device=device)
+
+        # 6. Append the terminal sigma value.
+        #    If a model requires inverted sigma schedule for denoising but timesteps without inversion, the
+        #    `invert_sigmas` flag can be set to `True`. This case is only required in Mochi
+        if self.config.invert_sigmas:
+            sigmas = 1.0 - sigmas
+            timesteps = sigmas * self.config.num_train_timesteps
+            sigmas = torch.cat([sigmas, torch.ones(1, device=sigmas.device)])
+        else:
+            sigmas = torch.cat([sigmas, torch.zeros(1, device=sigmas.device)])
+
+        self.timesteps = timesteps
+        self.sigmas = sigmas
+        self._step_index = None
+        self._begin_index = None
+
+    def index_for_timestep(self, timestep, schedule_timesteps=None):
+        if schedule_timesteps is None:
+            schedule_timesteps = self.timesteps
+
+        indices = (schedule_timesteps == timestep).nonzero()
+
+        # The sigma index that is taken for the **very** first `step`
+        # is always the second index (or the last index if there is only 1)
+        # This way we can ensure we don't accidentally skip a sigma in
+        # case we start in the middle of the denoising schedule (e.g. for image-to-image)
+        pos = 1 if len(indices) > 1 else 0
+
+        return indices[pos].item()
+
+    def _init_step_index(self, timestep):
+        if self.begin_index is None:
+            if isinstance(timestep, torch.Tensor):
+                timestep = timestep.to(self.timesteps.device)
+            self._step_index = self.index_for_timestep(timestep)
+        else:
+            self._step_index = self._begin_index
+
+    def step(
+        self,
+        model_output: torch.FloatTensor,
+        timestep: Union[float, torch.FloatTensor],
+        sample: torch.FloatTensor,
+        s_churn: float = 0.0,
+        s_tmin: float = 0.0,
+        s_tmax: float = float("inf"),
+        s_noise: float = 1.0,
+        generator: Optional[torch.Generator] = None,
+        per_token_timesteps: Optional[torch.Tensor] = None,
+        return_dict: bool = True,
+    ) -> Union[FlowMatchEulerDiscreteSchedulerOutput, Tuple]:
+        """
+        Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion
+        process from the learned model outputs (most often the predicted noise).
+
+        Args:
+            model_output (`torch.FloatTensor`):
+                The direct output from learned diffusion model.
+            timestep (`float`):
+                The current discrete timestep in the diffusion chain.
+            sample (`torch.FloatTensor`):
+                A current instance of a sample created by the diffusion process.
+            s_churn (`float`):
+            s_tmin  (`float`):
+            s_tmax  (`float`):
+            s_noise (`float`, defaults to 1.0):
+                Scaling factor for noise added to the sample.
+            generator (`torch.Generator`, *optional*):
+                A random number generator.
+            per_token_timesteps (`torch.Tensor`, *optional*):
+                The timesteps for each token in the sample.
+            return_dict (`bool`):
+                Whether or not to return a
+                [`~schedulers.scheduling_flow_match_euler_discrete.FlowMatchEulerDiscreteSchedulerOutput`] or tuple.
+
+        Returns:
+            [`~schedulers.scheduling_flow_match_euler_discrete.FlowMatchEulerDiscreteSchedulerOutput`] or `tuple`:
+                If return_dict is `True`,
+                [`~schedulers.scheduling_flow_match_euler_discrete.FlowMatchEulerDiscreteSchedulerOutput`] is returned,
+                otherwise a tuple is returned where the first element is the sample tensor.
+        """
+
+        if (
+            isinstance(timestep, int)
+            or isinstance(timestep, torch.IntTensor)
+            or isinstance(timestep, torch.LongTensor)
+        ):
+            raise ValueError(
+                (
+                    "Passing integer indices (e.g. from `enumerate(timesteps)`) as timesteps to"
+                    " `FlowMatchEulerDiscreteScheduler.step()` is not supported. Make sure to pass"
+                    " one of the `scheduler.timesteps` as a timestep."
+                ),
+            )
+
+        if self.step_index is None:
+            self._init_step_index(timestep)
+
+        # Upcast to avoid precision issues when computing prev_sample
+        sample = sample.to(torch.float32)
+
+        if per_token_timesteps is not None:
+            per_token_sigmas = per_token_timesteps / self.config.num_train_timesteps
+
+            sigmas = self.sigmas[:, None, None]
+            lower_mask = sigmas < per_token_sigmas[None] - 1e-6
+            lower_sigmas = lower_mask * sigmas
+            lower_sigmas, _ = lower_sigmas.max(dim=0)
+
+            current_sigma = per_token_sigmas[..., None]
+            next_sigma = lower_sigmas[..., None]
+            dt = current_sigma - next_sigma
+        else:
+            sigma_idx = self.step_index
+            sigma = self.sigmas[sigma_idx]
+            sigma_next = self.sigmas[sigma_idx + 1]
+
+            current_sigma = sigma
+            next_sigma = sigma_next
+            dt = sigma_next - sigma
+
+        if self.config.stochastic_sampling:
+            x0 = sample - current_sigma * lm_correct(prev_noise=self.prev_noise, noise_pred = model_output, lamb = self.lamb, kappa=self.kappa)
+            noise = torch.randn_like(sample)
+            prev_sample = (1.0 - next_sigma) * x0 + next_sigma * noise
+            self.prev_noise = model_output
+        else:
+            prev_sample = sample + dt * lm_correct(prev_noise=self.prev_noise, noise_pred = model_output, lamb = self.lamb, kappa=self.kappa)
+            self.prev_noise = model_output
+        # upon completion increase step index by one
+        self._step_index += 1
+        if per_token_timesteps is None:
+            # Cast sample back to model compatible dtype
+            prev_sample = prev_sample.to(model_output.dtype)
+
+        if not return_dict:
+            return (prev_sample,)
+
+        return FlowMatchEulerDiscreteSchedulerOutput(prev_sample=prev_sample)
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_karras
+    def _convert_to_karras(self, in_sigmas: torch.Tensor, num_inference_steps) -> torch.Tensor:
+        """Constructs the noise schedule of Karras et al. (2022)."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        rho = 7.0  # 7.0 is the value used in the paper
+        ramp = np.linspace(0, 1, num_inference_steps)
+        min_inv_rho = sigma_min ** (1 / rho)
+        max_inv_rho = sigma_max ** (1 / rho)
+        sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_exponential
+    def _convert_to_exponential(self, in_sigmas: torch.Tensor, num_inference_steps: int) -> torch.Tensor:
+        """Constructs an exponential noise schedule."""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.exp(np.linspace(math.log(sigma_max), math.log(sigma_min), num_inference_steps))
+        return sigmas
+
+    # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_beta
+    def _convert_to_beta(
+        self, in_sigmas: torch.Tensor, num_inference_steps: int, alpha: float = 0.6, beta: float = 0.6
+    ) -> torch.Tensor:
+        """From "Beta Sampling is All You Need" [arXiv:2407.12173] (Lee et. al, 2024)"""
+
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
+
+        sigmas = np.array(
+            [
+                sigma_min + (ppf * (sigma_max - sigma_min))
+                for ppf in [
+                    scipy.stats.beta.ppf(timestep, alpha, beta)
+                    for timestep in 1 - np.linspace(0, 1, num_inference_steps)
+                ]
+            ]
+        )
+        return sigmas
+
+    def _time_shift_exponential(self, mu, sigma, t):
+        return math.exp(mu) / (math.exp(mu) + (1 / t - 1) ** sigma)
+
+    def _time_shift_linear(self, mu, sigma, t):
+        return mu / (mu + (1 / t - 1) ** sigma)
+
+    def __len__(self):
+        return self.config.num_train_timesteps

