Duplicate from Intel/textual-inversion-training

Co-authored-by: lvkaokao <lvkaokao@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,34 @@
+*.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
+*.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

README.md

@@ -0,0 +1,14 @@
+---
+title: Textual Inversion Training
+emoji: 📉
+colorFrom: purple
+colorTo: red
+sdk: gradio
+sdk_version: 3.14.0
+app_file: app.py
+pinned: false
+license: apache-2.0
+duplicated_from: Intel/textual-inversion-training
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,559 @@
+import gradio as gr
+import os
+from pathlib import Path
+import argparse
+import shutil
+# from train_dreambooth import run_training
+from textual_inversion import run_training
+from convertosd import convert
+from PIL import Image
+from slugify import slugify
+import requests
+import torch
+import zipfile
+import tarfile
+import urllib.parse
+import gc
+from diffusers import StableDiffusionPipeline
+from huggingface_hub import snapshot_download
+
+
+is_spaces = True if "SPACE_ID" in os.environ else False
+#is_shared_ui = True if "IS_SHARED_UI" in os.environ else False
+if(is_spaces):
+    is_shared_ui = True if ("lvkaokao/textual-inversion-training" in os.environ['SPACE_ID'] or "Intel/textual-inversion-training" in os.environ['SPACE_ID']) else False
+else:
+    is_shared_ui = False
+
+css = '''
+    .instruction{position: absolute; top: 0;right: 0;margin-top: 0px !important}
+    .arrow{position: absolute;top: 0;right: -110px;margin-top: -8px !important}
+    #component-4, #component-3, #component-10{min-height: 0}
+    .duplicate-button img{margin: 0}
+'''
+maximum_concepts = 1
+
+#Pre download the files
+'''
+model_v1_4 = snapshot_download(repo_id="CompVis/stable-diffusion-v1-4")
+#model_v1_5 = snapshot_download(repo_id="runwayml/stable-diffusion-v1-5")
+model_v1_5 = snapshot_download(repo_id="stabilityai/stable-diffusion-2")
+model_v2_512 = snapshot_download(repo_id="stabilityai/stable-diffusion-2-base", revision="fp16")
+safety_checker = snapshot_download(repo_id="multimodalart/sd-sc")
+'''
+model_v1_4 = "CompVis/stable-diffusion-v1-4"
+model_v1_5 = "stabilityai/stable-diffusion-2"
+model_v2_512 = "stabilityai/stable-diffusion-2-base"
+
+model_to_load = model_v1_4
+
+
+with zipfile.ZipFile("mix.zip", 'r') as zip_ref:
+    zip_ref.extractall(".")
+
+def swap_text(option):
+    mandatory_liability = "You must have the right to do so and you are liable for the images you use, example:"
+    if(option == "object"):
+        instance_prompt_example = "cttoy"
+        freeze_for = 30
+        return [f"You are going to train `object`(s), upload 5-10 images of each object you are planning on training on from different angles/perspectives. {mandatory_liability}:", '''<img src="file/cat-toy.png" />''', f"You should name your concept with a unique made up word that has low chance of the model already knowing it (e.g.: `{instance_prompt_example}` here). Images will be automatically cropped to 512x512.", freeze_for, gr.update(visible=False)]
+    elif(option == "person"):
+       instance_prompt_example = "julcto"
+       freeze_for = 70
+       return [f"You are going to train a `person`(s), upload 10-20 images of each person you are planning on training on from different angles/perspectives. {mandatory_liability}:", '''<img src="file/person.png" />''', f"You should name your concept with a unique made up word that has low chance of the model already knowing it (e.g.: `{instance_prompt_example}` here). Images will be automatically cropped to 512x512.", freeze_for, gr.update(visible=True)]
+    elif(option == "style"):
+        instance_prompt_example = "trsldamrl"
+        freeze_for = 10
+        return [f"You are going to train a `style`, upload 10-20 images of the style you are planning on training on. Name the files with the words you would like  {mandatory_liability}:", '''<img src="file/trsl_style.png" />''', f"You should name your concept with a unique made up word that has low chance of the model already knowing it (e.g.: `{instance_prompt_example}` here). Images will be automatically cropped to 512x512.", freeze_for, gr.update(visible=False)]
+
+def swap_base_model(selected_model):
+    global model_to_load
+    if(selected_model == "v1-4"):
+        model_to_load = model_v1_4
+    elif(selected_model == "v1-5"):
+        model_to_load = model_v1_5
+    else:
+        model_to_load = model_v2_512
+
+def count_files(*inputs):
+    file_counter = 0
+    concept_counter = 0
+    for i, input in enumerate(inputs):
+        if(i < maximum_concepts-1):
+            files = inputs[i]
+            if(files):
+                concept_counter+=1
+                file_counter+=len(files)
+    uses_custom = inputs[-1] 
+    type_of_thing = inputs[-4]
+    if(uses_custom):
+        Training_Steps = int(inputs[-3])
+    else:
+        Training_Steps = file_counter*200
+        if(Training_Steps > 2400):
+            Training_Steps=2400
+        elif(Training_Steps < 1400):
+            Training_Steps=1400
+    if(is_spaces):
+        summary_sentence = f'''The training should take around 24 hours for 1000 steps using the default free CPU.<br><br>'''
+    else:
+        summary_sentence = f'''You are going to train {concept_counter} {type_of_thing}(s), with {file_counter} images for {Training_Steps} steps.<br><br>'''
