Upload 22 files

README.md

@@ -0,0 +1,14 @@
+---
+title: Train Dreambooth LoRa Sd-XL
+emoji: 🏆
+colorFrom: red
+colorTo: red
+python_version: 3.10.12
+sdk: gradio
+sdk_version: 3.44.2
+app_file: app.py
+pinned: false
+suggested_hardware: "a10g-small"
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,104 @@
+import gradio as gr
+import torch
+import os
+import shutil
+import requests
+import subprocess
+from subprocess import getoutput
+from huggingface_hub import login, HfFileSystem, snapshot_download, HfApi, create_repo
+from pathlib import Path
+from PIL import Image
+
+from app_train import create_training_demo
+from sdxl.app_inference import create_inference_demo
+from depthgltf.app_visualisations import create_visual_demo 
+
+from transformers import DPTFeatureExtractor, DPTForDepthEstimation
+import numpy as np
+import open3d as o3d
+
+
+css="""
+#col-container {max-width: 780px; margin-left: auto; margin-right: auto;}
+#upl-dataset-group {background-color: none!important;}
+
+div#warning-ready {
+    background-color: #ecfdf5;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+div#warning-ready > .gr-prose > h2, div#warning-ready > .gr-prose > p {
+    color: #057857!important;
+}
+
+div#warning-duplicate {
+    background-color: #ebf5ff;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+
+div#warning-duplicate > .gr-prose > h2, div#warning-duplicate > .gr-prose > p {
+    color: #0f4592!important;
+}
+
+div#warning-duplicate strong {
+    color: #0f4592;
+}
+
+p.actions {
+    display: flex;
+    align-items: center;
+    margin: 20px 0;
+}
+
+div#warning-duplicate .actions a {
+    display: inline-block;
+    margin-right: 10px;
+}
+
+div#warning-setgpu {
+    background-color: #fff4eb;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+
+div#warning-setgpu > .gr-prose > h2, div#warning-setgpu > .gr-prose > p {
+    color: #92220f!important;
+}
+
+div#warning-setgpu a, div#warning-setgpu b {
+    color: #91230f;
+}
+
+div#warning-setgpu p.actions > a {
+    display: inline-block;
+    background: #1f1f23;
+    border-radius: 40px;
+    padding: 6px 24px;
+    color: antiquewhite;
+    text-decoration: none;
+    font-weight: 600;
+    font-size: 1.2em;
+}
+
+button#load-dataset-btn{
+min-height: 60px;
+}
+"""
+
+
+with gr.Blocks(css=css) as demo:
+
+    gr.Markdown("SUTD x SUNS Shop Design Generator")
+    with gr.Tab("Training"):
+         create_training_demo()
+    with gr.Tab("Generation"):
+        create_inference_demo()
+       
+        #create_visual_demo(); 
+    with gr.Tab("Visualisation"):
+        create_visual_demo(); 
+
+
+demo.queue().launch(debug=True, share=True)
+