scripts/FLUX_T2i_Sampling.py

@@ -0,0 +1,108 @@
+import sys
+
+import torch
+import os
+import json
+import argparse
+
+sys.path.append(os.getcwd())
+
+from diffusers import StableDiffusion3Pipeline, FluxPipeline, FlowMatchHeunDiscreteScheduler, FlowMatchEulerDiscreteScheduler
+from scheduler.scheduling_flow_match_euler_discrete_lm import FlowMatchEulerDiscreteLMScheduler
+from tqdm import tqdm
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for T2I-Bench.")
+    parser.add_argument('--test_num', type=int, default=10)
+    parser.add_argument('--start_index', type=int, default=0)
+    parser.add_argument('--num_inference_steps', type=int, default=10)
+    parser.add_argument('--guidance', type=float, default=7.5)
+    parser.add_argument('--sampler_type', type = str, default='fm_euler')
+    parser.add_argument('--model_id', type=str, default='XXX')
+    parser.add_argument('--save_dir', type=str, default='results/')
+    parser.add_argument('--lamb', type=float, default=5.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--freeze', type=float, default=0.0)
+    parser.add_argument('--dataset_category', type=str, default="color")
+    parser.add_argument('--dataset_path', type=str, default="T2I-CompBench-main")
+    parser.add_argument('--dtype', type=str, default='bf16')
+    parser.add_argument('--device', type=str, default='cuda')
+
+    args = parser.parse_args()
+    dtype = None
+    if args.dtype in ['fp32']:
+        dtype = torch.float32
+    elif args.dtype in ['fp64']:
+        dtype = torch.float64
+    elif args.dtype in ['fp16']:
+        dtype = torch.float16
+    elif args.dtype in ['bf16']:
+        dtype = torch.bfloat16
+
+    start_index = args.start_index
+    sampler_type = args.sampler_type
+    test_num = args.test_num
+    guidance_scale = args.guidance
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    freeze = args.freeze
+    kappa = args.kappa
+    model_id = args.model_id
+    device = args.device
+
+    # load model
+    sd_pipe = FluxPipeline.from_pretrained(
+        model_id,
+        torch_dtype=dtype, safety_checker=None)
+    sd_pipe = sd_pipe.to(device)
+    print("flux model loaded")
+
+    if sampler_type in ['fm_euler']:
+        pass
+    elif sampler_type in ['lml_euler']:
+        sd_pipe.scheduler = FlowMatchEulerDiscreteLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = True
+        sd_pipe.scheduler.kappa = kappa
+    else:
+        raise ValueError(f"invalid: '{sampler_type}'.")
+
+    save_dir = args.save_dir
+    
+    if sampler_type in ['lml_euler']:
+        save_dir = os.path.join(save_dir, "flux", args.dataset_category, sampler_type + "_lamda_" + str(lamb))
+    else:
+        save_dir = os.path.join(save_dir, "flux", args.dataset_category, sampler_type)
+    
+    save_dir = os.path.join(save_dir, "samples")
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+    
+    def getT2IDataset(file_path):
+        with open(file_path, "r", encoding="utf-8") as file:
+            for line in file:
+                stripped_line = line.strip()
+                if stripped_line:
+                    yield stripped_line
+    
+    # T2I prompts
+    dataset_path = os.path.join(args.dataset_path, 'examples/dataset', args.dataset_category + '_val.txt')
+    count = 0
+    with tqdm(total=300 * test_num, desc="Generating Images") as pbar:
+        try:
+            for prompt in getT2IDataset(dataset_path):
+                for seed in range(start_index, start_index + test_num):
+                    torch.manual_seed(seed)
+                    res = sd_pipe(prompt=prompt, num_inference_steps=num_inference_steps,
+                            guidance_scale=guidance_scale, generator=None, width=512, height=512).images[0]
+                    res.save(os.path.join(save_dir, f"{prompt}_{count:06d}.png"))
+                    count += 1
+                    pbar.update(1)
+        except FileNotFoundError:
+            print(f"dataset can not be found: {dataset_path}")
+        except Exception as e:
+            print(f"unknown error: {str(e)}")
+    print(f"{dataset_path} finish")
+
+if __name__ == '__main__':
+    main()