+        
+    return([gr.update(visible=True), gr.update(visible=True, value=summary_sentence)])
+
+def update_steps(*files_list):
+    file_counter = 0
+    for i, files in enumerate(files_list):
+        if(files):
+            file_counter+=len(files)
+    return(gr.update(value=file_counter*200))
+
+def pad_image(image):
+    w, h = image.size
+    if w == h:
+        return image
+    elif w > h:
+        new_image = Image.new(image.mode, (w, w), (0, 0, 0))
+        new_image.paste(image, (0, (w - h) // 2))
+        return new_image
+    else:
+        new_image = Image.new(image.mode, (h, h), (0, 0, 0))
+        new_image.paste(image, ((h - w) // 2, 0))
+        return new_image
+
+def train(*inputs):
+    if is_shared_ui:
+        raise gr.Error("This Space only works in duplicated instances")
+    
+    torch.cuda.empty_cache()
+    if 'pipe' in globals():
+        global pipe, pipe_is_set
+        del pipe
+        pipe_is_set = False
+        gc.collect()
+        
+    if os.path.exists("output_model"): shutil.rmtree('output_model')
+    if os.path.exists("concept_images"): shutil.rmtree('concept_images')
+    if os.path.exists("diffusers_model.tar"): os.remove("diffusers_model.tar")
+    if os.path.exists("model.ckpt"): os.remove("model.ckpt")
+    if os.path.exists("hastrained.success"): os.remove("hastrained.success")
+    file_counter = 0
+    print(inputs)
+
+    os.makedirs('concept_images', exist_ok=True)
+    files = inputs[maximum_concepts*3]
+    init_word = inputs[maximum_concepts*2]
+    prompt = inputs[maximum_concepts]
+    if(prompt == "" or prompt == None):
+        raise gr.Error("You forgot to define your concept prompt")
+
+    for j, file_temp in enumerate(files):
+        file = Image.open(file_temp.name)
+        image = pad_image(file)
+        image = image.resize((512, 512))
+        extension = file_temp.name.split(".")[1]
+        image = image.convert('RGB')
+        image.save(f'concept_images/{j+1}.jpg', format="JPEG", quality = 100)
+        file_counter += 1
+
+
+    os.makedirs('output_model',exist_ok=True)
+    uses_custom = inputs[-1] 
+    type_of_thing = inputs[-4]
+    remove_attribution_after = inputs[-6]
+    experimental_face_improvement = inputs[-9]
+    which_model = inputs[-10]
+    if(uses_custom):
+        Training_Steps = int(inputs[-3])
+    else:
+        Training_Steps = 1000
+
+    print(os.listdir("concept_images"))
+
+    args_general = argparse.Namespace(
+        pretrained_model_name_or_path = model_to_load,
+        train_data_dir="concept_images",
+        learnable_property=type_of_thing,
+        placeholder_token=prompt,
+        initializer_token=init_word,
+        resolution=512,
+        train_batch_size=1,
+        gradient_accumulation_steps=2,
+        use_bf16=True,
+        max_train_steps=Training_Steps,
+        learning_rate=5.0e-4,
+        scale_lr=True,
+        lr_scheduler="constant",
+        lr_warmup_steps=0,
+        output_dir="output_model",
+    )
+    print("Starting single training...")
+    lock_file = open("intraining.lock", "w")
+    lock_file.close()
+    run_training(args_general)
+
+    gc.collect()
+    torch.cuda.empty_cache()
+    if(which_model in ["v1-5"]):
+        print("Adding Safety Checker to the model...")
+        shutil.copytree(f"{safety_checker}/feature_extractor", "output_model/feature_extractor")
+        shutil.copytree(f"{safety_checker}/safety_checker", "output_model/safety_checker")
+        shutil.copy(f"model_index.json", "output_model/model_index.json")
+    
+    if(not remove_attribution_after):
+        print("Archiving model file...")
+        with tarfile.open("diffusers_model.tar", "w") as tar:
+            tar.add("output_model", arcname=os.path.basename("output_model"))
+        if os.path.exists("intraining.lock"): os.remove("intraining.lock")
+        trained_file = open("hastrained.success", "w")
+        trained_file.close()
+        print(os.listdir("output_model"))
+        print("Training completed!")
+        return [
+            gr.update(visible=True, value=["diffusers_model.tar"]), #result
+            gr.update(visible=True), #try_your_model
+            gr.update(visible=True), #push_to_hub
+            gr.update(visible=True), #convert_button
+            gr.update(visible=False), #training_ongoing
+            gr.update(visible=True) #completed_training
+        ]
+    else:
+        hf_token = inputs[-5]
+        model_name = inputs[-7]
+        where_to_upload = inputs[-8]
+        push(model_name, where_to_upload, hf_token, which_model, True)
+        hardware_url = f"https://huggingface.co/spaces/{os.environ['SPACE_ID']}/hardware"
+        headers = { "authorization" : f"Bearer {hf_token}"}
+        body = {'flavor': 'cpu-basic'}
+        requests.post(hardware_url, json = body, headers=headers)
+
+import time
+pipe_is_set = False
+def generate(prompt, steps):
+
+    print("prompt: ", prompt)
+    print("steps: ", steps)
+
+    torch.cuda.empty_cache()
+    from diffusers import StableDiffusionPipeline
+    global pipe_is_set
+    if(not pipe_is_set):
+        global pipe
+        if torch.cuda.is_available():
+            pipe = StableDiffusionPipeline.from_pretrained("./output_model", torch_dtype=torch.float16)
+            pipe = pipe.to("cuda")
+        else:
+            pipe = StableDiffusionPipeline.from_pretrained("./output_model", torch_dtype=torch.float)
+        pipe_is_set = True
+    
+    start_time = time.time()
+    image = pipe(prompt, num_inference_steps=steps, guidance_scale=7.5).images[0]
+    print("cost: ", time.time() - start_time)
+    return(image)
+    
+def push(model_name, where_to_upload, hf_token, which_model, comes_from_automated=False):
+
+    if(not os.path.exists("model.ckpt")):
+        convert("output_model", "model.ckpt")
+    from huggingface_hub import HfApi, HfFolder, CommitOperationAdd
+    from huggingface_hub import create_repo
+    model_name_slug = slugify(model_name)
+    api = HfApi()
+    your_username = api.whoami(token=hf_token)["name"]
+    if(where_to_upload == "My personal profile"):    
+        model_id = f"{your_username}/{model_name_slug}"
+    else:
+        model_id = f"sd-dreambooth-library/{model_name_slug}"
+        headers = {"Authorization" : f"Bearer: {hf_token}", "Content-Type": "application/json"}
+        response = requests.post("https://huggingface.co/organizations/sd-dreambooth-library/share/SSeOwppVCscfTEzFGQaqpfcjukVeNrKNHX", headers=headers)
+    
+    images_upload = os.listdir("concept_images")
+    image_string = ""
+    instance_prompt_list = []
+    previous_instance_prompt = ''
+    for i, image in enumerate(images_upload):
+        instance_prompt = image.split("_")[0]
+        if(instance_prompt != previous_instance_prompt):
+            title_instance_prompt_string = instance_prompt
+            instance_prompt_list.append(instance_prompt)
+        else:
+            title_instance_prompt_string = ''
+        previous_instance_prompt = instance_prompt
+        image_string = f'''{title_instance_prompt_string} {"(use that on your prompt)" if title_instance_prompt_string != "" else ""} 
+{image_string}![{instance_prompt} {i}](https://huggingface.co/{model_id}/resolve/main/concept_images/{urllib.parse.quote(image)})'''
+    readme_text = f'''---
+license: creativeml-openrail-m
+tags:
+- text-to-image
+---
+### {model_name} Dreambooth model trained by {api.whoami(token=hf_token)["name"]} with [Hugging Face Dreambooth Training Space](https://huggingface.co/spaces/multimodalart/dreambooth-training) with the {which_model} base model
+
+You run your new concept via `diffusers` [Colab Notebook for Inference](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_dreambooth_inference.ipynb). Don't forget to use the concept prompts! 
+
+Sample pictures of:
+{image_string}
+'''
+    #Save the readme to a file
+    readme_file = open("model.README.md", "w")
+    readme_file.write(readme_text)
+    readme_file.close()
+    #Save the token identifier to a file
+    text_file = open("token_identifier.txt", "w")
+    text_file.write(', '.join(instance_prompt_list))
+    text_file.close()
+    try:
+        create_repo(model_id,private=True, token=hf_token)
+    except:
+        import time
+        epoch_time = str(int(time.time()))
+        create_repo(f"{model_id}-{epoch_time}", private=True,token=hf_token)
+    operations = [
+        CommitOperationAdd(path_in_repo="token_identifier.txt", path_or_fileobj="token_identifier.txt"),
+        CommitOperationAdd(path_in_repo="README.md", path_or_fileobj="model.README.md"),
+        CommitOperationAdd(path_in_repo=f"model.ckpt",path_or_fileobj="model.ckpt")
+    ]
+    api.create_commit(
+    repo_id=model_id,
+    operations=operations,
+    commit_message=f"Upload the model {model_name}",
+    token=hf_token
+    )
+    api.upload_folder(
+    folder_path="output_model",
+    repo_id=model_id,
+    token=hf_token
+    )
+    api.upload_folder(
+    folder_path="concept_images",
+    path_in_repo="concept_images",
+    repo_id=model_id,
+    token=hf_token
+    )
+    if is_spaces:
+        if(not comes_from_automated):
+            extra_message = "Don't forget to remove the GPU attribution after you play with it."
+        else:
+            extra_message = "The GPU has been removed automatically as requested, and you can try the model via the model page"
+        api.create_discussion(repo_id=os.environ['SPACE_ID'], title=f"Your model {model_name} has finished trained from the Dreambooth Train Spaces!", description=f"Your model has been successfully uploaded to: https://huggingface.co/{model_id}. {extra_message}",repo_type="space", token=hf_token)
+
+    return [gr.update(visible=True, value=f"Successfully uploaded your model. Access it [here](https://huggingface.co/{model_id})"), gr.update(visible=True, value=["diffusers_model.tar", "model.ckpt"])]
+
+def convert_to_ckpt():
+    convert("output_model", "model.ckpt")
+    return gr.update(visible=True, value=["diffusers_model.tar", "model.ckpt"])
+
+def check_status(top_description):
+    print('=='*20)
+    print(os.listdir("./"))
+
+    if os.path.exists("hastrained.success"):
+        if is_spaces:
+            update_top_tag = gr.update(value=f'''
+            <div class="gr-prose" style="max-width: 80%">
+                <h2>Your model has finished training ✅</h2>
+                <p>Yay, congratulations on training your model. Scroll down to play with with it, save it (either downloading it or on the Hugging Face Hub). Once you are done, your model is safe, and you don't want to train a new one, go to the <a href="https://huggingface.co/spaces/{os.environ['SPACE_ID']}">settings page</a> and downgrade your Space to a CPU Basic</p> 