app_train.py

@@ -0,0 +1,389 @@
+import gradio as gr
+import torch
+import os
+import shutil
+import requests
+import subprocess
+from subprocess import getoutput
+from huggingface_hub import login, HfFileSystem, snapshot_download, HfApi, create_repo
+
+is_gpu_associated = torch.cuda.is_available()
+
+is_shared_ui = False
+
+hf_token = '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'
+
+fs = HfFileSystem(token=hf_token)
+api = HfApi()
+
+if is_gpu_associated:
+    gpu_info = getoutput('nvidia-smi')
+    if("A10G" in gpu_info):
+        which_gpu = "A10G"
+    elif("T4" in gpu_info):
+        which_gpu = "T4"
+    else:
+        which_gpu = "CPU"
+
+def check_upload_or_no(value):
+    if value is True:
+        return gr.update(visible=True)
+    else:
+        return gr.update(visible=False)
+
+def load_images_to_dataset(images, dataset_name):
+
+    if is_shared_ui:
+        raise gr.Error("This Space only works in duplicated instances")
+
+    if dataset_name == "":
+        raise gr.Error("You forgot to name your new dataset. ")
+
+    # Create the directory if it doesn't exist
+    my_working_directory = f"my_working_directory_for_{dataset_name}"
+    if not os.path.exists(my_working_directory):
+        os.makedirs(my_working_directory)
+
+    # Assuming 'images' is a list of image file paths
+    for idx, image in enumerate(images):
+        # Get the base file name (without path) from the original location
+        image_name = os.path.basename(image.name)
+    
+        # Construct the destination path in the working directory
+        destination_path = os.path.join(my_working_directory, image_name)
+    
+        # Copy the image from the original location to the working directory
+        shutil.copy(image.name, destination_path)
+    
+        # Print the image name and its corresponding save path
+        print(f"Image {idx + 1}: {image_name} copied to {destination_path}")
+   
+    path_to_folder = my_working_directory
+    your_username = api.whoami(token=hf_token)["name"]
+    repo_id = f"{your_username}/{dataset_name}"
+    create_repo(repo_id=repo_id, repo_type="dataset", token=hf_token)
+    
+    api.upload_folder(
+        folder_path=path_to_folder,
+        repo_id=repo_id,
+        repo_type="dataset",
+        token=hf_token
+    )
+
+    return "Done, your dataset is ready and loaded for the training step!", repo_id
+
+def swap_hardware(hf_token, hardware="cpu-basic"):
+    hardware_url = f"https://huggingface.co/spaces/ClaireOzzz/train-dreambooth-lora-sdxl/hardware"
+    headers = { "authorization" : f"Bearer {hf_token}"}
+    body = {'flavor': hardware}
+    requests.post(hardware_url, json = body, headers=headers)
+
+def swap_sleep_time(hf_token,sleep_time):
+    sleep_time_url = f"https://huggingface.co/api/spaces/ClaireOzzz/train-dreambooth-lora-sdxl/sleeptime"
+    headers = { "authorization" : f"Bearer {hf_token}"}
+    body = {'seconds':sleep_time}
+    requests.post(sleep_time_url,json=body,headers=headers)
+
+def get_sleep_time(hf_token):
+    sleep_time_url = f"https://huggingface.co/api/spaces/ClaireOzzz/train-dreambooth-lora-sdxl"
+    headers = { "authorization" : f"Bearer {hf_token}"}
+    response = requests.get(sleep_time_url,headers=headers)
+    try:
+        gcTimeout = response.json()['runtime']['gcTimeout']
+    except:
+        gcTimeout = None
+    return gcTimeout
+
+def write_to_community(title, description,hf_token): 
+    
+    api.create_discussion(repo_id=os.environ['ClaireOzzz/train-dreambooth-lora-sdxl'], title=title, description=description,repo_type="space", token=hf_token)
+
+
+def set_accelerate_default_config():
+    try:
+        subprocess.run(["accelerate", "config", "default"], check=True)
+        print("Accelerate default config set successfully!")
+    except subprocess.CalledProcessError as e:
+        print(f"An error occurred: {e}")
+
+def train_dreambooth_lora_sdxl(dataset_id, instance_data_dir, lora_trained_xl_folder, instance_prompt, max_train_steps, checkpoint_steps, remove_gpu):
+    
+    script_filename = "train_dreambooth_lora_sdxl.py"  # Assuming it's in the same folder
+
+    command = [
+        "accelerate",
+        "launch",
+        script_filename,  # Use the local script
+        "--pretrained_model_name_or_path=stabilityai/stable-diffusion-xl-base-1.0",
+        "--pretrained_vae_model_name_or_path=madebyollin/sdxl-vae-fp16-fix",
+        f"--dataset_id={dataset_id}",
+        f"--instance_data_dir={instance_data_dir}",
+        f"--output_dir={lora_trained_xl_folder}",
+        "--mixed_precision=fp16",
+        f"--instance_prompt={instance_prompt}",
+        "--resolution=1024",
+        "--train_batch_size=2",
+        "--gradient_accumulation_steps=2",
+        "--gradient_checkpointing",
+        "--learning_rate=1e-4",
+        "--lr_scheduler=constant",
+        "--lr_warmup_steps=0",
+        "--enable_xformers_memory_efficient_attention",
+        "--mixed_precision=fp16",
+        "--use_8bit_adam",
+        f"--max_train_steps={max_train_steps}",
+        f"--checkpointing_steps={checkpoint_steps}",
+        "--seed=0",
+        "--push_to_hub",
+        f"--hub_token={hf_token}"
+    ]
+
+    try:
+        subprocess.run(command, check=True)
+        print("Training is finished!")
+        if remove_gpu:
+            swap_hardware(hf_token, "cpu-basic")
+        else:
+            swap_sleep_time(hf_token, 300)
+    except subprocess.CalledProcessError as e:
+        print(f"An error occurred: {e}")
+        
+        title="There was an error on during your training"
+        description=f'''
+        Unfortunately there was an error during training your {lora_trained_xl_folder} model. 
+        Please check it out below. Feel free to report this issue to [SD-XL Dreambooth LoRa Training](https://huggingface.co/spaces/fffiloni/train-dreambooth-lora-sdxl): 
+        ```
+        {str(e)}
+        ```
+        '''
+        if remove_gpu:
+            swap_hardware(hf_token, "cpu-basic")
+        else:
+            swap_sleep_time(hf_token, 300)
+        #write_to_community(title,description,hf_token)
+
+def main(dataset_id, 
+         lora_trained_xl_folder,
+         instance_prompt,
+         max_train_steps,
+         checkpoint_steps,
+         remove_gpu):
+
+    
+    if is_shared_ui:
+        raise gr.Error("This Space only works in duplicated instances")
+
+    if not is_gpu_associated:
+        raise gr.Error("Please associate a T4 or A10G GPU for this Space")
+
+    if dataset_id == "":
+        raise gr.Error("You forgot to specify an image dataset")
+
+    if instance_prompt == "":
+        raise gr.Error("You forgot to specify a concept prompt")
+
+    if lora_trained_xl_folder == "":
+        raise gr.Error("You forgot to name the output folder for your model")
+
+    sleep_time = get_sleep_time(hf_token)
+    if sleep_time:
+        swap_sleep_time(hf_token, -1)
+
+    gr.Warning("If you did not check the `Remove GPU After training`, don't forget to remove the GPU attribution after you are done. ")
+        
+    dataset_repo = dataset_id
+
+    # Automatically set local_dir based on the last part of dataset_repo
+    repo_parts = dataset_repo.split("/")
+    local_dir = f"./{repo_parts[-1]}"  # Use the last part of the split
+
+    # Check if the directory exists and create it if necessary
+    if not os.path.exists(local_dir):
+        os.makedirs(local_dir)
+
+    gr.Info("Downloading dataset ...")
+    
+    snapshot_download(
+        dataset_repo,
+        local_dir=local_dir,
+        repo_type="dataset",
+        ignore_patterns=".gitattributes",
+        token=hf_token
+    )
+
+    set_accelerate_default_config()
+
+    gr.Info("Training begins ...")
+
+    instance_data_dir = repo_parts[-1]
+    train_dreambooth_lora_sdxl(dataset_id, instance_data_dir, lora_trained_xl_folder, instance_prompt, max_train_steps, checkpoint_steps, remove_gpu)
+    
+    your_username = api.whoami(token=hf_token)["name"]
+    return f"Done, your trained model has been stored in your models library: {your_username}/{lora_trained_xl_folder}"
+
+css="""
+#col-container {max-width: 780px; margin-left: auto; margin-right: auto;}
+#upl-dataset-group {background-color: none!important;}
+
+div#warning-ready {
+    background-color: #ecfdf5;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+div#warning-ready > .gr-prose > h2, div#warning-ready > .gr-prose > p {
+    color: #057857!important;
+}
+
+div#warning-duplicate {
+    background-color: #ebf5ff;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+
+div#warning-duplicate > .gr-prose > h2, div#warning-duplicate > .gr-prose > p {
+    color: #0f4592!important;
+}
+
+div#warning-duplicate strong {
+    color: #0f4592;
+}
+
+p.actions {
+    display: flex;
+    align-items: center;
+    margin: 20px 0;
+}
+
+div#warning-duplicate .actions a {
+    display: inline-block;
+    margin-right: 10px;
+}
+
+div#warning-setgpu {
+    background-color: #fff4eb;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+
+div#warning-setgpu > .gr-prose > h2, div#warning-setgpu > .gr-prose > p {
+    color: #92220f!important;
+}
+
+div#warning-setgpu a, div#warning-setgpu b {
+    color: #91230f;
+}
+
+div#warning-setgpu p.actions > a {
+    display: inline-block;
+    background: #1f1f23;
+    border-radius: 40px;
+    padding: 6px 24px;
+    color: antiquewhite;
+    text-decoration: none;
+    font-weight: 600;
+    font-size: 1.2em;
+}
+
+button#load-dataset-btn{
+min-height: 60px;
+}
+"""
+def create_training_demo() -> gr.Blocks:
+  with gr.Blocks(css=css) as demo:
+    with gr.Column(elem_id="col-container"):
+        if is_shared_ui:
+            top_description = gr.HTML(f'''
+                <div class="gr-prose">
+                    <h2><svg xmlns="http://www.w3.org/2000/svg" width="18px" height="18px" style="margin-right: 0px;display: inline-block;"fill="none"><path fill="#fff" d="M7 13.2a6.3 6.3 0 0 0 4.4-10.7A6.3 6.3 0 0 0 .6 6.9 6.3 6.3 0 0 0 7 13.2Z"/><path fill="#fff" fill-rule="evenodd" d="M7 0a6.9 6.9 0 0 1 4.8 11.8A6.9 6.9 0 0 1 0 7 6.9 6.9 0 0 1 7 0Zm0 0v.7V0ZM0 7h.6H0Zm7 6.8v-.6.6ZM13.7 7h-.6.6ZM9.1 1.7c-.7-.3-1.4-.4-2.2-.4a5.6 5.6 0 0 0-4 1.6 5.6 5.6 0 0 0-1.6 4 5.6 5.6 0 0 0 1.6 4 5.6 5.6 0 0 0 4 1.7 5.6 5.6 0 0 0 4-1.7 5.6 5.6 0 0 0 1.7-4 5.6 5.6 0 0 0-1.7-4c-.5-.5-1.1-.9-1.8-1.2Z" clip-rule="evenodd"/><path fill="#000" fill-rule="evenodd" d="M7 2.9a.8.8 0 1 1 0 1.5A.8.8 0 0 1 7 3ZM5.8 5.7c0-.4.3-.6.6-.6h.7c.3 0 .6.2.6.6v3.7h.5a.6.6 0 0 1 0 1.3H6a.6.6 0 0 1 0-1.3h.4v-3a.6.6 0 0 1-.6-.7Z" clip-rule="evenodd"/></svg>
+                    Attention: this Space need to be duplicated to work</h2>
+                    <p class="main-message">
+                        To make it work, <strong>duplicate the Space</strong> and run it on your own profile using a <strong>private</strong> GPU (T4-small or A10G-small).<br />
+                        A T4 costs <strong>US$0.60/h</strong>, so it should cost < US$1 to train most models.
+                    </p>
+                    <p class="actions">
+                       
+                        to start training your own image model
+                    </p>
+                </div>
+            ''', elem_id="warning-duplicate")
+        # else:
+        #     if(is_gpu_associated):
+        #         top_description = gr.HTML(f'''
+        #                 <div class="gr-prose">
+        #                     <h2><svg xmlns="http://www.w3.org/2000/svg" width="18px" height="18px" style="margin-right: 0px;display: inline-block;"fill="none"><path fill="#fff" d="M7 13.2a6.3 6.3 0 0 0 4.4-10.7A6.3 6.3 0 0 0 .6 6.9 6.3 6.3 0 0 0 7 13.2Z"/><path fill="#fff" fill-rule="evenodd" d="M7 0a6.9 6.9 0 0 1 4.8 11.8A6.9 6.9 0 0 1 0 7 6.9 6.9 0 0 1 7 0Zm0 0v.7V0ZM0 7h.6H0Zm7 6.8v-.6.6ZM13.7 7h-.6.6ZM9.1 1.7c-.7-.3-1.4-.4-2.2-.4a5.6 5.6 0 0 0-4 1.6 5.6 5.6 0 0 0-1.6 4 5.6 5.6 0 0 0 1.6 4 5.6 5.6 0 0 0 4 1.7 5.6 5.6 0 0 0 4-1.7 5.6 5.6 0 0 0 1.7-4 5.6 5.6 0 0 0-1.7-4c-.5-.5-1.1-.9-1.8-1.2Z" clip-rule="evenodd"/><path fill="#000" fill-rule="evenodd" d="M7 2.9a.8.8 0 1 1 0 1.5A.8.8 0 0 1 7 3ZM5.8 5.7c0-.4.3-.6.6-.6h.7c.3 0 .6.2.6.6v3.7h.5a.6.6 0 0 1 0 1.3H6a.6.6 0 0 1 0-1.3h.4v-3a.6.6 0 0 1-.6-.7Z" clip-rule="evenodd"/></svg>
+        #                     You have successfully associated a {which_gpu} GPU to the SD-XL Training Space 🎉</h2>
+        #                     <p>
+        #                         You can now train your model! You will be billed by the minute from when you activated the GPU until when it is turned off.
+        #                     </p> 
+        #                 </div>
+        #         ''', elem_id="warning-ready")
+        #     else:
+        #         top_description = gr.HTML(f'''
+        #                 <div class="gr-prose">
+        #                 <h2><svg xmlns="http://www.w3.org/2000/svg" width="18px" height="18px" style="margin-right: 0px;display: inline-block;"fill="none"><path fill="#fff" d="M7 13.2a6.3 6.3 0 0 0 4.4-10.7A6.3 6.3 0 0 0 .6 6.9 6.3 6.3 0 0 0 7 13.2Z"/><path fill="#fff" fill-rule="evenodd" d="M7 0a6.9 6.9 0 0 1 4.8 11.8A6.9 6.9 0 0 1 0 7 6.9 6.9 0 0 1 7 0Zm0 0v.7V0ZM0 7h.6H0Zm7 6.8v-.6.6ZM13.7 7h-.6.6ZM9.1 1.7c-.7-.3-1.4-.4-2.2-.4a5.6 5.6 0 0 0-4 1.6 5.6 5.6 0 0 0-1.6 4 5.6 5.6 0 0 0 1.6 4 5.6 5.6 0 0 0 4 1.7 5.6 5.6 0 0 0 4-1.7 5.6 5.6 0 0 0 1.7-4 5.6 5.6 0 0 0-1.7-4c-.5-.5-1.1-.9-1.8-1.2Z" clip-rule="evenodd"/><path fill="#000" fill-rule="evenodd" d="M7 2.9a.8.8 0 1 1 0 1.5A.8.8 0 0 1 7 3ZM5.8 5.7c0-.4.3-.6.6-.6h.7c.3 0 .6.2.6.6v3.7h.5a.6.6 0 0 1 0 1.3H6a.6.6 0 0 1 0-1.3h.4v-3a.6.6 0 0 1-.6-.7Z" clip-rule="evenodd"/></svg>
+        #                 You have successfully duplicated the SD-XL Training Space 🎉</h2>
+        #                 <p>There's only one step left before you can train your model: <a href="https://huggingface.co/spaces/{os.environ['SPACE_ID']}/settings" style="text-decoration: underline" target="_blank">attribute a <b>T4-small or A10G-small GPU</b> to it (via the Settings tab)</a> and run the training below.
+        #                 You will be billed by the minute from when you activate the GPU until when it is turned off.</p> 
+        #                 <p class="actions">
+        #                     <a href="https://huggingface.co/spaces/ClaireOzzz/train-dreambooth-lora-sdxl/settings">🔥 &nbsp; Set recommended GPU</a>
+        #                 </p>
+        #                 </div>
+        #         ''', elem_id="warning-setgpu")
+        
+        gr.Markdown("# SD-XL Dreambooth LoRa Training UI 💭")
+        
+        upload_my_images = gr.Checkbox(label="Drop your training images ? (optional)", value=False)
+        gr.Markdown("Use this step to upload your training images and create a new dataset. If you already have a dataset stored on your HF profile, you can skip this step, and provide your dataset ID in the training `Datased ID` input below.")
+         
+        with gr.Group(visible=False, elem_id="upl-dataset-group") as upload_group:
+            with gr.Row():
+                images = gr.File(file_types=["image"], label="Upload your images", file_count="multiple", interactive=True, visible=True)
+                with gr.Column():
+                    new_dataset_name = gr.Textbox(label="Set new dataset name", placeholder="e.g.: my_awesome_dataset")
+                    dataset_status = gr.Textbox(label="dataset status")
+                    load_btn = gr.Button("Load images to new dataset", elem_id="load-dataset-btn")
+        
+        gr.Markdown("## Training ")
+        gr.Markdown("You can use an existing image dataset, find a dataset example here: [https://huggingface.co/datasets/diffusers/dog-example](https://huggingface.co/datasets/diffusers/dog-example) ;)")
+        
+        with gr.Row():
+            dataset_id = gr.Textbox(label="Dataset ID", info="use one of your previously uploaded image datasets on your HF profile", placeholder="diffusers/dog-example")
+            instance_prompt = gr.Textbox(label="Concept prompt", info="concept prompt - use a unique, made up word to avoid collisions")
+        
+        with gr.Row():
+            model_output_folder = gr.Textbox(label="Output model folder name", placeholder="lora-trained-xl-folder")
+            max_train_steps = gr.Number(label="Max Training Steps", value=500, precision=0, step=10)
+            checkpoint_steps = gr.Number(label="Checkpoints Steps", value=100, precision=0, step=10)
+
+        remove_gpu = gr.Checkbox(label="Remove GPU After Training", value=True, info="If NOT enabled, don't forget to remove the GPU attribution after you are done.")
+        train_button = gr.Button("Train !")
+
+        train_status = gr.Textbox(label="Training status")
+
+    upload_my_images.change(
+        fn = check_upload_or_no,
+        inputs =[upload_my_images],
+        outputs = [upload_group]        
+    )
+    
+    load_btn.click(
+        fn = load_images_to_dataset,
+        inputs = [images, new_dataset_name],
+        outputs = [dataset_status, dataset_id]
+    )
+    
+    train_button.click(
+        fn = main,
+        inputs = [
+            dataset_id,
+            model_output_folder,
+            instance_prompt,
+            max_train_steps,
+            checkpoint_steps,
+            remove_gpu
+        ],
+        outputs = [train_status]
+    )
+  return demo
+
+#demo.launch(debug=True, share=True)

depthgltf/.gitattributes

@@ -0,0 +1,27 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zstandard filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