scripts/StableDiffusion_COCO.py

@@ -0,0 +1,115 @@
+import sys
+
+import torch
+import os
+import json
+import argparse
+sys.path.append(os.getcwd())
+
+from diffusers import StableDiffusionPipeline, DPMSolverMultistepLMScheduler, DDIMLMScheduler, PNDMScheduler, UniPCMultistepScheduler
+
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for COCO14.")
+    parser.add_argument('--test_num', type=int, default=1000)
+    parser.add_argument('--start_index', type=int, default=0)
+    parser.add_argument('--seed', type=int, default=1)
+    parser.add_argument('--num_inference_steps', type=int, default=20)
+    parser.add_argument('--guidance', type=float, default=7.5)
+    parser.add_argument('--sampler_type', type = str, default='ddim')
+    parser.add_argument('--model_id', type=str, default='/xxx/xxx/stable-diffusion-v1-5')
+    parser.add_argument('--save_dir', type=str, default='/xxx/xxx')
+    parser.add_argument('--lamb', type=float, default=5.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--device', type=str, default='cuda')
+
+
+    args = parser.parse_args()
+
+    start_index = args.start_index
+    sampler_type = args.sampler_type
+    test_num = args.test_num
+    guidance_scale = args.guidance
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    kappa = args.kappa
+    device = args.device
+    model_id = args.model_id
+
+
+    # load model
+    sd_pipe = None
+
+    sd_pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float32, safety_checker=None)
+    sd_pipe = sd_pipe.to(device)
+    print("sd model loaded")
+    
+
+    if sampler_type in ['dpm_lm']:
+        sd_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.config.solver_order = 3
+        sd_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = True
+    elif sampler_type in ['dpm']:
+        sd_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.config.solver_order = 3
+        sd_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++']:
+        sd_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.config.solver_order = 3
+        sd_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++_lm']:
+        sd_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.config.solver_order = 3
+        sd_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = True
+    elif sampler_type in ['pndm']:
+        sd_pipe.scheduler = PNDMScheduler.from_config(sd_pipe.scheduler.config)
+    elif sampler_type in ['ddim']:
+        sd_pipe.scheduler = DDIMLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = False
+        sd_pipe.scheduler.kappa = kappa
+    elif sampler_type in ['ddim_lm']:
+        sd_pipe.scheduler = DDIMLMScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe.scheduler.lamb = lamb
+        sd_pipe.scheduler.lm = True
+        sd_pipe.scheduler.kappa = kappa
+    elif sampler_type in ['unipc']:
+        sd_pipe.scheduler = UniPCMultistepScheduler.from_config(sd_pipe.scheduler.config)
+
+    save_dir = args.save_dir
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+
+    # COCO prompts
+    with open('/mnt/chongqinggeminiceph1fs/geminicephfs/mm-base-vision/pazelzhang/make_dataset/fid_3W_json.json') as fr:
+        COCO_prompts_dict = json.load(fr)
+    image_id = COCO_prompts_dict.keys()
+    with torch.no_grad():
+        for pi, key in enumerate(image_id):
+            if pi >= start_index  and pi < start_index + test_num:
+                print(key)
+                print(COCO_prompts_dict[key])
+                prompt = COCO_prompts_dict[key]
+                negative_prompt = None
+
+                for seed in [1]:
+                    generator = torch.Generator(device='cuda')
+                    generator = generator.manual_seed(args.seed)
+                    res = sd_pipe(prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=num_inference_steps,
+                            guidance_scale=guidance_scale, generator=generator).images[0]
+                    res.save(os.path.join(save_dir, f"{pi:05d}_{key}_guidance{guidance_scale}_inference{num_inference_steps}_seed{seed}_{sampler_type}.jpg"))
+                    print(f"{sampler_type}##{key},done")
+
+
+if __name__ == '__main__':
+    main()