+            </div>
+            ''')
+        else:
+            update_top_tag = gr.update(value=f'''
+            <div class="gr-prose" style="max-width: 80%">
+                <h2>Your model has finished training ✅</h2>
+                <p>Yay, congratulations on training your model. Scroll down to play with with it, save it (either downloading it or on the Hugging Face Hub).</p> 
+            </div>
+            ''')
+        show_outputs = True
+    elif os.path.exists("intraining.lock"):
+        update_top_tag = gr.update(value='''
+        <div class="gr-prose" style="max-width: 80%">
+            <h2>Don't worry, your model is still training! ⌛</h2>
+            <p>You closed the tab while your model was training, but it's all good! It is still training right now. You can click the "Open logs" button above here to check the training status. Once training is done, reload this tab to interact with your model</p> 
+        </div>
+        ''')
+        show_outputs = False
+    else:
+        update_top_tag = gr.update(value=top_description)
+        show_outputs = False
+    if os.path.exists("diffusers_model.tar"):
+       update_files_tag = gr.update(visible=show_outputs, value=["diffusers_model.tar"])
+    else:
+       update_files_tag = gr.update(visible=show_outputs)
+    return [
+        update_top_tag, #top_description
+        gr.update(visible=show_outputs), #try_your_model
+        gr.update(visible=show_outputs), #push_to_hub
+        update_files_tag, #result
+        gr.update(visible=show_outputs), #convert_button
+    ]
+
+def checkbox_swap(checkbox):
+    return [gr.update(visible=checkbox), gr.update(visible=checkbox), gr.update(visible=checkbox), gr.update(visible=checkbox)]
+
+with gr.Blocks(css=css) as demo:
+    with gr.Box():
+        if is_shared_ui:
+            top_description = gr.HTML(f'''
+                <div class="gr-prose" style="max-width: 80%">
+                <h2>Attention - This Space doesn't work in this shared UI</h2>
+                <p>For it to work, you can either run locally or duplicate the Space and run it on your own profile using the free CPU or a (paid) private T4 GPU for training. CPU training takes a long time while each T4 costs US$0.60/h which should cost < US$1 to train most models using default settings!&nbsp;&nbsp;<a class="duplicate-button" style="display:inline-block" href="https://huggingface.co/spaces/{os.environ['SPACE_ID']}?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a></p>
+                <img class="instruction" src="file/duplicate.png"> 
+                <img class="arrow" src="file/arrow.png" />
+                </div>
+            ''')
+        elif(is_spaces):
+            top_description = gr.HTML(f'''
+                    <div class="gr-prose" style="max-width: 80%">
+                    <h2>You have successfully duplicated the Textual Inversion Training Space 🎉</h2>
+                    <p>If you want to use CPU, it will take a long time to run the training below. If you want to use GPU, please get this ready: <a href="https://huggingface.co/spaces/{os.environ['SPACE_ID']}/settings">attribute a T4 GPU 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 it off.</p> 
+                    </div>
+            ''')
+        else:
+            top_description = gr.HTML(f'''
+                    <div class="gr-prose" style="max-width: 80%">
+                    <h2>You have successfully cloned the Dreambooth Training Space locally 🎉</h2>
+                    <p>Do a <code>pip install requirements-local.txt</code></p> 
+                    </div>
+                ''')
+    gr.Markdown("# Textual Inversion Training UI 💭")
+    gr.Markdown("Customize Stable Diffusion by training it on a new concept. This Space is based on [Intel® Neural Compressor](https://github.com/intel/neural-compressor/tree/master/examples/pytorch/diffusion_model/diffusers/textual_inversion) with [🧨 diffusers](https://github.com/huggingface/diffusers)")
+    
+    with gr.Row() as what_are_you_training:
+        type_of_thing = gr.Dropdown(label="What would you like to train?", choices=["object", "person", "style"], value="object", interactive=True)
+        base_model_to_use = gr.Dropdown(label="Which base model would you like to use?", choices=["v1-4", "v1-5", "v2-512"], value="v1-4", interactive=True)
+    
+    #Very hacky approach to emulate dynamically created Gradio components   
+    with gr.Row() as upload_your_concept:
+        with gr.Column():
+            thing_description = gr.Markdown("You are going to train an `object`, please upload 1-5 images of the object to teach new concepts to Stable Diffusion, example")
+            thing_experimental = gr.Checkbox(label="Improve faces (prior preservation) - can take longer training but can improve faces", visible=False, value=False)
+            thing_image_example = gr.HTML('''<img src="file/dicoo-toy.png" class="aligncenter" height="128" width="128" />''')
+            things_naming = gr.Markdown("You should name your concept with a unique made up word that never appears in the model vocab (e.g.: `dicoo*` here). **The meaning of the initial word** is to initialize the concept word embedding which will make training easy (e.g.: `toy` here). Images will be automatically cropped to 512x512.")