depthgltf/.gitignore

@@ -0,0 +1,46 @@
+# Python build
+.eggs/
+gradio.egg-info/*
+!gradio.egg-info/requires.txt
+!gradio.egg-info/PKG-INFO
+dist/
+*.pyc
+__pycache__/
+*.py[cod]
+*$py.class
+build/
+
+# JS build
+gradio/templates/frontend
+# Secrets
+.env
+
+# Gradio run artifacts
+*.db
+*.sqlite3
+gradio/launches.json
+flagged/
+gradio_cached_examples/
+
+# Tests
+.coverage
+coverage.xml
+test.txt
+
+# Demos
+demo/tmp.zip
+demo/files/*.avi
+demo/files/*.mp4
+
+# Etc
+.idea/*
+.DS_Store
+*.bak
+workspace.code-workspace
+*.h5
+.vscode/
+
+# log files
+.pnpm-debug.log
+venv/
+*.db-journal

depthgltf/README.md

@@ -0,0 +1,12 @@
+---
+title: Dpt Depth Estimation + 3D
+emoji: ⚡
+colorFrom: blue
+colorTo: red
+sdk: gradio
+sdk_version: 3.0b8
+app_file: app.py
+pinned: false
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces#reference

depthgltf/app_visualisations.py

@@ -0,0 +1,133 @@
+import gradio as gr
+from transformers import DPTFeatureExtractor, DPTForDepthEstimation
+import torch
+import numpy as np
+from PIL import Image
+import open3d as o3d
+from pathlib import Path
+import os
+
+feature_extractor = DPTFeatureExtractor.from_pretrained("Intel/dpt-large")
+model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large")
+
+
+def process_image(image_path):
+    image_path = Path(image_path)
+    image_raw = Image.open(image_path)
+    image = image_raw.resize(
+        (800, int(800 * image_raw.size[1] / image_raw.size[0])),
+        Image.Resampling.LANCZOS)
+
+    # prepare image for the model
+    encoding = feature_extractor(image, return_tensors="pt")
+
+    # forward pass
+    with torch.no_grad():
+        outputs = model(**encoding)
+        predicted_depth = outputs.predicted_depth
+
+    # interpolate to original size
+    prediction = torch.nn.functional.interpolate(
+        predicted_depth.unsqueeze(1),
+        size=image.size[::-1],
+        mode="bicubic",
+        align_corners=False,
+    ).squeeze()
+    output = prediction.cpu().numpy()
+    depth_image = (output * 255 / np.max(output)).astype('uint8')
+    try:
+        gltf_path = create_3d_obj(np.array(image), depth_image, image_path)
+        img = Image.fromarray(depth_image)
+        return [img, gltf_path, gltf_path]
+    except Exception as e:
+        gltf_path = create_3d_obj(
+            np.array(image), depth_image, image_path, depth=8)
+        img = Image.fromarray(depth_image)
+        return [img, gltf_path, gltf_path]
+    except:
+        print("Error reconstructing 3D model")
+        raise Exception("Error reconstructing 3D model")
+
+
+def create_3d_obj(rgb_image, depth_image, image_path, depth=10):
+    depth_o3d = o3d.geometry.Image(depth_image)
+    image_o3d = o3d.geometry.Image(rgb_image)
+    rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(
+        image_o3d, depth_o3d, convert_rgb_to_intensity=False)
+    w = int(depth_image.shape[1])
+    h = int(depth_image.shape[0])
+
+    camera_intrinsic = o3d.camera.PinholeCameraIntrinsic()
+    camera_intrinsic.set_intrinsics(w, h, 500, 500, w/2, h/2)
+
+    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
+        rgbd_image, camera_intrinsic)
+
+    print('normals')
+    pcd.normals = o3d.utility.Vector3dVector(
+        np.zeros((1, 3)))  # invalidate existing normals
+    pcd.estimate_normals(
+        search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.01, max_nn=30))
+    pcd.orient_normals_towards_camera_location(
+        camera_location=np.array([0., 0., 1000.]))
+    pcd.transform([[1, 0, 0, 0],
+                   [0, -1, 0, 0],
+                   [0, 0, -1, 0],
+                   [0, 0, 0, 1]])
+    pcd.transform([[-1, 0, 0, 0],
+                   [0, 1, 0, 0],
+                   [0, 0, 1, 0],
+                   [0, 0, 0, 1]])
+
+    print('run Poisson surface reconstruction')
+    with o3d.utility.VerbosityContextManager(o3d.utility.VerbosityLevel.Debug) as cm:
+        mesh_raw, densities = o3d.geometry.TriangleMesh.create_from_point_cloud_poisson(
+            pcd, depth=depth, width=0, scale=1.1, linear_fit=True)
+
+    voxel_size = max(mesh_raw.get_max_bound() - mesh_raw.get_min_bound()) / 256
+    print(f'voxel_size = {voxel_size:e}')
+    mesh = mesh_raw.simplify_vertex_clustering(
+        voxel_size=voxel_size,
+        contraction=o3d.geometry.SimplificationContraction.Average)
+
+    # vertices_to_remove = densities < np.quantile(densities, 0.001)
+    # mesh.remove_vertices_by_mask(vertices_to_remove)
+    bbox = pcd.get_axis_aligned_bounding_box()
+    mesh_crop = mesh.crop(bbox)
+    gltf_path = f'./{image_path.stem}.gltf'
+    o3d.io.write_triangle_mesh(
+        gltf_path, mesh_crop, write_triangle_uvs=True)
+    return gltf_path
+
+
+current_directory = os.path.dirname(__file__)
+
+title = "Demo: zero-shot depth estimation with DPT + 3D Point Cloud"
+description = "This demo is a variation from the original <a href='https://huggingface.co/spaces/nielsr/dpt-depth-estimation' target='_blank'>DPT Demo</a>. It uses the DPT model to predict the depth of an image and then uses 3D Point Cloud to create a 3D object."
+#examples = [["examples/" + img] for img in os.listdir("examples/")]
+
+# result_image_path = os.path.join(current_directory, '..', 'result.png')
+# image_path = Path(result_image_path)
+
+
+# Load the image
+# rawimage = Image.open(image_path)
+# image_r = gr.Image(value=rawimage, type="pil", label="Input Image")
+#image_r.change(create_visual_demo, [],[])
+ 
+def create_visual_demo():
+  iface = gr.Interface(fn=process_image,
+    inputs=[gr.Image(
+        type="filepath", label="Input Image")],
+    outputs=[gr.Image(label="predicted depth", type="pil"),
+              gr.Model3D(label="3d mesh reconstruction", clear_color=[
+                                1.0, 1.0, 1.0, 1.0]),
+              gr.File(label="3d gLTF")],
+    title=title,
+    description=description,
+    #examples=examples,
+    live=True,
+    allow_flagging="never",
+    cache_examples=False)  
+
+#iface.launch(debug=True, enable_queue=False, share=True)

depthgltf/packages.txt

@@ -0,0 +1 @@
+libgl1-mesa-glx

depthgltf/requirements.txt

@@ -0,0 +1,7 @@
+torch
+git+https://github.com/nielsrogge/transformers.git@add_dpt_redesign#egg=transformers
+numpy
+Pillow
+gradio
+jinja2
+open3d

requirements.txt

@@ -0,0 +1,10 @@
+torch==2.0.1
+torchvision==0.15.2
+xformers
+bitsandbytes
+transformers
+accelerate
+git+https://github.com/huggingface/diffusers.git
+scipy
+huggingface_hub
+datasets

sdxl/.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

sdxl/README.md

@@ -0,0 +1,12 @@
+---
+title: SD-XL + Control LoRas
+emoji: 🦀
+colorFrom: green
+colorTo: indigo
+sdk: gradio
+sdk_version: 3.44.4
+app_file: app.py
+pinned: false
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