scripts/StableDiffusion_PixArt_T2i_Sampling.py

@@ -0,0 +1,159 @@
+import sys
+
+import torch
+import os
+import json
+import argparse
+
+sys.path.append(os.getcwd())
+
+from diffusers import StableDiffusionPipeline, PNDMScheduler, UniPCMultistepScheduler, DDIMScheduler, DiffusionPipeline, PixArtAlphaPipeline
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+from tqdm import tqdm
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for T2I-Bench.")
+    parser.add_argument('--test_num', type=int, default=10)
+    parser.add_argument('--start_index', type=int, default=0)
+    parser.add_argument('--num_inference_steps', type=int, default=10)
+    parser.add_argument('--guidance', type=float, default=7.5)
+    parser.add_argument('--sampler_type', type = str, default='dpm_lm')
+    parser.add_argument('--model', type=str, default='sd15', choices=['sd15', 'sd2_base', 'sdxl', 'pixart'])
+    parser.add_argument('--model_dir', type=str, default='XXX')
+    parser.add_argument('--save_dir', type=str, default='results/')
+    parser.add_argument('--lamb', type=float, default=5.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--freeze', type=float, default=0.0)
+    parser.add_argument('--dataset_category', type=str, default="color")
+    parser.add_argument('--dataset_path', type=str, default="../T2I-CompBench-main")
+    parser.add_argument('--dtype', type=str, default='fp32')
+    parser.add_argument('--device', type=str, default='cuda')
+
+    args = parser.parse_args()
+    dtype = None
+    if args.dtype in ['fp32']:
+        dtype = torch.float32
+    elif args.dtype in ['fp64']:
+        dtype = torch.float64
+    elif args.dtype in ['fp16']:
+        dtype = torch.float16
+    elif args.dtype in ['bf16']:
+        dtype = torch.bfloat16
+
+    device = args.device
+    start_index = args.start_index
+    sampler_type = args.sampler_type
+    test_num = args.test_num
+    guidance_scale = args.guidance
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    freeze = args.freeze
+    kappa = args.kappa
+    model_dir = args.model_dir
+
+    # load model
+    sd_pipe = None
+    if args.model in ['sd15']:
+        sd_pipe = StableDiffusionPipeline.from_pretrained(
+            model_dir,
+            torch_dtype=dtype, safety_checker=None)
+        sd_pipe = sd_pipe.to(device)
+        print("sd-1.5 model loaded")
+    elif args.model in ['sd2_base']:
+        sd_pipe = StableDiffusionPipeline.from_pretrained(
+            model_dir,
+            torch_dtype=dtype, safety_checker=None)
+        sd_pipe = sd_pipe.to(device)
+        print("sd-2-base model loaded")
+    elif args.model in ['sdxl']:
+        sd_pipe = DiffusionPipeline.from_pretrained(
+            model_dir,
+            torch_dtype=dtype, safety_checker=None)
+        sd_pipe = sd_pipe.to(device)
+        print("sd-xl-base model loaded")
+    elif args.model in ['pixart']:
+        sd_pipe = PixArtAlphaPipeline.from_pretrained(
+            model_dir,
+            torch_dtype=dtype, safety_checker=None)
+        sd_pipe = sd_pipe.to(device)
+        print("PixArt-XL-2-512x512 model loaded")
+
+    SAMPLER_CONFIG = {
+        'dpm_lm': {
+            'scheduler': DPMSolverMultistepLMScheduler,
+            'params': {'solver_order': 3, 'algorithm_type': "dpmsolver", 'lm': True, 'lamb': lamb, 'kappa': kappa, 'freeze': freeze}
+        },
+        'dpm': {
+            'scheduler': DPMSolverMultistepLMScheduler,
+            'params': {'solver_order': 3, 'algorithm_type': "dpmsolver", 'lm': False}
+        },
+        'dpm++': {
+            'scheduler': DPMSolverMultistepLMScheduler,
+            'params': {'solver_order': 3, 'algorithm_type': "dpmsolver++", 'lm': False}
+        },
+        'dpm++_lm': {
+            'scheduler': DPMSolverMultistepLMScheduler,
+            'params': {'solver_order': 3, 'algorithm_type': "dpmsolver++", 'lm': True, 'lamb': lamb, 'kappa': kappa, 'freeze': freeze}
+        },
+        'pndm': {'scheduler': PNDMScheduler, 'params': {}},
+        'ddim': {'scheduler': DDIMScheduler, 'params': {}},
+        'ddim_lm': {
+            'scheduler': DDIMLMScheduler,
+            'params': {'lm': True, 'lamb': lamb, 'kappa': kappa, 'freeze': freeze}
+        },
+        'unipc': {'scheduler': UniPCMultistepScheduler, 'params': {}},
+    }
+
+    if sampler_type in SAMPLER_CONFIG:
+        config = SAMPLER_CONFIG[sampler_type]
+        scheduler_class = config['scheduler']
+        sd_pipe.scheduler = scheduler_class.from_config(sd_pipe.scheduler.config)
+        
+        for param, value in config['params'].items():
+            if hasattr(sd_pipe.scheduler, param):
+                setattr(sd_pipe.scheduler, param, value)
+            elif hasattr(sd_pipe.scheduler.config, param):
+                 setattr(sd_pipe.scheduler.config, param, value)
+    else:
+        raise ValueError(f"invalid: '{sampler_type}'.")
+        
+    save_dir = args.save_dir
+    
+    if sampler_type in ['ddim_lm', 'dpm++_lm', 'dpm_lm']:
+        save_dir = os.path.join(save_dir, args.model, args.dataset_category, sampler_type + "_lambda_" + str(lamb))
+    else:
+        save_dir = os.path.join(save_dir, args.model, args.dataset_category, sampler_type)
+    save_dir = os.path.join(save_dir, "samples")
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+    
+    def getT2IDataset(file_path):
+        with open(file_path, "r", encoding="utf-8") as file:
+            for line in file:
+                stripped_line = line.strip()
+                if stripped_line:
+                    yield stripped_line
+    
+    # T2I prompts
+    dataset_path = os.path.join(args.dataset_path, 'examples/dataset', args.dataset_category + '_val.txt')
+    count = 0
+    with tqdm(total=300 * test_num, desc="Generating Images") as pbar:
+        try:
+            for prompt in getT2IDataset(dataset_path):
+                for seed in range(start_index, start_index + test_num):
+                    torch.manual_seed(seed)
+                    res = sd_pipe(prompt=prompt, num_inference_steps=num_inference_steps,
+                            guidance_scale=guidance_scale, generator=None).images[0]
+                    res.save(os.path.join(save_dir, f"{prompt}_{count:06d}.png"))
+                    count += 1
+                    pbar.update(1)
+        except FileNotFoundError:
+            print(f"dataset can not be found: {dataset_path}")
+        except Exception as e:
+            print(f"unknown error: {str(e)}")
+    print(f"{dataset_path} finish")
+
+if __name__ == '__main__':
+    main()