+            
+        with gr.Column():
+            file_collection = []
+            concept_collection = []
+            init_collection = []
+            buttons_collection = []
+            delete_collection = []
+            is_visible = []
+
+            row = [None] * maximum_concepts
+            for x in range(maximum_concepts):
+                ordinal = lambda n: "%d%s" % (n, "tsnrhtdd"[(n // 10 % 10 != 1) * (n % 10 < 4) * n % 10::4])
+                if(x == 0):
+                    visible = True
+                    is_visible.append(gr.State(value=True))
+                else:
+                    visible = False
+                    is_visible.append(gr.State(value=False))
+
+                file_collection.append(gr.File(label=f'''Upload the images for your {ordinal(x+1) if (x>0) else ""} concept''', file_count="multiple", interactive=True, visible=visible))
+                with gr.Column(visible=visible) as row[x]:
+                    concept_collection.append(gr.Textbox(label=f'''{ordinal(x+1) if (x>0) else ""} concept word - use a unique, made up word to avoid collisions'''))
+                    init_collection.append(gr.Textbox(label=f'''{ordinal(x+1) if (x>0) else ""} initial word - to init the concept embedding'''))
+                    with gr.Row():
+                        if(x < maximum_concepts-1):
+                            buttons_collection.append(gr.Button(value="Add +1 concept", visible=visible))
+                        if(x > 0):
+                            delete_collection.append(gr.Button(value=f"Delete {ordinal(x+1)} concept"))
+            
+            counter_add = 1
+            for button in buttons_collection:
+                if(counter_add < len(buttons_collection)):
+                    button.click(lambda:
+                    [gr.update(visible=True),gr.update(visible=True), gr.update(visible=False), gr.update(visible=True), True, None],
+                    None, 
+                    [row[counter_add], file_collection[counter_add], buttons_collection[counter_add-1], buttons_collection[counter_add], is_visible[counter_add], file_collection[counter_add]], queue=False)
+                else:
+                    button.click(lambda:[gr.update(visible=True),gr.update(visible=True), gr.update(visible=False), True], None, [row[counter_add], file_collection[counter_add], buttons_collection[counter_add-1], is_visible[counter_add]], queue=False)
+                counter_add += 1
+            
+            counter_delete = 1
+            for delete_button in delete_collection:
+                if(counter_delete < len(delete_collection)+1):
+                    delete_button.click(lambda:[gr.update(visible=False),gr.update(visible=False), gr.update(visible=True), False], None, [file_collection[counter_delete], row[counter_delete], buttons_collection[counter_delete-1], is_visible[counter_delete]], queue=False)
+                counter_delete += 1
+            
+    with gr.Accordion("Custom Settings", open=False):
+        swap_auto_calculated = gr.Checkbox(label="Use custom settings")
+        gr.Markdown("The default steps is 1000. If your results aren't really what you wanted, it may be underfitting and you need more steps.")
+        steps = gr.Number(label="How many steps", value=1000)
+        # need to remove
+        perc_txt_encoder = gr.Number(label="Percentage of the training steps the text-encoder should be trained as well", value=30, visible=False)
+        # perc_txt_encoder = 30
+        
+    with gr.Box(visible=False) as training_summary:
+        training_summary_text = gr.HTML("", visible=False, label="Training Summary")
+        is_advanced_visible = True if is_spaces else False
+        training_summary_checkbox = gr.Checkbox(label="Automatically remove paid GPU attribution and upload model to the Hugging Face Hub after training", value=False, visible=is_advanced_visible)
+        training_summary_model_name = gr.Textbox(label="Name of your model", visible=False)
+        training_summary_where_to_upload = gr.Dropdown(["My personal profile", "Public Library"], label="Upload to", visible=False)
+        training_summary_token_message = gr.Markdown("[A Hugging Face write access token](https://huggingface.co/settings/tokens), go to \"New token\" -> Role : Write. A regular read token won't work here.", visible=False)            
+        training_summary_token = gr.Textbox(label="Hugging Face Write Token", type="password", visible=False)
+        
+    train_btn = gr.Button("Start Training")
+    
+    training_ongoing = gr.Markdown("## Training is ongoing ⌛... You can close this tab if you like or just wait. If you did not check the `Remove GPU After training`, you can come back here to try your model and upload it after training. Don't forget to remove the GPU attribution after you are done. ", visible=False)
+    
+    #Post-training UI
+    completed_training = gr.Markdown('''# ✅ Training completed. 
+    ### Don't forget to remove the GPU attribution after you are done trying and uploading your model''', visible=False)
+    
+    with gr.Row():
+        with gr.Box(visible=True) as try_your_model:
+            gr.Markdown("## Try your model")
+            prompt = gr.Textbox(label="Type your prompt")
+            result_image = gr.Image()
+            inference_steps = gr.Slider(minimum=1, maximum=150, value=50, step=1)
+            generate_button = gr.Button("Generate Image")
+        
+        with gr.Box(visible=False) as push_to_hub:
+            gr.Markdown("## Push to Hugging Face Hub")
+            model_name = gr.Textbox(label="Name of your model", placeholder="Tarsila do Amaral Style")
+            where_to_upload = gr.Dropdown(["My personal profile", "Public Library"], label="Upload to")
+            gr.Markdown("[A Hugging Face write access token](https://huggingface.co/settings/tokens), go to \"New token\" -> Role : Write. A regular read token won't work here.")
+            hf_token = gr.Textbox(label="Hugging Face Write Token", type="password")
+            
+            push_button = gr.Button("Push to the Hub")
+    
+    result = gr.File(label="Download the uploaded models in the diffusers format", visible=True)
+    success_message_upload = gr.Markdown(visible=False)
+    convert_button = gr.Button("Convert to CKPT", visible=False)
+    