sdxl/app_inference.py

@@ -0,0 +1,363 @@
+import gradio as gr
+from huggingface_hub import login, HfFileSystem, HfApi, ModelCard
+import os
+import spaces
+import random
+import torch
+
+is_shared_ui = False
+
+hf_token = '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'
+login(token=hf_token)
+
+fs = HfFileSystem(token=hf_token)
+api = HfApi()
+
+device="cuda" if torch.cuda.is_available() else "cpu"
+
+from diffusers import ControlNetModel, StableDiffusionXLControlNetPipeline, AutoencoderKL
+from diffusers.utils import load_image
+from PIL import Image
+import torch
+import numpy as np
+import cv2
+
+vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+
+controlnet = ControlNetModel.from_pretrained(
+    "diffusers/controlnet-canny-sdxl-1.0",
+    torch_dtype=torch.float16
+)
+
+def check_use_custom_or_no(value):
+    if value is True:
+        return gr.update(visible=True)
+    else:
+        return gr.update(visible=False)
+
+def get_files(file_paths):
+    last_files = {}  # Dictionary to store the last file for each path
+
+    for file_path in file_paths:
+        # Split the file path into directory and file components
+        directory, file_name = file_path.rsplit('/', 1)
+    
+        # Update the last file for the current path
+        last_files[directory] = file_name
+    
+    # Extract the last files from the dictionary
+    result = list(last_files.values())
+
+    return result
+
+def load_model(model_name):
+
+    if model_name == "":
+        gr.Warning("If you want to use a private model, you need to duplicate this space on your personal account.")
+        raise gr.Error("You forgot to define Model ID.")
+
+    # Get instance_prompt a.k.a trigger word
+    card = ModelCard.load(model_name)
+    repo_data = card.data.to_dict()
+    instance_prompt = repo_data.get("instance_prompt")
+
+    if instance_prompt is not None:
+        print(f"Trigger word: {instance_prompt}")
+    else:
+        instance_prompt = "no trigger word needed"
+        print(f"Trigger word: no trigger word needed")
+
+    # List all ".safetensors" files in repo
+    sfts_available_files = fs.glob(f"{model_name}/*safetensors")
+    sfts_available_files = get_files(sfts_available_files)
+
+    if sfts_available_files == []:
+        sfts_available_files = ["NO SAFETENSORS FILE"]
+
+    print(f"Safetensors available: {sfts_available_files}")
+
+    return model_name, "Model Ready", gr.update(choices=sfts_available_files, value=sfts_available_files[0], visible=True), gr.update(value=instance_prompt, visible=True)
+
+def custom_model_changed(model_name, previous_model):
+    if model_name == "" and previous_model == "" :
+        status_message = ""      
+    elif model_name != previous_model:
+        status_message = "model changed, please reload before any new run"
+    else:
+        status_message = "model ready"
+    return status_message
+
+def resize_image(input_path, output_path, target_height):
+    # Open the input image
+    img = Image.open(input_path)
+
+    # Calculate the aspect ratio of the original image
+    original_width, original_height = img.size
+    original_aspect_ratio = original_width / original_height
+
+    # Calculate the new width while maintaining the aspect ratio and the target height
+    new_width = int(target_height * original_aspect_ratio)
+
+    # Resize the image while maintaining the aspect ratio and fixing the height
+    img = img.resize((new_width, target_height), Image.LANCZOS)
+
+    # Save the resized image
+    img.save(output_path)
+
+    return output_path
+
+@spaces.GPU
+def infer(use_custom_model, model_name, weight_name, custom_lora_weight, image_in, prompt, negative_prompt, preprocessor, controlnet_conditioning_scale, guidance_scale, inf_steps, seed, progress=gr.Progress(track_tqdm=True)):
+
+    pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
+        "stabilityai/stable-diffusion-xl-base-1.0",
+        controlnet=controlnet,
+        vae=vae,
+        torch_dtype=torch.float16, 
+        variant="fp16",
+        use_safetensors=True
+    )
+    
+    pipe.to(device)
+    
+    prompt = prompt
+    negative_prompt = negative_prompt
+
+    if seed < 0 :
+        seed = random.randint(0, 423538377342)
+    
+    generator = torch.Generator(device=device).manual_seed(seed)
+
+    if image_in == None:
+        raise gr.Error("You forgot to upload a source image.")
+    
+    image_in = resize_image(image_in, "resized_input.jpg", 1024)
+    
+    if preprocessor == "canny":
+
+        image = load_image(image_in)
+
+        image = np.array(image)
+        image = cv2.Canny(image, 100, 200)
+        image = image[:, :, None]
+        image = np.concatenate([image, image, image], axis=2)
+        image = Image.fromarray(image)
+    
+    if use_custom_model:
+        
+        if model_name == "":
+            raise gr.Error("you forgot to set a custom model name.")
+        
+        custom_model = model_name
+
+        # This is where you load your trained weights
+        if weight_name == "NO SAFETENSORS FILE": 
+            pipe.load_lora_weights(
+                custom_model,     
+                low_cpu_mem_usage = True,
+                use_auth_token = True
+            )
+    
+        else:
+            pipe.load_lora_weights(
+                custom_model,
+                weight_name = weight_name,        
+                low_cpu_mem_usage = True,
+                use_auth_token = True
+            )
+    
+        lora_scale=custom_lora_weight
+
+        images = pipe(
+            prompt, 
+            negative_prompt=negative_prompt, 
+            image=image, 
+            controlnet_conditioning_scale=float(controlnet_conditioning_scale),
+            guidance_scale = float(guidance_scale),
+            num_inference_steps=inf_steps,
+            generator=generator,
+            cross_attention_kwargs={"scale": lora_scale}
+        ).images
+    else:
+        images = pipe(
+            prompt, 
+            negative_prompt=negative_prompt, 
+            image=image, 
+            controlnet_conditioning_scale=float(controlnet_conditioning_scale),
+            guidance_scale = float(guidance_scale),
+            num_inference_steps=inf_steps,
+            generator=generator,
+        ).images
+
+    images[0].save(f"result.png")
+
+    return f"result.png", seed
+
+css="""
+#col-container{
+    margin: 0 auto;
+    max-width: 720px;
+    text-align: left;
+}
+div#warning-duplicate {
+    background-color: #ebf5ff;
+    padding: 0 10px 5px;
+    margin: 20px 0;
+}
+div#warning-duplicate > .gr-prose > h2, div#warning-duplicate > .gr-prose > p {
+    color: #0f4592!important;
+}
+div#warning-duplicate strong {
+    color: #0f4592;
+}
+p.actions {
+    display: flex;
+    align-items: center;
+    margin: 20px 0;
+}
+div#warning-duplicate .actions a {
+    display: inline-block;
+    margin-right: 10px;
+}
+button#load_model_btn{
+    height: 46px;
+}
+#status_info{
+    font-size: 0.9em;
+}
+"""
+def create_inference_demo() -> gr.Blocks:
+  
+  with gr.Blocks(css=css) as demo:
+      with gr.Column(elem_id="col-container"):
+          if is_shared_ui:
+              top_description = gr.HTML(f'''
+                  <div class="gr-prose">
+                      <h2><svg xmlns="http://www.w3.org/2000/svg" width="18px" height="18px" style="margin-right: 0px;display: inline-block;"fill="none"><path fill="#fff" d="M7 13.2a6.3 6.3 0 0 0 4.4-10.7A6.3 6.3 0 0 0 .6 6.9 6.3 6.3 0 0 0 7 13.2Z"/><path fill="#fff" fill-rule="evenodd" d="M7 0a6.9 6.9 0 0 1 4.8 11.8A6.9 6.9 0 0 1 0 7 6.9 6.9 0 0 1 7 0Zm0 0v.7V0ZM0 7h.6H0Zm7 6.8v-.6.6ZM13.7 7h-.6.6ZM9.1 1.7c-.7-.3-1.4-.4-2.2-.4a5.6 5.6 0 0 0-4 1.6 5.6 5.6 0 0 0-1.6 4 5.6 5.6 0 0 0 1.6 4 5.6 5.6 0 0 0 4 1.7 5.6 5.6 0 0 0 4-1.7 5.6 5.6 0 0 0 1.7-4 5.6 5.6 0 0 0-1.7-4c-.5-.5-1.1-.9-1.8-1.2Z" clip-rule="evenodd"/><path fill="#000" fill-rule="evenodd" d="M7 2.9a.8.8 0 1 1 0 1.5A.8.8 0 0 1 7 3ZM5.8 5.7c0-.4.3-.6.6-.6h.7c.3 0 .6.2.6.6v3.7h.5a.6.6 0 0 1 0 1.3H6a.6.6 0 0 1 0-1.3h.4v-3a.6.6 0 0 1-.6-.7Z" clip-rule="evenodd"/></svg>
+                      Note: you might want to use a <strong>private</strong> custom LoRa model</h2>
+                      <p class="main-message">
+                          To do so, <strong>duplicate the Space</strong> and run it on your own profile using <strong>your own access token</strong> and eventually a GPU (T4-small or A10G-small) for faster inference without waiting in the queue.<br />
+                      </p>
+                      <p class="actions">
+                          <a href="https://huggingface.co/spaces/{os.environ['SPACE_ID']}?duplicate=true">
+                              <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-lg-dark.svg" alt="Duplicate this Space" />
+                          </a>
+                          to start using private models and skip the queue
+                      </p>
+                  </div>
+              ''', elem_id="warning-duplicate")
+          gr.HTML("""
+  <h2 style="text-align: center;">SD-XL Control LoRas</h2>
+  <p style="text-align: center;">Use StableDiffusion XL with <a href="https://huggingface.co/collections/diffusers/sdxl-controlnets-64f9c35846f3f06f5abe351f">Diffusers' SDXL ControlNets</a></p>
+
+          """)
+
+          use_custom_model = gr.Checkbox(label="Use a custom pre-trained LoRa model ? (optional)", value=False, info="To use a private model, you'll need to duplicate the space with your own access token.")
+          
+          with gr.Box(visible=False) as custom_model_box:
+              with gr.Row():
+                  with gr.Column():
+                      if not is_shared_ui:
+                          your_username = api.whoami()["name"]
+                          my_models = api.list_models(author=your_username, filter=["diffusers", "stable-diffusion-xl", 'lora'])
+                          model_names = [item.modelId for item in my_models]
+      
+                      if not is_shared_ui:
+                          custom_model = gr.Dropdown(
+                              label = "Your custom model ID",
+                              info="You can pick one of your private models",
+                              choices = model_names,
+                              allow_custom_value = True
+                              #placeholder = "username/model_id"
+                          )
+                      else:
+                          custom_model = gr.Textbox(
+                              label="Your custom model ID", 
+                              placeholder="your_username/your_trained_model_name", 
+                              info="Make sure your model is set to PUBLIC"
+                          )
+                      
+                      weight_name = gr.Dropdown(
+                          label="Safetensors file", 
+                          #value="pytorch_lora_weights.safetensors", 
+                          info="specify which one if model has several .safetensors files",
+                          allow_custom_value=True,
+                          visible = False
+                      )
+                  with gr.Column():
+                      with gr.Group():
+                          load_model_btn = gr.Button("Load my model", elem_id="load_model_btn")
+                          previous_model = gr.Textbox(
+                              visible = False
+                          )
+                          model_status = gr.Textbox(
+                              label = "model status",
+                              show_label = False,
+                              elem_id = "status_info"
+                          )
+                      trigger_word = gr.Textbox(label="Trigger word", interactive=False, visible=False)
+          
+          image_in = gr.Image(source="upload", type="filepath")
+          
+          with gr.Row():
+              
+              with gr.Column():
+                  with gr.Group():
+                      prompt = gr.Textbox(label="Prompt")
+                      negative_prompt = gr.Textbox(label="Negative prompt", value="extra digit, fewer digits, cropped, worst quality, low quality, glitch, deformed, mutated, ugly, disfigured")
+                  with gr.Group():
+                      guidance_scale = gr.Slider(label="Guidance Scale", minimum=1.0, maximum=10.0, step=0.1, value=7.5)
+                      inf_steps = gr.Slider(label="Inference Steps", minimum="25", maximum="50", step=1, value=25)
+                      custom_lora_weight = gr.Slider(label="Custom model weights", minimum=0.1, maximum=0.9, step=0.1, value=0.9)
+
+              with gr.Column():
+                  with gr.Group():
+                      preprocessor = gr.Dropdown(label="Preprocessor", choices=["canny"], value="canny", interactive=False, info="For the moment, only canny is available")  
+                      controlnet_conditioning_scale = gr.Slider(label="Controlnet conditioning Scale", minimum=0.1, maximum=0.9, step=0.01, value=0.5)
+                  with gr.Group():
+                      seed = gr.Slider(
+                          label="Seed",
+                          info = "-1 denotes a random seed",
+                          minimum=-1,
+                          maximum=423538377342,
+                          step=1,
+                          value=-1
+                      )
+                      last_used_seed = gr.Number(
+                          label = "Last used seed",
+                          info = "the seed used in the last generation",
+                      )
+          
+          
+          submit_btn = gr.Button("Submit")
+          
+          result = gr.Image(label="Result")
+      
+      use_custom_model.change(
+          fn = check_use_custom_or_no,
+          inputs =[use_custom_model],
+          outputs = [custom_model_box],
+          queue = False
+      )
+      custom_model.blur(
+          fn=custom_model_changed,
+          inputs = [custom_model, previous_model],
+          outputs = [model_status],
+          queue = False
+      )
+      load_model_btn.click(
+          fn = load_model,
+          inputs=[custom_model],
+          outputs = [previous_model, model_status, weight_name, trigger_word],
+          queue = False
+      )
+      submit_btn.click(
+          fn = infer,
+          inputs = [use_custom_model, custom_model, weight_name, custom_lora_weight, image_in, prompt, negative_prompt, preprocessor, controlnet_conditioning_scale, guidance_scale, inf_steps, seed],
+          outputs = [result, last_used_seed]
+      )
+
+  return demo 
+  
+
+#demo.queue(max_size=12).launch(share=True)

sdxl/requirements.txt

@@ -0,0 +1,8 @@
+torch==2.0.1
+torchvision==0.15.2
+invisible_watermark
+accelerate
+transformers
+safetensors
+opencv-python
+git+https://github.com/huggingface/diffusers.git