scripts/celeba.py

@@ -0,0 +1,110 @@
+import sys
+import time
+import torch
+import os
+
+import json
+import argparse
+sys.path.append(os.getcwd())
+from diffusers import LDMPipeline, DDIMScheduler, PNDMScheduler, UniPCMultistepScheduler, DPMSolverMultistepScheduler
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for CelebA-HQ.")
+    parser.add_argument('--test_num', type=int, default=1)
+    parser.add_argument('--start_index', type=int, default=0)
+    parser.add_argument('--batch_size', type=int, default=4)
+    parser.add_argument('--num_inference_steps', type=int, default=20)
+    parser.add_argument('--sampler_type', type = str,default='lag', choices=[ 'pndm', 'ddim_lm', 'ddim', 'dpm++', 'dpm','dpm_lm', 'unipc'])
+    parser.add_argument('--save_dir', type=str, default='/xxx/xxx')
+    parser.add_argument('--model_id', type=str,
+                        default='/xxx/xxx/ddpm_ema_cifar10')
+    parser.add_argument('--lamb', type=float, default=1.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--dtype', type=str, default='fp32')
+    parser.add_argument('--device', type=str, default='cuda')
+
+    args = parser.parse_args()
+
+    dtype = None
+    if args.dtype in ['fp32']:
+        dtype = torch.float32
+    elif args.dtype in ['fp64']:
+        dtype = torch.float64
+    elif args.dtype in ['fp16']:
+        dtype = torch.float16
+    elif args.dtype in ['bf16']:
+        dtype = torch.bfloat16
+
+    start_index = args.start_index
+    batch_size = args.batch_size
+    sampler_type = args.sampler_type
+    test_num = args.test_num
+    num_inference_steps = args.num_inference_steps
+    device = args.device
+    lamb = args.lamb
+    kappa = args.kappa
+    model_id = args.model_id
+
+    save_dir = args.save_dir
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+
+    with torch.no_grad():
+        # load pipeline
+        pipe = LDMPipeline.from_pretrained(model_id, torch_dtype=dtype)
+        pipe.unet.to(device)
+        pipe.vqvae.to(device)
+
+        # load scheduler
+        if sampler_type in ['pndm']:
+            pipe.scheduler = PNDMScheduler.from_config(pipe.scheduler.config)
+        elif sampler_type in ['dpm++']:
+            pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.config.solver_order = 3
+            pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        elif sampler_type in ['dpm_lm']:
+            pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.config.solver_order = 3
+            pipe.scheduler.config.algorithm_type = "dpmsolver"
+            pipe.scheduler.lamb = lamb
+            pipe.scheduler.lm = True
+            pipe.scheduler.kappa = kappa
+        elif sampler_type in ['dpm']:
+            pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.config.solver_order = 3
+            pipe.scheduler.config.algorithm_type = "dpmsolver"
+            pipe.scheduler.lm = False
+        elif sampler_type in ['ddim']:
+            pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+        elif sampler_type in ['ddim_lm']:
+            pipe.scheduler = DDIMLMScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.lamb = lamb
+            pipe.scheduler.lm = True
+            pipe.scheduler.kappa = kappa
+        elif sampler_type in ['unipc']:
+            pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+
+        for seed in range(start_index, start_index + test_num):
+            print('prepare to sample')
+            start_time = time.time()
+            torch.manual_seed(seed)
+            
+            # sampling process
+            images = pipe(batch_size=batch_size, num_inference_steps=num_inference_steps).images
+
+            # store the generated images
+            for i, image in enumerate(images):
+                image.save(
+                    os.path.join(save_dir, f"cifar10_{sampler_type}_inference{num_inference_steps}_seed{seed}_{i}.png"))
+            print(f"{sampler_type} batch##{seed},done")
+
+            # output the sampling time-costs
+            end_time = time.time()
+            time_difference = end_time - start_time
+            print(f"The code took {time_difference} seconds to run.")
+
+if __name__ == '__main__':
+    main()
+

scripts/cifar10.py

@@ -0,0 +1,104 @@
+import sys
+import time
+import torch
+import os
+
+import json
+import argparse
+sys.path.append(os.getcwd())
+from diffusers import DDPMPipeline, DDIMScheduler, PNDMScheduler, UniPCMultistepScheduler, DPMSolverMultistepScheduler
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for CIFAR-10.")
+    parser.add_argument('--test_num', type=int, default=1)
+    parser.add_argument('--start_index', type=int, default=0)
+    parser.add_argument('--batch_size', type=int, default=4)
+    parser.add_argument('--num_inference_steps', type=int, default=20)
+    parser.add_argument('--sampler_type', type = str,default='lag', choices=[ 'pndm', 'ddim', 'dpm++', 'dpm','dpm_lm', 'unipc'])
+    parser.add_argument('--save_dir', type=str, default='/xxx/xxx')
+    parser.add_argument('--model_id', type=str,
+                        default='/xxx/xxx/ddpm_ema_cifar10')
+    parser.add_argument('--lamb', type=float, default=1.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--dtype', type=str, default='fp32')
+    parser.add_argument('--device', type=str, default='cuda')
+
+    args = parser.parse_args()
+
+    dtype = None
+    if args.dtype in ['fp32']:
+        dtype = torch.float32
+    elif args.dtype in ['fp64']:
+        dtype = torch.float64
+    elif args.dtype in ['fp16']:
+        dtype = torch.float16
+    elif args.dtype in ['bf16']:
+        dtype = torch.bfloat16
+
+    start_index = args.start_index
+    device = args.device
+    batch_size = args.batch_size
+    sampler_type = args.sampler_type
+    test_num = args.test_num
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    kappa = args.kappa
+    model_id = args.model_id
+
+    save_dir = args.save_dir
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+
+    with torch.no_grad():
+        # load pipeline
+        pipe = DDPMPipeline.from_pretrained(model_id, torch_dtype=dtype)
+        pipe.unet.to(device)
+
+        # load scheduler
+        if sampler_type in ['pndm']:
+            pipe.scheduler = PNDMScheduler.from_config(pipe.scheduler.config)
+        elif sampler_type in ['dpm++']:
+            pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.config.solver_order = 3
+            pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        elif sampler_type in ['dpm_lm']:
+            pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.config.solver_order = 3
+            pipe.scheduler.config.algorithm_type = "dpmsolver"
+            pipe.scheduler.lamb = lamb
+            pipe.scheduler.lm = True
+            pipe.scheduler.kappa = kappa
+        elif sampler_type in ['dpm']:
+            pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(pipe.scheduler.config)
+            pipe.scheduler.config.solver_order = 3
+            pipe.scheduler.config.algorithm_type = "dpmsolver"
+            pipe.scheduler.lm = False
+        elif sampler_type in ['ddim']:
+            pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+        elif sampler_type in ['unipc']:
+            pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+
+        for seed in range(start_index, start_index + test_num):
+            print('prepare to sample')
+            start_time = time.time()
+            torch.manual_seed(seed)
+            
+            # sampling process
+            images = pipe(batch_size=batch_size, num_inference_steps=num_inference_steps).images
+
+            # store the generated images
+            for i, image in enumerate(images):
+                image.save(
+                    os.path.join(save_dir, f"cifar10_{sampler_type}_inference{num_inference_steps}_seed{seed}_{i}.png"))
+            print(f"{sampler_type} batch##{seed},done")
+
+            # output the sampling time-costs
+            end_time = time.time()
+            time_difference = end_time - start_time
+            print(f"The code took {time_difference} seconds to run.")
+
+if __name__ == '__main__':
+    main()
+