+    #Swap the examples and the % of text encoder trained depending if it is an object, person or style
+    type_of_thing.change(fn=swap_text, inputs=[type_of_thing], outputs=[thing_description, thing_image_example, things_naming, perc_txt_encoder, thing_experimental], queue=False, show_progress=False)
+    
+    #Swap the base model
+    base_model_to_use.change(fn=swap_base_model, inputs=base_model_to_use, outputs=[])
+
+    #Update the summary box below the UI according to how many images are uploaded and whether users are using custom settings or not 
+    for file in file_collection:
+        #file.change(fn=update_steps,inputs=file_collection, outputs=steps)
+        file.change(fn=count_files, inputs=file_collection+[type_of_thing]+[steps]+[perc_txt_encoder]+[swap_auto_calculated], outputs=[training_summary, training_summary_text], queue=False)
+        
+    steps.change(fn=count_files, inputs=file_collection+[type_of_thing]+[steps]+[perc_txt_encoder]+[swap_auto_calculated], outputs=[training_summary, training_summary_text], queue=False)
+    perc_txt_encoder.change(fn=count_files, inputs=file_collection+[type_of_thing]+[steps]+[perc_txt_encoder]+[swap_auto_calculated], outputs=[training_summary, training_summary_text], queue=False)
+    
+    #Give more options if the user wants to finish everything after training
+    if(is_spaces):
+        training_summary_checkbox.change(fn=checkbox_swap, inputs=training_summary_checkbox, outputs=[training_summary_token_message, training_summary_token, training_summary_model_name, training_summary_where_to_upload],queue=False, show_progress=False)
+    #Add a message for while it is in training
+    train_btn.click(lambda:gr.update(visible=True), inputs=None, outputs=training_ongoing)
+
+    #The main train function
+    train_btn.click(fn=train, inputs=is_visible+concept_collection+init_collection+file_collection+[base_model_to_use]+[thing_experimental]+[training_summary_where_to_upload]+[training_summary_model_name]+[training_summary_checkbox]+[training_summary_token]+[type_of_thing]+[steps]+[perc_txt_encoder]+[swap_auto_calculated], outputs=[result, try_your_model, push_to_hub, convert_button, training_ongoing, completed_training], queue=False)
+    
+    #Button to generate an image from your trained model after training
+    print('=='*20)
+    print(prompt)
+    print(inference_steps)
+    generate_button.click(fn=generate, inputs=[prompt, inference_steps], outputs=result_image, queue=False)
+
+    #Button to push the model to the Hugging Face Hub
+    push_button.click(fn=push, inputs=[model_name, where_to_upload, hf_token, base_model_to_use], outputs=[success_message_upload, result], queue=False)
+    #Button to convert the model to ckpt format 
+    convert_button.click(fn=convert_to_ckpt, inputs=[], outputs=result, queue=False)
+    
+    #Checks if the training is running
+    demo.load(fn=check_status, inputs=top_description, outputs=[top_description, try_your_model, push_to_hub, result, convert_button], queue=False, show_progress=False)
+
+demo.queue(default_enabled=False).launch(debug=True)

convertosd.py

@@ -0,0 +1,226 @@
+# Script for converting a HF Diffusers saved pipeline to a Stable Diffusion checkpoint.
+# *Only* converts the UNet, VAE, and Text Encoder.
+# Does not convert optimizer state or any other thing.
+# Written by jachiam
+
+import argparse
+import os.path as osp
+
+import torch
+import gc 
+
+# =================#
+# UNet Conversion #
+# =================#
+
+unet_conversion_map = [
+    # (stable-diffusion, HF Diffusers)
+    ("time_embed.0.weight", "time_embedding.linear_1.weight"),
+    ("time_embed.0.bias", "time_embedding.linear_1.bias"),
+    ("time_embed.2.weight", "time_embedding.linear_2.weight"),
+    ("time_embed.2.bias", "time_embedding.linear_2.bias"),
+    ("input_blocks.0.0.weight", "conv_in.weight"),
+    ("input_blocks.0.0.bias", "conv_in.bias"),
+    ("out.0.weight", "conv_norm_out.weight"),
+    ("out.0.bias", "conv_norm_out.bias"),
+    ("out.2.weight", "conv_out.weight"),
+    ("out.2.bias", "conv_out.bias"),
+]
+
+unet_conversion_map_resnet = [
+    # (stable-diffusion, HF Diffusers)
+    ("in_layers.0", "norm1"),
+    ("in_layers.2", "conv1"),
+    ("out_layers.0", "norm2"),
+    ("out_layers.3", "conv2"),
+    ("emb_layers.1", "time_emb_proj"),
+    ("skip_connection", "conv_shortcut"),
+]
+
+unet_conversion_map_layer = []
+# hardcoded number of downblocks and resnets/attentions...
+# would need smarter logic for other networks.
+for i in range(4):
+    # loop over downblocks/upblocks
+
+    for j in range(2):
+        # loop over resnets/attentions for downblocks
+        hf_down_res_prefix = f"down_blocks.{i}.resnets.{j}."
+        sd_down_res_prefix = f"input_blocks.{3*i + j + 1}.0."
+        unet_conversion_map_layer.append((sd_down_res_prefix, hf_down_res_prefix))
+
+        if i < 3:
+            # no attention layers in down_blocks.3
+            hf_down_atn_prefix = f"down_blocks.{i}.attentions.{j}."
+            sd_down_atn_prefix = f"input_blocks.{3*i + j + 1}.1."
+            unet_conversion_map_layer.append((sd_down_atn_prefix, hf_down_atn_prefix))
+
+    for j in range(3):
+        # loop over resnets/attentions for upblocks
+        hf_up_res_prefix = f"up_blocks.{i}.resnets.{j}."
+        sd_up_res_prefix = f"output_blocks.{3*i + j}.0."
+        unet_conversion_map_layer.append((sd_up_res_prefix, hf_up_res_prefix))
+
+        if i > 0:
+            # no attention layers in up_blocks.0
+            hf_up_atn_prefix = f"up_blocks.{i}.attentions.{j}."
+            sd_up_atn_prefix = f"output_blocks.{3*i + j}.1."
+            unet_conversion_map_layer.append((sd_up_atn_prefix, hf_up_atn_prefix))
+
+    if i < 3:
+        # no downsample in down_blocks.3
+        hf_downsample_prefix = f"down_blocks.{i}.downsamplers.0.conv."
+        sd_downsample_prefix = f"input_blocks.{3*(i+1)}.0.op."
+        unet_conversion_map_layer.append((sd_downsample_prefix, hf_downsample_prefix))
+
+        # no upsample in up_blocks.3
+        hf_upsample_prefix = f"up_blocks.{i}.upsamplers.0."
+        sd_upsample_prefix = f"output_blocks.{3*i + 2}.{1 if i == 0 else 2}."
+        unet_conversion_map_layer.append((sd_upsample_prefix, hf_upsample_prefix))
+
+hf_mid_atn_prefix = "mid_block.attentions.0."
+sd_mid_atn_prefix = "middle_block.1."