train_dreambooth_lora_sdxl.py

@@ -0,0 +1,1508 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2023 The HuggingFace Inc. 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
+
+import argparse
+import gc
+import hashlib
+import itertools
+import logging
+import math
+import os
+import shutil
+import warnings
+from pathlib import Path
+from typing import Dict
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+import transformers
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import ProjectConfiguration, set_seed
+from huggingface_hub import create_repo, upload_folder
+from packaging import version
+from PIL import Image
+from PIL.ImageOps import exif_transpose
+from torch.utils.data import Dataset
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import AutoTokenizer, PretrainedConfig
+
+import diffusers
+from diffusers import (
+    AutoencoderKL,
+    DDPMScheduler,
+    DPMSolverMultistepScheduler,
+    StableDiffusionXLPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.loaders import LoraLoaderMixin, text_encoder_lora_state_dict
+from diffusers.models.attention_processor import LoRAAttnProcessor, LoRAAttnProcessor2_0
+from diffusers.optimization import get_scheduler
+from diffusers.utils import check_min_version, is_wandb_available
+from diffusers.utils.import_utils import is_xformers_available
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.22.0.dev0")
+
+logger = get_logger(__name__)
+
+def save_tempo_model_card(
+    repo_id: str, dataset_id=str, base_model=str, train_text_encoder=False, prompt=str, repo_folder=None, vae_path=None, last_checkpoint=str
+):
+
+    yaml = f"""
+---
+base_model: {base_model}
+instance_prompt: {prompt}
+tags:
+- stable-diffusion-xl
+- stable-diffusion-xl-diffusers
+- text-to-image
+- diffusers
+- lora
+inference: false
+datasets:
+- {dataset_id}
+---
+    """
+    model_card = f"""
+# LoRA DreamBooth - {repo_id}
+## MODEL IS CURRENTLY TRAINING ...
+Last checkpoint saved: {last_checkpoint}
+These are LoRA adaption weights for {base_model} trained on @fffiloni's SD-XL trainer. 
+The weights were trained on the concept prompt: 
+```
+{prompt}
+```  
+Use this keyword to trigger your custom model in your prompts. 
+LoRA for the text encoder was enabled: {train_text_encoder}.
+Special VAE used for training: {vae_path}.
+## Usage
+Make sure to upgrade diffusers to >= 0.19.0:
+```
+pip install diffusers --upgrade
+```
+In addition make sure to install transformers, safetensors, accelerate as well as the invisible watermark:
+```
+pip install invisible_watermark transformers accelerate safetensors
+```
+To just use the base model, you can run:
+```python
+import torch
+from diffusers import DiffusionPipeline, AutoencoderKL
+device = "cuda" if torch.cuda.is_available() else "cpu"
+vae = AutoencoderKL.from_pretrained('{vae_path}', torch_dtype=torch.float16)
+pipe = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-1.0",
+    vae=vae, torch_dtype=torch.float16, variant="fp16",
+    use_safetensors=True
+)
+pipe.to(device)
+# This is where you load your trained weights
+specific_safetensors = "pytorch_lora_weights.safetensors"
+lora_scale = 0.9
+pipe.load_lora_weights(
+    '{repo_id}', 
+    weight_name = specific_safetensors,
+    # use_auth_token = True 
+)
+prompt = "A majestic {prompt} jumping from a big stone at night"
+image = pipe(
+    prompt=prompt, 
+    num_inference_steps=50,
+    cross_attention_kwargs={{"scale": lora_scale}}
+).images[0]
+```
+"""
+    with open(os.path.join(repo_folder, "README.md"), "w") as f:
+        f.write(yaml + model_card)
+
+def save_model_card(
+    repo_id: str, images=None, dataset_id=str, base_model=str, train_text_encoder=False, prompt=str, repo_folder=None, vae_path=None
+):
+    img_str = ""
+    for i, image in enumerate(images):
+        image.save(os.path.join(repo_folder, f"image_{i}.png"))
+        img_str += f"![img_{i}](./image_{i}.png)\n"
+
+    yaml = f"""
+---
+base_model: {base_model}
+instance_prompt: {prompt}
+tags:
+- stable-diffusion-xl
+- stable-diffusion-xl-diffusers
+- text-to-image
+- diffusers
+- lora
+inference: false
+datasets:
+- {dataset_id}
+---
+    """
+    model_card = f"""
+# LoRA DreamBooth - {repo_id}
+These are LoRA adaption weights for {base_model} trained on @fffiloni's SD-XL trainer. 
+The weights were trained on the concept prompt: 
+```
+{prompt}
+```  
+Use this keyword to trigger your custom model in your prompts. 
+LoRA for the text encoder was enabled: {train_text_encoder}.
+Special VAE used for training: {vae_path}.
+## Usage
+Make sure to upgrade diffusers to >= 0.19.0:
+```
+pip install diffusers --upgrade
+```
+In addition make sure to install transformers, safetensors, accelerate as well as the invisible watermark:
+```
+pip install invisible_watermark transformers accelerate safetensors
+```
+To just use the base model, you can run:
+```python
+import torch
+from diffusers import DiffusionPipeline, AutoencoderKL
+device = "cuda" if torch.cuda.is_available() else "cpu"
+vae = AutoencoderKL.from_pretrained('{vae_path}', torch_dtype=torch.float16)
+pipe = DiffusionPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-1.0",
+    vae=vae, torch_dtype=torch.float16, variant="fp16",
+    use_safetensors=True
+)
+pipe.to(device)
+# This is where you load your trained weights
+specific_safetensors = "pytorch_lora_weights.safetensors"
+lora_scale = 0.9
+pipe.load_lora_weights(
+    '{repo_id}', 
+    weight_name = specific_safetensors,
+    # use_auth_token = True 
+)
+prompt = "A majestic {prompt} jumping from a big stone at night"
+image = pipe(
+    prompt=prompt, 
+    num_inference_steps=50,
+    cross_attention_kwargs={{"scale": lora_scale}}
+).images[0]
+```
+"""
+    with open(os.path.join(repo_folder, "README.md"), "w") as f:
+        f.write(yaml + model_card)
+
+
+def import_model_class_from_model_name_or_path(
+    pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
+):
+    text_encoder_config = PretrainedConfig.from_pretrained(
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
+    )
+    model_class = text_encoder_config.architectures[0]
+
+    if model_class == "CLIPTextModel":
+        from transformers import CLIPTextModel
+
+        return CLIPTextModel
+    elif model_class == "CLIPTextModelWithProjection":
+        from transformers import CLIPTextModelWithProjection
+
+        return CLIPTextModelWithProjection
+    else:
+        raise ValueError(f"{model_class} is not supported.")
+
+
+def parse_args(input_args=None):
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        required=True,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--pretrained_vae_model_name_or_path",
+        type=str,
+        default=None,
+        help="Path to pretrained VAE model with better numerical stability. More details: https://github.com/huggingface/diffusers/pull/4038.",
+    )
+    parser.add_argument(
+        "--revision",
+        type=str,
+        default=None,
+        required=False,
+        help="Revision of pretrained model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--dataset_id",
+        type=str,
+        default=None,
+        required=True,
+        help="The dataset ID you want to train images from",
+    )
+    parser.add_argument(
+        "--instance_data_dir",
+        type=str,
+        default=None,
+        required=True,
+        help="A folder containing the training data of instance images.",
+    )
+    parser.add_argument(
+        "--class_data_dir",
+        type=str,
+        default=None,
+        required=False,
+        help="A folder containing the training data of class images.",
+    )
+    parser.add_argument(
+        "--instance_prompt",
+        type=str,
+        default=None,
+        required=True,
+        help="The prompt with identifier specifying the instance",
+    )
+    parser.add_argument(
+        "--class_prompt",
+        type=str,
+        default=None,
+        help="The prompt to specify images in the same class as provided instance images.",
+    )
+    parser.add_argument(
+        "--validation_prompt",
+        type=str,
+        default=None,
+        help="A prompt that is used during validation to verify that the model is learning.",
+    )
+    parser.add_argument(
+        "--num_validation_images",
+        type=int,
+        default=4,
+        help="Number of images that should be generated during validation with `validation_prompt`.",
+    )
+    parser.add_argument(
+        "--validation_epochs",
+        type=int,
+        default=50,
+        help=(
+            "Run dreambooth validation every X epochs. Dreambooth validation consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`."
+        ),
+    )
+    parser.add_argument(
+        "--with_prior_preservation",
+        default=False,
+        action="store_true",
+        help="Flag to add prior preservation loss.",
+    )
+    parser.add_argument("--prior_loss_weight", type=float, default=1.0, help="The weight of prior preservation loss.")
+    parser.add_argument(
+        "--num_class_images",
+        type=int,
+        default=100,
+        help=(
+            "Minimal class images for prior preservation loss. If there are not enough images already present in"
+            " class_data_dir, additional images will be sampled with class_prompt."
+        ),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="lora-dreambooth-model",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
+    parser.add_argument(
+        "--resolution",
+        type=int,
+        default=1024,
+        help=(
+            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
+            " resolution"
+        ),
+    )
+    parser.add_argument(
+        "--crops_coords_top_left_h",
+        type=int,
+        default=0,
+        help=("Coordinate for (the height) to be included in the crop coordinate embeddings needed by SDXL UNet."),
+    )
+    parser.add_argument(
+        "--crops_coords_top_left_w",
+        type=int,
+        default=0,
+        help=("Coordinate for (the height) to be included in the crop coordinate embeddings needed by SDXL UNet."),
+    )
+    parser.add_argument(
+        "--center_crop",
+        default=False,
+        action="store_true",
+        help=(
+            "Whether to center crop the input images to the resolution. If not set, the images will be randomly"
+            " cropped. The images will be resized to the resolution first before cropping."
+        ),
+    )
+    parser.add_argument(
+        "--train_text_encoder",
+        action="store_true",
+        help="Whether to train the text encoder. If set, the text encoder should be float32 precision.",
+    )
+    parser.add_argument(
+        "--train_batch_size", type=int, default=4, help="Batch size (per device) for the training dataloader."
+    )
+    parser.add_argument(
+        "--sample_batch_size", type=int, default=4, help="Batch size (per device) for sampling images."
+    )
+    parser.add_argument("--num_train_epochs", type=int, default=1)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints can be used both as final"
+            " checkpoints in case they are better than the last checkpoint, and are also suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--checkpoints_total_limit",
+        type=int,
+        default=None,
+        help=("Max number of checkpoints to store."),
+    )
+    parser.add_argument(
+        "--resume_from_checkpoint",
+        type=str,
+        default=None,
+        help=(
+            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
+            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
+        ),
+    )
+    parser.add_argument(
+        "--gradient_accumulation_steps",
+        type=int,
+        default=1,
+        help="Number of updates steps to accumulate before performing a backward/update pass.",
+    )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action="store_true",
+        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
+    )
+    parser.add_argument(
+        "--learning_rate",
+        type=float,
+        default=5e-4,
+        help="Initial learning rate (after the potential warmup period) to use.",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        action="store_true",
+        default=False,
+        help="Scale the learning rate by the number of GPUs, gradient accumulation steps, and batch size.",
+    )
+    parser.add_argument(
+        "--lr_scheduler",
+        type=str,
+        default="constant",
+        help=(
+            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
+            ' "constant", "constant_with_warmup"]'
+        ),
+    )
+    parser.add_argument(
+        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
+    )
+    parser.add_argument(
+        "--lr_num_cycles",
+        type=int,
+        default=1,
+        help="Number of hard resets of the lr in cosine_with_restarts scheduler.",
+    )
+    parser.add_argument("--lr_power", type=float, default=1.0, help="Power factor of the polynomial scheduler.")
+    parser.add_argument(
+        "--dataloader_num_workers",
+        type=int,
+        default=0,
+        help=(
+            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
+        ),
+    )
+    parser.add_argument(
+        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
+    )
+    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_weight_decay", type=float, default=1e-2, help="Weight decay to use.")
+    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
+    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--hub_model_id",
+        type=str,
+        default=None,
+        help="The name of the repository to keep in sync with the local `output_dir`.",
+    )
+    parser.add_argument(
+        "--logging_dir",
+        type=str,
+        default="logs",
+        help=(
+            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