scripts/control_net_canny.py

@@ -0,0 +1,147 @@
+import sys
+
+import torch
+import os
+import json
+import argparse
+sys.path.append(os.getcwd())
+
+from PIL import Image
+import torchvision.transforms as transforms
+import numpy as np
+import glob
+
+
+from diffusers import StableDiffusionPipeline, DDIMScheduler
+from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
+from diffusers.utils import load_image
+from diffusers import StableDiffusionPipeline, DDPMPipeline, DDIMPipeline, PNDMPipeline, PNDMLMPipeline, DDPMLMPipeline, DPMLMPipeline, UniPCPipeline, LDMPipeline, PNDMScheduler, UniPCMultistepScheduler,DDIMScheduler
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+import cv2
+import numpy as np
+
+
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for ControlNet-canny.")
+    parser.add_argument('--seed', type=int, default=1)
+    parser.add_argument('--num_inference_steps', type=int, default=20)
+    parser.add_argument('--guidance', type=float, default=7.5)
+    parser.add_argument('--sampler_type', type = str,default='lag')
+    parser.add_argument('--prompt', type=str, default='an asian girl')
+    parser.add_argument('--original_image_path', type=str, default="/xxx/xxx/data/input_image_vermeer.png")
+    parser.add_argument('--lamb', type=float, default=5.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--freeze', type=float, default=0.0)
+    # parser.add_argument('--prompt_list', nargs='+', type=str,
+    #                     default=['an asian girl'])
+    parser.add_argument('--save_dir', type=str, default='/xxx/xxx/result/0402')
+    parser.add_argument('--controlnet_dir', type=str, default="/xxx/xxx/sd-controlnet-canny")
+    parser.add_argument('--sd_dir', type=str, default="/xxx/xxx/stable-diffusion-v1-5")
+
+
+
+    args = parser.parse_args()
+    if args.sampler_type in ['bdia']:
+        parser.add_argument('--bdia_gamma', type=float, default=0.5)
+    if args.sampler_type in ['edict']:
+        parser.add_argument('--edict_p', type=float, default=0.93)
+    args = parser.parse_args()
+    device = 'cuda'
+    sampler_type = args.sampler_type
+    guidance_scale = args.guidance
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    freeze = args.freeze
+    kappa = args.kappa
+
+    save_dir = args.save_dir
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+
+    torch.manual_seed(args.seed)
+    controlnet = ControlNetModel.from_pretrained(args.controlnet_dir, torch_dtype=torch.float16,use_safetensors=True)
+
+    control_pipe = StableDiffusionControlNetPipeline.from_pretrained(
+        args.sd_dir, controlnet=controlnet, torch_dtype=torch.float16, use_safetensors=True
+    )
+    control_pipe.enable_model_cpu_offload()
+    control_pipe.safety_checker = None
+
+    if sampler_type in ['dpm_lm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+    elif sampler_type in ['dpm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++_lm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+    elif sampler_type in ['pndm']:
+        control_pipe.scheduler = PNDMScheduler.from_config(control_pipe.scheduler.config)
+    elif sampler_type in ['ddim']:
+        control_pipe.scheduler = DDIMScheduler.from_config(control_pipe.scheduler.config)
+    elif sampler_type in ['ddim_lm']:
+        control_pipe.scheduler = DDIMLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+        control_pipe.scheduler.kappa = kappa
+        control_pipe.scheduler.freeze = freeze
+    elif sampler_type in ['unipc']:
+        control_pipe.scheduler = UniPCMultistepScheduler.from_config(control_pipe.scheduler.config)
+
+    original_image = load_image(
+        args.original_image_path
+    )
+    image = np.array(original_image)
+    low_threshold = 100
+    high_threshold = 200
+
+    image = cv2.Canny(image, low_threshold, high_threshold)
+    image = image[:, :, None]
+    image = np.concatenate([image, image, image], axis=2)
+    canny_image = Image.fromarray(image)
+
+
+    for prompt, negative_prompt in [['the mona lisa',''],
+                                    ['an asian girl',''],
+                                    ['an asian princess',''],
+                                    ['a portrait of a beautiful woman standing amidst a bed of vibrant tulips.',''],
+                                    ['a stunning Arabic woman dressed in traditional clothing',''],
+                                    ['a stunning Asian woman dressed in traditional clothing',''],
+                                    ['a stunning Black woman dressed in traditional clothing', ''],
+                                    ['a stunning German woman dressed in traditional clothing', ''],
+                                    ['a stunning Japan woman dressed in traditional clothing', ''],
+                                    ['a stunning Chinese woman dressed in traditional clothing', ''],
+                                    ['a stunning Jewish woman dressed in traditional clothing', ''],
+                                    ]:
+        for seed in range(1):
+            torch.manual_seed(seed)
+            res = control_pipe(
+                prompt=prompt, negative_prompt=negative_prompt, image=canny_image,num_inference_steps=num_inference_steps,
+            ).images[0]
+
+            res.save(os.path.join(save_dir,
+                                  f"{args.model}_{prompt[:20]}_seed{seed}_{sampler_type}_infer{num_inference_steps}_g{guidance_scale}_lamb{args.lamb}.png"))
+
+
+
+if __name__ == '__main__':
+    main()