+unet_conversion_map_layer.append((sd_mid_atn_prefix, hf_mid_atn_prefix))
+
+for j in range(2):
+    hf_mid_res_prefix = f"mid_block.resnets.{j}."
+    sd_mid_res_prefix = f"middle_block.{2*j}."
+    unet_conversion_map_layer.append((sd_mid_res_prefix, hf_mid_res_prefix))
+
+
+def convert_unet_state_dict(unet_state_dict):
+    # buyer beware: this is a *brittle* function,
+    # and correct output requires that all of these pieces interact in
+    # the exact order in which I have arranged them.
+    mapping = {k: k for k in unet_state_dict.keys()}
+    for sd_name, hf_name in unet_conversion_map:
+        mapping[hf_name] = sd_name
+    for k, v in mapping.items():
+        if "resnets" in k:
+            for sd_part, hf_part in unet_conversion_map_resnet:
+                v = v.replace(hf_part, sd_part)
+            mapping[k] = v
+    for k, v in mapping.items():
+        for sd_part, hf_part in unet_conversion_map_layer:
+            v = v.replace(hf_part, sd_part)
+        mapping[k] = v
+    new_state_dict = {v: unet_state_dict[k] for k, v in mapping.items()}
+    return new_state_dict
+
+
+# ================#
+# VAE Conversion #
+# ================#
+
+vae_conversion_map = [
+    # (stable-diffusion, HF Diffusers)
+    ("nin_shortcut", "conv_shortcut"),
+    ("norm_out", "conv_norm_out"),
+    ("mid.attn_1.", "mid_block.attentions.0."),
+]
+
+for i in range(4):
+    # down_blocks have two resnets
+    for j in range(2):
+        hf_down_prefix = f"encoder.down_blocks.{i}.resnets.{j}."
+        sd_down_prefix = f"encoder.down.{i}.block.{j}."
+        vae_conversion_map.append((sd_down_prefix, hf_down_prefix))
+
+    if i < 3:
+        hf_downsample_prefix = f"down_blocks.{i}.downsamplers.0."
+        sd_downsample_prefix = f"down.{i}.downsample."
+        vae_conversion_map.append((sd_downsample_prefix, hf_downsample_prefix))
+
+        hf_upsample_prefix = f"up_blocks.{i}.upsamplers.0."
+        sd_upsample_prefix = f"up.{3-i}.upsample."
+        vae_conversion_map.append((sd_upsample_prefix, hf_upsample_prefix))
+
+    # up_blocks have three resnets
+    # also, up blocks in hf are numbered in reverse from sd
+    for j in range(3):
+        hf_up_prefix = f"decoder.up_blocks.{i}.resnets.{j}."
+        sd_up_prefix = f"decoder.up.{3-i}.block.{j}."
+        vae_conversion_map.append((sd_up_prefix, hf_up_prefix))
+
+# this part accounts for mid blocks in both the encoder and the decoder
+for i in range(2):
+    hf_mid_res_prefix = f"mid_block.resnets.{i}."
+    sd_mid_res_prefix = f"mid.block_{i+1}."
+    vae_conversion_map.append((sd_mid_res_prefix, hf_mid_res_prefix))
+
+
+vae_conversion_map_attn = [
+    # (stable-diffusion, HF Diffusers)
+    ("norm.", "group_norm."),
+    ("q.", "query."),
+    ("k.", "key."),
+    ("v.", "value."),
+    ("proj_out.", "proj_attn."),
+]
+
+
+def reshape_weight_for_sd(w):
+    # convert HF linear weights to SD conv2d weights
+    return w.reshape(*w.shape, 1, 1)
+
+
+def convert_vae_state_dict(vae_state_dict):
+    mapping = {k: k for k in vae_state_dict.keys()}
+    for k, v in mapping.items():
+        for sd_part, hf_part in vae_conversion_map:
+            v = v.replace(hf_part, sd_part)
+        mapping[k] = v
+    for k, v in mapping.items():
+        if "attentions" in k:
+            for sd_part, hf_part in vae_conversion_map_attn:
+                v = v.replace(hf_part, sd_part)
+            mapping[k] = v
+    new_state_dict = {v: vae_state_dict[k] for k, v in mapping.items()}
+    weights_to_convert = ["q", "k", "v", "proj_out"]
+    print("Converting to CKPT ...")    