+            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
+        ),
+    )
+    parser.add_argument(
+        "--allow_tf32",
+        action="store_true",
+        help=(
+            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
+            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
+        ),
+    )
+    parser.add_argument(
+        "--report_to",
+        type=str,
+        default="tensorboard",
+        help=(
+            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
+            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default=None,
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
+        ),
+    )
+    parser.add_argument(
+        "--prior_generation_precision",
+        type=str,
+        default=None,
+        choices=["no", "fp32", "fp16", "bf16"],
+        help=(
+            "Choose prior generation precision between fp32, fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to  fp16 if a GPU is available else fp32."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+    parser.add_argument(
+        "--rank",
+        type=int,
+        default=4,
+        help=("The dimension of the LoRA update matrices."),
+    )
+
+    if input_args is not None:
+        args = parser.parse_args(input_args)
+    else:
+        args = parser.parse_args()
+
+    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
+    if env_local_rank != -1 and env_local_rank != args.local_rank:
+        args.local_rank = env_local_rank
+
+    if args.with_prior_preservation:
+        if args.class_data_dir is None:
+            raise ValueError("You must specify a data directory for class images.")
+        if args.class_prompt is None:
+            raise ValueError("You must specify prompt for class images.")
+    else:
+        # logger is not available yet
+        if args.class_data_dir is not None:
+            warnings.warn("You need not use --class_data_dir without --with_prior_preservation.")
+        if args.class_prompt is not None:
+            warnings.warn("You need not use --class_prompt without --with_prior_preservation.")
+
+    return args
+
+
+class DreamBoothDataset(Dataset):
+    """
+    A dataset to prepare the instance and class images with the prompts for fine-tuning the model.
+    It pre-processes the images.
+    """
+
+    def __init__(
+        self,
+        instance_data_root,
+        class_data_root=None,
+        class_num=None,
+        size=1024,
+        center_crop=False,
+    ):
+        self.size = size
+        self.center_crop = center_crop
+
+        self.instance_data_root = Path(instance_data_root)
+        if not self.instance_data_root.exists():
+            raise ValueError("Instance images root doesn't exists.")
+
+        self.instance_images_path = list(Path(instance_data_root).iterdir())
+        self.num_instance_images = len(self.instance_images_path)
+        self._length = self.num_instance_images
+
+        if class_data_root is not None:
+            self.class_data_root = Path(class_data_root)
+            self.class_data_root.mkdir(parents=True, exist_ok=True)
+            self.class_images_path = list(self.class_data_root.iterdir())
+            if class_num is not None:
+                self.num_class_images = min(len(self.class_images_path), class_num)
+            else:
+                self.num_class_images = len(self.class_images_path)
+            self._length = max(self.num_class_images, self.num_instance_images)
+        else:
+            self.class_data_root = None
+
+        self.image_transforms = transforms.Compose(
+            [
+                transforms.Resize(size, interpolation=transforms.InterpolationMode.BILINEAR),
+                transforms.CenterCrop(size) if center_crop else transforms.RandomCrop(size),
+                transforms.ToTensor(),
+                transforms.Normalize([0.5], [0.5]),
+            ]
+        )
+
+    def __len__(self):
+        return self._length
+
+    def __getitem__(self, index):
+        example = {}
+        instance_image = Image.open(self.instance_images_path[index % self.num_instance_images])
+        instance_image = exif_transpose(instance_image)
+
+        if not instance_image.mode == "RGB":
+            instance_image = instance_image.convert("RGB")
+        example["instance_images"] = self.image_transforms(instance_image)
+
+        if self.class_data_root:
+            class_image = Image.open(self.class_images_path[index % self.num_class_images])
+            class_image = exif_transpose(class_image)
+
+            if not class_image.mode == "RGB":
+                class_image = class_image.convert("RGB")
+            example["class_images"] = self.image_transforms(class_image)
+
+        return example
+
+
+def collate_fn(examples, with_prior_preservation=False):
+    pixel_values = [example["instance_images"] for example in examples]
+
+    # Concat class and instance examples for prior preservation.
+    # We do this to avoid doing two forward passes.
+    if with_prior_preservation:
+        pixel_values += [example["class_images"] for example in examples]
+
+    pixel_values = torch.stack(pixel_values)
+    pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
+
+    batch = {"pixel_values": pixel_values}
+    return batch
+
+
+class PromptDataset(Dataset):
+    "A simple dataset to prepare the prompts to generate class images on multiple GPUs."
+
+    def __init__(self, prompt, num_samples):
+        self.prompt = prompt
+        self.num_samples = num_samples
+
+    def __len__(self):
+        return self.num_samples
+
+    def __getitem__(self, index):
+        example = {}
+        example["prompt"] = self.prompt
+        example["index"] = index
+        return example
+
+
+def tokenize_prompt(tokenizer, prompt):
+    text_inputs = tokenizer(
+        prompt,
+        padding="max_length",
+        max_length=tokenizer.model_max_length,
+        truncation=True,
+        return_tensors="pt",
+    )
+    text_input_ids = text_inputs.input_ids
+    return text_input_ids
+
+
+# Adapted from pipelines.StableDiffusionXLPipeline.encode_prompt
+def encode_prompt(text_encoders, tokenizers, prompt, text_input_ids_list=None):
+    prompt_embeds_list = []
+
+    for i, text_encoder in enumerate(text_encoders):
+        if tokenizers is not None:
+            tokenizer = tokenizers[i]
+            text_input_ids = tokenize_prompt(tokenizer, prompt)
+        else:
+            assert text_input_ids_list is not None
+            text_input_ids = text_input_ids_list[i]
+
+        prompt_embeds = text_encoder(
+            text_input_ids.to(text_encoder.device),
+            output_hidden_states=True,
+        )
+
+        # We are only ALWAYS interested in the pooled output of the final text encoder
+        pooled_prompt_embeds = prompt_embeds[0]
+        prompt_embeds = prompt_embeds.hidden_states[-2]
+        bs_embed, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.view(bs_embed, seq_len, -1)
+        prompt_embeds_list.append(prompt_embeds)
+
+    prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+    pooled_prompt_embeds = pooled_prompt_embeds.view(bs_embed, -1)
+    return prompt_embeds, pooled_prompt_embeds
+
+
+def unet_attn_processors_state_dict(unet) -> Dict[str, torch.tensor]:
+    """
+    Returns:
+        a state dict containing just the attention processor parameters.
+    """
+    attn_processors = unet.attn_processors
+
+    attn_processors_state_dict = {}
+
+    for attn_processor_key, attn_processor in attn_processors.items():
+        for parameter_key, parameter in attn_processor.state_dict().items():
+            attn_processors_state_dict[f"{attn_processor_key}.{parameter_key}"] = parameter
+
+    return attn_processors_state_dict
+
+
+def main(args):
+    logging_dir = Path(args.output_dir, args.logging_dir)
+
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
+
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.report_to,
+        project_config=accelerator_project_config,
+    )
+
+    if args.report_to == "wandb":
+        if not is_wandb_available():
+            raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
+        import wandb
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    logger.info(accelerator.state, main_process_only=False)
+    if accelerator.is_local_main_process:
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        transformers.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # Generate class images if prior preservation is enabled.
+    if args.with_prior_preservation:
+        class_images_dir = Path(args.class_data_dir)
+        if not class_images_dir.exists():
+            class_images_dir.mkdir(parents=True)
+        cur_class_images = len(list(class_images_dir.iterdir()))
+
+        if cur_class_images < args.num_class_images:
+            torch_dtype = torch.float16 if accelerator.device.type == "cuda" else torch.float32
+            if args.prior_generation_precision == "fp32":
+                torch_dtype = torch.float32
+            elif args.prior_generation_precision == "fp16":
+                torch_dtype = torch.float16
+            elif args.prior_generation_precision == "bf16":
+                torch_dtype = torch.bfloat16
+            pipeline = StableDiffusionXLPipeline.from_pretrained(
+                args.pretrained_model_name_or_path,
+                torch_dtype=torch_dtype,
+                revision=args.revision,
+            )
+            pipeline.set_progress_bar_config(disable=True)
+
+            num_new_images = args.num_class_images - cur_class_images
+            logger.info(f"Number of class images to sample: {num_new_images}.")
+
+            sample_dataset = PromptDataset(args.class_prompt, num_new_images)
+            sample_dataloader = torch.utils.data.DataLoader(sample_dataset, batch_size=args.sample_batch_size)
+
+            sample_dataloader = accelerator.prepare(sample_dataloader)
+            pipeline.to(accelerator.device)
+
+            for example in tqdm(
+                sample_dataloader, desc="Generating class images", disable=not accelerator.is_local_main_process
+            ):
+                images = pipeline(example["prompt"]).images
+
+                for i, image in enumerate(images):
+                    hash_image = hashlib.sha1(image.tobytes()).hexdigest()
+                    image_filename = class_images_dir / f"{example['index'][i] + cur_class_images}-{hash_image}.jpg"
+                    image.save(image_filename)
+
+            del pipeline
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+
+    # Handle the repository creation
+    if accelerator.is_main_process:
+        if args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+        if args.push_to_hub:
+            repo_id = create_repo(
+                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
+            ).repo_id
+
+    # Load the tokenizers
+    tokenizer_one = AutoTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="tokenizer", revision=args.revision, use_fast=False
+    )
+    tokenizer_two = AutoTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="tokenizer_2", revision=args.revision, use_fast=False
+    )
+
+    # import correct text encoder classes
+    text_encoder_cls_one = import_model_class_from_model_name_or_path(
+        args.pretrained_model_name_or_path, args.revision
+    )
+    text_encoder_cls_two = import_model_class_from_model_name_or_path(
+        args.pretrained_model_name_or_path, args.revision, subfolder="text_encoder_2"
+    )
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    text_encoder_one = text_encoder_cls_one.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision
+    )
+    text_encoder_two = text_encoder_cls_two.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder_2", revision=args.revision
+    )
+    vae_path = (
+        args.pretrained_model_name_or_path
+        if args.pretrained_vae_model_name_or_path is None
+        else args.pretrained_vae_model_name_or_path
+    )
+    vae = AutoencoderKL.from_pretrained(
+        vae_path, subfolder="vae" if args.pretrained_vae_model_name_or_path is None else None, revision=args.revision
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision
+    )
+
+    # We only train the additional adapter LoRA layers
+    vae.requires_grad_(False)
+    text_encoder_one.requires_grad_(False)
+    text_encoder_two.requires_grad_(False)
+    unet.requires_grad_(False)
+
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights are only used for inference, keeping weights in full precision is not required.
+    weight_dtype = torch.float32
+    if accelerator.mixed_precision == "fp16":
+        weight_dtype = torch.float16
+    elif accelerator.mixed_precision == "bf16":
+        weight_dtype = torch.bfloat16
+
+    # Move unet, vae and text_encoder to device and cast to weight_dtype
+    unet.to(accelerator.device, dtype=weight_dtype)
+
+    # The VAE is always in float32 to avoid NaN losses.
+    vae.to(accelerator.device, dtype=torch.float32)
+
+    text_encoder_one.to(accelerator.device, dtype=weight_dtype)
+    text_encoder_two.to(accelerator.device, dtype=weight_dtype)
+
+    if args.enable_xformers_memory_efficient_attention:
+        if is_xformers_available():
+            import xformers
+
+            xformers_version = version.parse(xformers.__version__)
+            if xformers_version == version.parse("0.0.16"):
+                logger.warn(
+                    "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
+                )
+            unet.enable_xformers_memory_efficient_attention()
+        else:
+            raise ValueError("xformers is not available. Make sure it is installed correctly")
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        if args.train_text_encoder:
+            text_encoder_one.gradient_checkpointing_enable()
+            text_encoder_two.gradient_checkpointing_enable()