scripts/control_net_depth.py

@@ -0,0 +1,151 @@
+import sys
+
+import torch
+import os
+import argparse
+sys.path.append(os.getcwd())
+
+from PIL import Image
+import torchvision.transforms as transforms
+import numpy as np
+
+
+from diffusers import StableDiffusionPipeline, DDIMScheduler
+from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
+from diffusers.utils import load_image
+from diffusers import PNDMScheduler, UniPCMultistepScheduler,DDIMScheduler
+from diffusers import StableDiffusionControlNetImg2ImgPipeline, ControlNetModel
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+from transformers import pipeline
+import cv2
+import numpy as np
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for ControlNet-depth.")
+    parser.add_argument('--seed', type=int, default=1)
+    parser.add_argument('--num_inference_steps', type=int, default=20)
+    parser.add_argument('--guidance', type=float, default=7.5)
+    parser.add_argument('--sampler_type', type = str,default='lag')
+    parser.add_argument('--model', type=str, default='sd2_base', choices=['sd15', 'sd2_base'])
+    parser.add_argument('--prompt', type=str, default='an asian girl')
+    parser.add_argument('--lamb', type=float, default=5.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--freeze', type=float, default=0.0)
+    parser.add_argument('--prompt_list', nargs='+', type=str,
+                        default=['an asian girl'])
+    parser.add_argument('--save_dir', type=str, default='/xxx/xxx/result/0402')
+    parser.add_argument('--controlnet_dir', type=str, default="/xxx/xxx/control_v11f1p_sd15_depth")
+    parser.add_argument('--sd_dir', type=str, default="/xxx/xxx/stable-diffusion-v1-5")
+
+
+    args = parser.parse_args()
+    if args.sampler_type in ['bdia']:
+        parser.add_argument('--bdia_gamma', type=float, default=0.5)
+    if args.sampler_type in ['edict']:
+        parser.add_argument('--edict_p', type=float, default=0.93)
+    args = parser.parse_args()
+    device = 'cuda'
+    sampler_type = args.sampler_type
+    guidance_scale = args.guidance
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    freeze = args.freeze
+    kappa = args.kappa
+
+    save_dir = args.save_dir
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+
+    # torch.manual_seed(args.seed)
+    controlnet = ControlNetModel.from_pretrained(args.controlnet_dir, torch_dtype=torch.float16, use_safetensors=True)
+
+    control_pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
+        args.sd_dir,
+        controlnet=controlnet, torch_dtype=torch.float16, use_safetensors=True
+    )
+    control_pipe.enable_model_cpu_offload()
+    control_pipe.safety_checker = None
+
+    if sampler_type in ['dpm_lm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+    elif sampler_type in ['dpm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++_lm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+    elif sampler_type in ['pndm']:
+        control_pipe.scheduler = PNDMScheduler.from_config(control_pipe.scheduler.config)
+    elif sampler_type in ['ddim']:
+        control_pipe.scheduler = DDIMScheduler.from_config(control_pipe.scheduler.config)
+        # control_pipe.scheduler.lamb = lamb
+        # control_pipe.scheduler.lm = False
+        # control_pipe.scheduler.kappa = kappa
+    elif sampler_type in ['ddim_lm']:
+        control_pipe.scheduler = DDIMLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+        control_pipe.scheduler.kappa = kappa
+        control_pipe.scheduler.freeze = freeze
+    elif sampler_type in ['unipc']:
+        control_pipe.scheduler = UniPCMultistepScheduler.from_config(control_pipe.scheduler.config)
+
+    image = load_image(
+        "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet-img2img.jpg"
+    )
+
+    def get_depth_map(image, depth_estimator):
+        image = depth_estimator(image)["depth"]
+        image = np.array(image)
+        image = image[:, :, None]
+        image = np.concatenate([image, image, image], axis=2)
+        detected_map = torch.from_numpy(image).float() / 255.0
+        depth_map = detected_map.permute(2, 0, 1)
+        return depth_map
+
+    depth_estimator = pipeline("depth-estimation")
+    depth_map = get_depth_map(image, depth_estimator).unsqueeze(0).half().to("cuda")
+    transforms.ToPILImage()(depth_map[0]).save(os.path.join(save_dir,
+                                  f"depth_map.png"))
+
+
+    for prompt, negative_prompt in [["lego batman and robin",''],
+                                    ["Spider-Man and Superman", ''],
+                                    ["A girl and a boy", ''],
+                                    ["asian woman and asian man", ''],
+                                    ["American Indian woman and American Indian man", ''],
+                                    ["A girl and a girl, monalisa style", ''],
+                                    ["Elsa and Anna, in the movie Frozen", ''],
+                                    ["A woman and a man, wearing suit", ''],
+                                    ]:
+        for seed in range(20):
+            torch.manual_seed(seed)
+            res = control_pipe(
+                prompt = prompt, image=image, control_image=depth_map,num_inference_steps=num_inference_steps,
+            ).images[0]
+
+            res.save(os.path.join(save_dir,
+                                  f"{prompt[:20]}_seed{seed}_{sampler_type}_infer{num_inference_steps}_g{guidance_scale}_lamb{args.lamb}.png"))
+
+
+
+if __name__ == '__main__':
+    main()