+
+    # now we will add new LoRA weights to the attention layers
+    # Set correct lora layers
+    unet_lora_attn_procs = {}
+    unet_lora_parameters = []
+    for name, attn_processor in unet.attn_processors.items():
+        cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+        if name.startswith("mid_block"):
+            hidden_size = unet.config.block_out_channels[-1]
+        elif name.startswith("up_blocks"):
+            block_id = int(name[len("up_blocks.")])
+            hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+        elif name.startswith("down_blocks"):
+            block_id = int(name[len("down_blocks.")])
+            hidden_size = unet.config.block_out_channels[block_id]
+
+        lora_attn_processor_class = (
+            LoRAAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else LoRAAttnProcessor
+        )
+        module = lora_attn_processor_class(
+            hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, rank=args.rank
+        )
+        unet_lora_attn_procs[name] = module
+        unet_lora_parameters.extend(module.parameters())
+
+    unet.set_attn_processor(unet_lora_attn_procs)
+
+    # The text encoder comes from 🤗 transformers, so we cannot directly modify it.
+    # So, instead, we monkey-patch the forward calls of its attention-blocks.
+    if args.train_text_encoder:
+        # ensure that dtype is float32, even if rest of the model that isn't trained is loaded in fp16
+        text_lora_parameters_one = LoraLoaderMixin._modify_text_encoder(
+            text_encoder_one, dtype=torch.float32, rank=args.rank
+        )
+        text_lora_parameters_two = LoraLoaderMixin._modify_text_encoder(
+            text_encoder_two, dtype=torch.float32, rank=args.rank
+        )
+
+    # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
+    def save_model_hook(models, weights, output_dir):
+        if accelerator.is_main_process:
+            # there are only two options here. Either are just the unet attn processor layers
+            # or there are the unet and text encoder atten layers
+            unet_lora_layers_to_save = None
+            text_encoder_one_lora_layers_to_save = None
+            text_encoder_two_lora_layers_to_save = None
+
+            for model in models:
+                if isinstance(model, type(accelerator.unwrap_model(unet))):
+                    unet_lora_layers_to_save = unet_attn_processors_state_dict(model)
+                elif isinstance(model, type(accelerator.unwrap_model(text_encoder_one))):
+                    text_encoder_one_lora_layers_to_save = text_encoder_lora_state_dict(model)
+                elif isinstance(model, type(accelerator.unwrap_model(text_encoder_two))):
+                    text_encoder_two_lora_layers_to_save = text_encoder_lora_state_dict(model)
+                else:
+                    raise ValueError(f"unexpected save model: {model.__class__}")
+
+                # make sure to pop weight so that corresponding model is not saved again
+                weights.pop()
+
+            StableDiffusionXLPipeline.save_lora_weights(
+                output_dir,
+                unet_lora_layers=unet_lora_layers_to_save,
+                text_encoder_lora_layers=text_encoder_one_lora_layers_to_save,
+                text_encoder_2_lora_layers=text_encoder_two_lora_layers_to_save,
+            )
+
+    def load_model_hook(models, input_dir):
+        unet_ = None
+        text_encoder_one_ = None
+        text_encoder_two_ = None
+
+        while len(models) > 0:
+            model = models.pop()
+
+            if isinstance(model, type(accelerator.unwrap_model(unet))):
+                unet_ = model
+            elif isinstance(model, type(accelerator.unwrap_model(text_encoder_one))):
+                text_encoder_one_ = model
+            elif isinstance(model, type(accelerator.unwrap_model(text_encoder_two))):
+                text_encoder_two_ = model
+            else:
+                raise ValueError(f"unexpected save model: {model.__class__}")
+
+        lora_state_dict, network_alphas = LoraLoaderMixin.lora_state_dict(input_dir)
+        LoraLoaderMixin.load_lora_into_unet(lora_state_dict, network_alphas=network_alphas, unet=unet_)
+
+        text_encoder_state_dict = {k: v for k, v in lora_state_dict.items() if "text_encoder." in k}
+        LoraLoaderMixin.load_lora_into_text_encoder(
+            text_encoder_state_dict, network_alphas=network_alphas, text_encoder=text_encoder_one_
+        )
+
+        text_encoder_2_state_dict = {k: v for k, v in lora_state_dict.items() if "text_encoder_2." in k}
+        LoraLoaderMixin.load_lora_into_text_encoder(
+            text_encoder_2_state_dict, network_alphas=network_alphas, text_encoder=text_encoder_two_
+        )
+
+    accelerator.register_save_state_pre_hook(save_model_hook)
+    accelerator.register_load_state_pre_hook(load_model_hook)
+
+    # Enable TF32 for faster training on Ampere GPUs,
+    # cf https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices
+    if args.allow_tf32:
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    if args.scale_lr:
+        args.learning_rate = (
+            args.learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
+        )
+
+    # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
+    if args.use_8bit_adam:
+        try:
+            import bitsandbytes as bnb
+        except ImportError:
+            raise ImportError(
+                "To use 8-bit Adam, please install the bitsandbytes library: `pip install bitsandbytes`."
+            )
+
+        optimizer_class = bnb.optim.AdamW8bit
+    else:
+        optimizer_class = torch.optim.AdamW
+
+    # Optimizer creation
+    params_to_optimize = (
+        itertools.chain(unet_lora_parameters, text_lora_parameters_one, text_lora_parameters_two)
+        if args.train_text_encoder
+        else unet_lora_parameters
+    )
+    optimizer = optimizer_class(
+        params_to_optimize,
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    # Computes additional embeddings/ids required by the SDXL UNet.
+    # regular text emebddings (when `train_text_encoder` is not True)
+    # pooled text embeddings
+    # time ids
+
+    def compute_time_ids():
+        # Adapted from pipeline.StableDiffusionXLPipeline._get_add_time_ids
+        original_size = (args.resolution, args.resolution)
+        target_size = (args.resolution, args.resolution)
+        crops_coords_top_left = (args.crops_coords_top_left_h, args.crops_coords_top_left_w)
+        add_time_ids = list(original_size + crops_coords_top_left + target_size)
+        add_time_ids = torch.tensor([add_time_ids])
+        add_time_ids = add_time_ids.to(accelerator.device, dtype=weight_dtype)
+        return add_time_ids
+
+    if not args.train_text_encoder:
+        tokenizers = [tokenizer_one, tokenizer_two]
+        text_encoders = [text_encoder_one, text_encoder_two]
+
+        def compute_text_embeddings(prompt, text_encoders, tokenizers):
+            with torch.no_grad():
+                prompt_embeds, pooled_prompt_embeds = encode_prompt(text_encoders, tokenizers, prompt)
+                prompt_embeds = prompt_embeds.to(accelerator.device)
+                pooled_prompt_embeds = pooled_prompt_embeds.to(accelerator.device)
+            return prompt_embeds, pooled_prompt_embeds
+
+    # Handle instance prompt.
+    instance_time_ids = compute_time_ids()
+    if not args.train_text_encoder:
+        instance_prompt_hidden_states, instance_pooled_prompt_embeds = compute_text_embeddings(
+            args.instance_prompt, text_encoders, tokenizers
+        )
+
+    # Handle class prompt for prior-preservation.
+    if args.with_prior_preservation:
+        class_time_ids = compute_time_ids()
+        if not args.train_text_encoder:
+            class_prompt_hidden_states, class_pooled_prompt_embeds = compute_text_embeddings(
+                args.class_prompt, text_encoders, tokenizers
+            )
+
+    # Clear the memory here.
+    if not args.train_text_encoder:
+        del tokenizers, text_encoders
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    # Pack the statically computed variables appropriately. This is so that we don't
+    # have to pass them to the dataloader.
+    add_time_ids = instance_time_ids
+    if args.with_prior_preservation:
+        add_time_ids = torch.cat([add_time_ids, class_time_ids], dim=0)
+
+    if not args.train_text_encoder:
+        prompt_embeds = instance_prompt_hidden_states
+        unet_add_text_embeds = instance_pooled_prompt_embeds
+        if args.with_prior_preservation:
+            prompt_embeds = torch.cat([prompt_embeds, class_prompt_hidden_states], dim=0)
+            unet_add_text_embeds = torch.cat([unet_add_text_embeds, class_pooled_prompt_embeds], dim=0)
+    else:
+        tokens_one = tokenize_prompt(tokenizer_one, args.instance_prompt)
+        tokens_two = tokenize_prompt(tokenizer_two, args.instance_prompt)
+        if args.with_prior_preservation:
+            class_tokens_one = tokenize_prompt(tokenizer_one, args.class_prompt)
+            class_tokens_two = tokenize_prompt(tokenizer_two, args.class_prompt)
+            tokens_one = torch.cat([tokens_one, class_tokens_one], dim=0)
+            tokens_two = torch.cat([tokens_two, class_tokens_two], dim=0)
+
+    # Dataset and DataLoaders creation:
+    train_dataset = DreamBoothDataset(
+        instance_data_root=args.instance_data_dir,
+        class_data_root=args.class_data_dir if args.with_prior_preservation else None,
+        class_num=args.num_class_images,
+        size=args.resolution,
+        center_crop=args.center_crop,
+    )
+
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset,
+        batch_size=args.train_batch_size,
+        shuffle=True,
+        collate_fn=lambda examples: collate_fn(examples, args.with_prior_preservation),
+        num_workers=args.dataloader_num_workers,
+    )
+
+    # Scheduler and math around the number of training steps.
+    overrode_max_train_steps = False
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    if args.max_train_steps is None:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        overrode_max_train_steps = True
+
+    lr_scheduler = get_scheduler(
+        args.lr_scheduler,
+        optimizer=optimizer,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
+        num_cycles=args.lr_num_cycles,
+        power=args.lr_power,
+    )
+
+    # Prepare everything with our `accelerator`.
+    if args.train_text_encoder:
+        unet, text_encoder_one, text_encoder_two, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, text_encoder_one, text_encoder_two, optimizer, train_dataloader, lr_scheduler
+        )
+    else:
+        unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, optimizer, train_dataloader, lr_scheduler
+        )
+
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    if overrode_max_train_steps:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+    # Afterwards we recalculate our number of training epochs
+    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+
+    # We need to initialize the trackers we use, and also store our configuration.
+    # The trackers initializes automatically on the main process.
+    if accelerator.is_main_process:
+        accelerator.init_trackers("dreambooth-lora-sd-xl", config=vars(args))
+
+    # Train!
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+
+    logger.info("***** Running training *****")
+    logger.info(f"  Num examples = {len(train_dataset)}")
+    logger.info(f"  Num batches each epoch = {len(train_dataloader)}")
+    logger.info(f"  Num Epochs = {args.num_train_epochs}")
+    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
+    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
+    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
+    logger.info(f"  Total optimization steps = {args.max_train_steps}")
+    global_step = 0
+    first_epoch = 0
+
+    # Potentially load in the weights and states from a previous save
+    if args.resume_from_checkpoint:
+        if args.resume_from_checkpoint != "latest":
+            path = os.path.basename(args.resume_from_checkpoint)
+        else:
+            # Get the mos recent checkpoint
+            dirs = os.listdir(args.output_dir)
+            dirs = [d for d in dirs if d.startswith("checkpoint")]
+            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
+            path = dirs[-1] if len(dirs) > 0 else None
+
+        if path is None:
+            accelerator.print(
+                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
+            )
+            args.resume_from_checkpoint = None
+            initial_global_step = 0
+        else:
+            accelerator.print(f"Resuming from checkpoint {path}")
+            accelerator.load_state(os.path.join(args.output_dir, path))
+            global_step = int(path.split("-")[1])
+
+            initial_global_step = global_step
+            first_epoch = global_step // num_update_steps_per_epoch
+
+    else:
+        initial_global_step = 0
+
+    progress_bar = tqdm(
+        range(0, args.max_train_steps),
+        initial=initial_global_step,
+        desc="Steps",
+        # Only show the progress bar once on each machine.
+        disable=not accelerator.is_local_main_process,
+    )
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        # Print a message for each epoch
+        print(f"Epoch {epoch}: Training in progress...")
+        unet.train()
+        if args.train_text_encoder:
+            text_encoder_one.train()
+            text_encoder_two.train()
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet):
+                pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
+
+                # Convert images to latent space
+                model_input = vae.encode(pixel_values).latent_dist.sample()
+                model_input = model_input * vae.config.scaling_factor
+                if args.pretrained_vae_model_name_or_path is None:
+                    model_input = model_input.to(weight_dtype)
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(model_input)
+                bsz = model_input.shape[0]
+                # Sample a random timestep for each image
+                timesteps = torch.randint(
+                    0, noise_scheduler.config.num_train_timesteps, (bsz,), device=model_input.device
+                )
+                timesteps = timesteps.long()
+
+                # Add noise to the model input according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_model_input = noise_scheduler.add_noise(model_input, noise, timesteps)
+
+                # Calculate the elements to repeat depending on the use of prior-preservation.
+                elems_to_repeat = bsz // 2 if args.with_prior_preservation else bsz
+
+                # Predict the noise residual
+                if not args.train_text_encoder:
+                    unet_added_conditions = {
+                        "time_ids": add_time_ids.repeat(elems_to_repeat, 1),
+                        "text_embeds": unet_add_text_embeds.repeat(elems_to_repeat, 1),
+                    }
+                    prompt_embeds_input = prompt_embeds.repeat(elems_to_repeat, 1, 1)
+                    model_pred = unet(
+                        noisy_model_input,
+                        timesteps,
+                        prompt_embeds_input,
+                        added_cond_kwargs=unet_added_conditions,
+                    ).sample
+                else:
+                    unet_added_conditions = {"time_ids": add_time_ids.repeat(elems_to_repeat, 1)}
+                    prompt_embeds, pooled_prompt_embeds = encode_prompt(
+                        text_encoders=[text_encoder_one, text_encoder_two],
+                        tokenizers=None,
+                        prompt=None,
+                        text_input_ids_list=[tokens_one, tokens_two],
+                    )
+                    unet_added_conditions.update({"text_embeds": pooled_prompt_embeds.repeat(elems_to_repeat, 1)})
+                    prompt_embeds_input = prompt_embeds.repeat(elems_to_repeat, 1, 1)
+                    model_pred = unet(
+                        noisy_model_input, timesteps, prompt_embeds_input, added_cond_kwargs=unet_added_conditions
+                    ).sample
+
+                # Get the target for loss depending on the prediction type
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(model_input, noise, timesteps)
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+                if args.with_prior_preservation:
+                    # Chunk the noise and model_pred into two parts and compute the loss on each part separately.
+                    model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
+                    target, target_prior = torch.chunk(target, 2, dim=0)
+
+                    # Compute instance loss
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                    # Compute prior loss
+                    prior_loss = F.mse_loss(model_pred_prior.float(), target_prior.float(), reduction="mean")
+
+                    # Add the prior loss to the instance loss.
+                    loss = loss + args.prior_loss_weight * prior_loss
+                else:
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    params_to_clip = (
+                        itertools.chain(unet_lora_parameters, text_lora_parameters_one, text_lora_parameters_two)
+                        if args.train_text_encoder
+                        else unet_lora_parameters
+                    )
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                # Print a message for each step
+                print(f"Step {global_step}/{args.max_train_steps}: Done") 
+                progress_bar.update(1)
+                global_step += 1
+
+                if accelerator.is_main_process:
+                    if global_step % args.checkpointing_steps == 0:
+                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
+                        if args.checkpoints_total_limit is not None:
+                            checkpoints = os.listdir(args.output_dir)
+                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
+                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
+
+                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
+                            if len(checkpoints) >= args.checkpoints_total_limit:
+                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
+                                removing_checkpoints = checkpoints[0:num_to_remove]
+
+                                logger.info(
+                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
+                                )
+                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
+
+                                for removing_checkpoint in removing_checkpoints:
+                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
+                                    shutil.rmtree(removing_checkpoint)
+
+                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
+                        accelerator.save_state(save_path)
+                        logger.info(f"Saved state to {save_path}")
+
+                        save_tempo_model_card(
+                            repo_id,
+                            dataset_id=args.dataset_id,
+                            base_model=args.pretrained_model_name_or_path,
+                            train_text_encoder=args.train_text_encoder,
+                            prompt=args.instance_prompt,
+                            repo_folder=args.output_dir,
+                            vae_path=args.pretrained_vae_model_name_or_path,
+                            last_checkpoint = f"checkpoint-{global_step}"
+                        )
+                        
+                        upload_folder(
+                            repo_id=repo_id,
+                            folder_path=args.output_dir,
+                            commit_message=f"saving checkpoint-{global_step}",
+                            ignore_patterns=["step_*", "epoch_*"],
+                            token=args.hub_token
+                        )
+
+            logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+            accelerator.log(logs, step=global_step)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if accelerator.is_main_process:
+            if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
+                logger.info(
+                    f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+                    f" {args.validation_prompt}."
+                )
+                # create pipeline
+                if not args.train_text_encoder:
+                    text_encoder_one = text_encoder_cls_one.from_pretrained(
+                        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision
+                    )
+                    text_encoder_two = text_encoder_cls_two.from_pretrained(
+                        args.pretrained_model_name_or_path, subfolder="text_encoder_2", revision=args.revision
+                    )
+                pipeline = StableDiffusionXLPipeline.from_pretrained(
+                    args.pretrained_model_name_or_path,
+                    vae=vae,
+                    text_encoder=accelerator.unwrap_model(text_encoder_one),
+                    text_encoder_2=accelerator.unwrap_model(text_encoder_two),
+                    unet=accelerator.unwrap_model(unet),
+                    revision=args.revision,
+                    torch_dtype=weight_dtype,
+                )
+
+                # We train on the simplified learning objective. If we were previously predicting a variance, we need the scheduler to ignore it
+                scheduler_args = {}
+
+                if "variance_type" in pipeline.scheduler.config:
+                    variance_type = pipeline.scheduler.config.variance_type
+
+                    if variance_type in ["learned", "learned_range"]:
+                        variance_type = "fixed_small"
+
+                    scheduler_args["variance_type"] = variance_type
+
+                pipeline.scheduler = DPMSolverMultistepScheduler.from_config(
+                    pipeline.scheduler.config, **scheduler_args
+                )
+
+                pipeline = pipeline.to(accelerator.device)
+                pipeline.set_progress_bar_config(disable=True)
+
+                # run inference
+                generator = torch.Generator(device=accelerator.device).manual_seed(args.seed) if args.seed else None
+                pipeline_args = {"prompt": args.validation_prompt}
+
+                with torch.cuda.amp.autocast():
+                    images = [
+                        pipeline(**pipeline_args, generator=generator).images[0]
+                        for _ in range(args.num_validation_images)
+                    ]
+
+                for tracker in accelerator.trackers:
+                    if tracker.name == "tensorboard":
+                        np_images = np.stack([np.asarray(img) for img in images])
+                        tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
+                    if tracker.name == "wandb":
+                        tracker.log(
+                            {
+                                "validation": [
+                                    wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
+                                    for i, image in enumerate(images)
+                                ]
+                            }
+                        )
+
+                del pipeline
+                torch.cuda.empty_cache()
+
+    # Save the lora layers
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        unet = accelerator.unwrap_model(unet)
+        unet = unet.to(torch.float32)
+        unet_lora_layers = unet_attn_processors_state_dict(unet)
+
+        if args.train_text_encoder:
+            text_encoder_one = accelerator.unwrap_model(text_encoder_one)
+            text_encoder_lora_layers = text_encoder_lora_state_dict(text_encoder_one.to(torch.float32))
+            text_encoder_two = accelerator.unwrap_model(text_encoder_two)
+            text_encoder_2_lora_layers = text_encoder_lora_state_dict(text_encoder_two.to(torch.float32))
+        else:
+            text_encoder_lora_layers = None
+            text_encoder_2_lora_layers = None
+
+        StableDiffusionXLPipeline.save_lora_weights(
+            save_directory=args.output_dir,
+            unet_lora_layers=unet_lora_layers,
+            text_encoder_lora_layers=text_encoder_lora_layers,
+            text_encoder_2_lora_layers=text_encoder_2_lora_layers,
+        )
+
+        # Final inference
+        # Load previous pipeline
+        vae = AutoencoderKL.from_pretrained(
+            vae_path,
+            subfolder="vae" if args.pretrained_vae_model_name_or_path is None else None,
+            revision=args.revision,
+            torch_dtype=weight_dtype,
+        )
+        pipeline = StableDiffusionXLPipeline.from_pretrained(
+            args.pretrained_model_name_or_path, vae=vae, revision=args.revision, torch_dtype=weight_dtype
+        )
+
+        # We train on the simplified learning objective. If we were previously predicting a variance, we need the scheduler to ignore it
+        scheduler_args = {}
+
+        if "variance_type" in pipeline.scheduler.config:
+            variance_type = pipeline.scheduler.config.variance_type
+
+            if variance_type in ["learned", "learned_range"]:
+                variance_type = "fixed_small"
+
+            scheduler_args["variance_type"] = variance_type
+
+        pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config, **scheduler_args)
+
+        # load attention processors
+        pipeline.load_lora_weights(args.output_dir)
+
+        # run inference
+        images = []
+        if args.validation_prompt and args.num_validation_images > 0:
+            pipeline = pipeline.to(accelerator.device)
+            generator = torch.Generator(device=accelerator.device).manual_seed(args.seed) if args.seed else None
+            images = [
+                pipeline(args.validation_prompt, num_inference_steps=25, generator=generator).images[0]
+                for _ in range(args.num_validation_images)
+            ]
+
+            for tracker in accelerator.trackers:
+                if tracker.name == "tensorboard":
+                    np_images = np.stack([np.asarray(img) for img in images])
+                    tracker.writer.add_images("test", np_images, epoch, dataformats="NHWC")
+                if tracker.name == "wandb":
+                    tracker.log(
+                        {
+                            "test": [
+                                wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
+                                for i, image in enumerate(images)
+                            ]
+                        }
+                    )
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id,
+                images=images,
+                dataset_id=args.dataset_id,
+                base_model=args.pretrained_model_name_or_path,
+                train_text_encoder=args.train_text_encoder,
+                prompt=args.instance_prompt,
+                repo_folder=args.output_dir,
+                vae_path=args.pretrained_vae_model_name_or_path,
+            )
+            upload_folder(
+                repo_id=repo_id,
+                folder_path=args.output_dir,
+                commit_message="End of training",
+                ignore_patterns=["step_*", "epoch_*"],
+                token=args.hub_token
+            )
+
+    accelerator.end_training()
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)