scripts/control_net_pose.py

@@ -0,0 +1,154 @@
+import sys
+
+import torch
+import os
+import json
+import argparse
+sys.path.append(os.getcwd())
+
+from PIL import Image
+import torchvision.transforms as transforms
+import numpy as np
+import glob
+
+from diffusers import StableDiffusionPipeline, DDIMScheduler
+from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
+from diffusers.utils import load_image
+from diffusers import PNDMScheduler, UniPCMultistepScheduler,DDIMScheduler
+from diffusers import StableDiffusionControlNetImg2ImgPipeline, ControlNetModel
+from scheduler.scheduling_dpmsolver_multistep_lm import DPMSolverMultistepLMScheduler
+from scheduler.scheduling_ddim_lm import DDIMLMScheduler
+
+from controlnet_aux import OpenposeDetector
+import cv2
+import numpy as np
+
+
+def main():
+    parser = argparse.ArgumentParser(description="sampling script for ControlNet-pose.")
+    parser.add_argument('--seed', type=int, default=1)
+    parser.add_argument('--num_inference_steps', type=int, default=20)
+    parser.add_argument('--guidance', type=float, default=7.5)
+    parser.add_argument('--sampler_type', type = str,default='lag')
+    parser.add_argument('--prompt', type=str, default='an asian girl')
+    parser.add_argument('--lamb', type=float, default=5.0)
+    parser.add_argument('--kappa', type=float, default=0.0)
+    parser.add_argument('--freeze', type=float, default=0.0)
+    parser.add_argument('--prompt_list', nargs='+', type=str,
+                        default=['an asian girl'])
+    parser.add_argument('--save_dir', type=str, default='/xxx/xxx/result/0402')
+    parser.add_argument('--controlnet_dir', type=str, default="lllyasviel/sd-controlnet-openpose")
+    parser.add_argument('--sd_dir', type=str, default="/xxx/xxx/stable-diffusion-v1-5")
+
+
+    args = parser.parse_args()
+    if args.sampler_type in ['bdia']:
+        parser.add_argument('--bdia_gamma', type=float, default=0.5)
+    if args.sampler_type in ['edict']:
+        parser.add_argument('--edict_p', type=float, default=0.93)
+    args = parser.parse_args()
+    device = 'cuda'
+    sampler_type = args.sampler_type
+    guidance_scale = args.guidance
+    num_inference_steps = args.num_inference_steps
+    lamb = args.lamb
+    freeze = args.freeze
+    kappa = args.kappa
+
+    save_dir = args.save_dir
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir, exist_ok=True)
+
+    # torch.manual_seed(args.seed)
+    controlnet = ControlNetModel.from_pretrained(
+        args.controlnet_dir, torch_dtype=torch.float16
+    )
+
+    control_pipe = StableDiffusionControlNetPipeline.from_pretrained(
+        args.sd_dir,
+        controlnet=controlnet, torch_dtype=torch.float16, use_safetensors=True
+    )
+    control_pipe.enable_model_cpu_offload()
+    control_pipe.safety_checker = None
+
+    if sampler_type in ['dpm_lm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+    elif sampler_type in ['dpm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = False
+    elif sampler_type in ['dpm++_lm']:
+        control_pipe.scheduler = DPMSolverMultistepLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.config.solver_order = 3
+        control_pipe.scheduler.config.algorithm_type = "dpmsolver++"
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+    elif sampler_type in ['pndm']:
+        control_pipe.scheduler = PNDMScheduler.from_config(control_pipe.scheduler.config)
+    elif sampler_type in ['ddim']:
+        control_pipe.scheduler = DDIMScheduler.from_config(control_pipe.scheduler.config)
+        # control_pipe.scheduler.lamb = lamb
+        # control_pipe.scheduler.lm = False
+        # control_pipe.scheduler.kappa = kappa
+    elif sampler_type in ['ddim_lm']:
+        control_pipe.scheduler = DDIMLMScheduler.from_config(control_pipe.scheduler.config)
+        control_pipe.scheduler.lamb = lamb
+        control_pipe.scheduler.lm = True
+        control_pipe.scheduler.kappa = kappa
+        control_pipe.scheduler.freeze = freeze
+    elif sampler_type in ['unipc']:
+        control_pipe.scheduler = UniPCMultistepScheduler.from_config(control_pipe.scheduler.config)
+
+    openpose = OpenposeDetector.from_pretrained('lllyasviel/ControlNet')
+
+    image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/person.png")
+
+    image = openpose(image)
+
+
+    for prompt, negative_prompt in [["chef in the kitchen",''],
+                                    ["Captain America", ''],
+                                    ["Spider-Man", ''],
+                                    ["Superman", ''],
+                                    ["Hulk", ''],
+                                    ["Batman", ''],
+                                    ["Iron Man", ''],
+                                    ["Deadpool", ''],
+                                    ["Winnie-the-Pooh", ''],
+                                    ["Snow White", ''],
+                                    ["Buzz Lightyear", ''],
+                                    ["Cinderella", ''],
+                                    ["Donald Duck", ''],
+                                    ["policeman", ''],
+                                    ["a doctor", ''],
+                                    ["a teacher", ''],
+                                    ['woman standing amidst a sea of wildflowers, with the warm sun shining down on her.',
+                                        ''],
+                                    ['a stunning Arabic woman dressed in traditional clothing', ''],
+                                    ['a stunning Asian woman dressed in traditional clothing', ''],
+                                    ]:
+        for seed in range(15):
+            torch.manual_seed(seed)
+            res = control_pipe(
+                prompt = prompt, image=image, num_inference_steps=num_inference_steps,
+            ).images[0]
+
+            res.save(os.path.join(save_dir,
+                                  f"{prompt[:20]}_seed{seed}_{sampler_type}_infer{num_inference_steps}_g{guidance_scale}_lamb{args.lamb}.png"))
+
+
+
+if __name__ == '__main__':
+    main()