+    for k, v in new_state_dict.items():
+        for weight_name in weights_to_convert:
+            if f"mid.attn_1.{weight_name}.weight" in k:
+                new_state_dict[k] = reshape_weight_for_sd(v)
+    return new_state_dict
+
+
+# =========================#
+# Text Encoder Conversion #
+# =========================#
+# pretty much a no-op
+
+
+def convert_text_enc_state_dict(text_enc_dict):
+    return text_enc_dict
+
+
+def convert(model_path, checkpoint_path):    
+    unet_path = osp.join(model_path, "unet", "diffusion_pytorch_model.bin")
+    vae_path = osp.join(model_path, "vae", "diffusion_pytorch_model.bin")
+    text_enc_path = osp.join(model_path, "text_encoder", "pytorch_model.bin")
+
+    # Convert the UNet model
+    unet_state_dict = torch.load(unet_path, map_location='cpu')
+    unet_state_dict = convert_unet_state_dict(unet_state_dict)
+    unet_state_dict = {"model.diffusion_model." + k: v for k, v in unet_state_dict.items()}
+
+    # Convert the VAE model
+    vae_state_dict = torch.load(vae_path, map_location='cpu')
+    vae_state_dict = convert_vae_state_dict(vae_state_dict)
+    vae_state_dict = {"first_stage_model." + k: v for k, v in vae_state_dict.items()}
+
+    # Convert the text encoder model
+    text_enc_dict = torch.load(text_enc_path, map_location='cpu')
+    text_enc_dict = convert_text_enc_state_dict(text_enc_dict)
+    text_enc_dict = {"cond_stage_model.transformer." + k: v for k, v in text_enc_dict.items()}
+
+    # Put together new checkpoint
+    state_dict = {**unet_state_dict, **vae_state_dict, **text_enc_dict}
+    
+    state_dict = {k:v.half() for k,v in state_dict.items()}
+    state_dict = {"state_dict": state_dict}
+    torch.save(state_dict, checkpoint_path)
+    del state_dict, text_enc_dict, vae_state_dict, unet_state_dict
+    torch.cuda.empty_cache()
+    gc.collect()

mix.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09207c4e95fcf5296eb0ff708fdc672da960aeb2864d298810db5094b072a0d4
+size 28022653

model_index.json

@@ -0,0 +1,32 @@
+{
+  "_class_name": "StableDiffusionPipeline",
+  "_diffusers_version": "0.6.0",
+  "feature_extractor": [
+    "transformers",
+    "CLIPFeatureExtractor"
+  ],
+  "safety_checker": [
+    "stable_diffusion",
+    "StableDiffusionSafetyChecker"
+  ],
+  "scheduler": [
+    "diffusers",
+    "PNDMScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "CLIPTextModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "CLIPTokenizer"
+  ],
+  "unet": [
+    "diffusers",
+    "UNet2DConditionModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKL"
+  ]
+}

requirements-local.txt

@@ -0,0 +1,18 @@
+--extra-index-url https://download.pytorch.org/whl/cu113
+torch==1.12.1+cu113
+torchvision==0.13.1+cu113
+diffusers==0.9.0
+accelerate==0.12.0
+OmegaConf
+wget
+pytorch_lightning
+huggingface_hub
+ftfy
+transformers
+pyfiglet
+triton==2.0.0.dev20220701
+bitsandbytes
+python-slugify
+requests
+tensorboard
+pip install git+https://github.com/facebookresearch/xformers@7e4c02c#egg=xformers

requirements.txt

@@ -0,0 +1,18 @@
+--extra-index-url https://download.pytorch.org/whl/cu113
+torch==1.12.1+cu113
+torchvision==0.13.1+cu113
+diffusers==0.9.0
+accelerate==0.12.0
+OmegaConf
+wget
+pytorch_lightning
+huggingface_hub
+ftfy
+transformers
+pyfiglet
+triton==2.0.0.dev20220701
+bitsandbytes
+python-slugify
+requests
+tensorboard
+https://github.com/apolinario/xformers/releases/download/0.0.2/xformers-0.0.14.dev0-cp38-cp38-linux_x86_64.whl

textual_inversion.py

@@ -0,0 +1,612 @@
+import argparse
+import itertools
+import math
+import os
+import random
+from pathlib import Path
+from typing import Optional
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch.utils.data import Dataset
+
+import PIL
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import set_seed
+from diffusers import AutoencoderKL, DDPMScheduler, PNDMScheduler, StableDiffusionPipeline, UNet2DConditionModel
+from diffusers.optimization import get_scheduler
+from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
+from huggingface_hub import HfFolder, Repository, whoami
+from PIL import Image
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+import gc
+
+logger = get_logger(__name__)
+
+
+def save_progress(text_encoder, placeholder_token_id, accelerator, args):
+    logger.info("Saving embeddings")
+    learned_embeds = accelerator.unwrap_model(text_encoder).get_input_embeddings().weight[placeholder_token_id]
+    learned_embeds_dict = {args.placeholder_token: learned_embeds.detach().cpu()}
+    torch.save(learned_embeds_dict, os.path.join(args.output_dir, "learned_embeds.bin"))
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--save_steps",
+        type=int,
+        default=500,
+        help="Save learned_embeds.bin every X updates steps.",
+    )
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--tokenizer_name",
+        type=str,
+        default=None,
+        help="Pretrained tokenizer name or path if not the same as model_name",
+    )
+    parser.add_argument(
+        "--train_data_dir", type=str, default=None, help="A folder containing the training data."
+    )
+    parser.add_argument(
+        "--placeholder_token",
+        type=str,
+        default=None,
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--initializer_token", type=str, default=None, help="A token to use as initializer word."
+    )
+    parser.add_argument("--learnable_property", type=str, default="object", help="Choose between 'object' and 'style'")
+    parser.add_argument("--repeats", type=int, default=100, help="How many times to repeat the training data.")
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="text-inversion-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=512,
+        help=(
+            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
+            " resolution"
+        ),
+    )
+    parser.add_argument(
+        "--center_crop", action="store_true", help="Whether to center crop images before resizing to resolution"
+    )
+    parser.add_argument(
+        "--train_batch_size", type=int, default=16, help="Batch size (per device) for the training dataloader."
+    )
+    parser.add_argument("--num_train_epochs", type=int, default=100)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=5000,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    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(
+        "--learning_rate",
+        type=float,
+        default=1e-4,
+        help="Initial learning rate (after the potential warmup period) to use.",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        action="store_true",
+        default=True,
+        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("--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("--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(
+        "--mixed_precision",
+        type=str,
+        default="no",
+        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."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+
+    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.train_data_dir is None:
+        raise ValueError("You must specify a train data directory.")
+    '''
+
+    return args
+
+
+imagenet_templates_small = [
+    "a photo of a {}",
+    "a rendering of a {}",
+    "a cropped photo of the {}",
+    "the photo of a {}",
+    "a photo of a clean {}",
+    "a photo of a dirty {}",
+    "a dark photo of the {}",
+    "a photo of my {}",
+    "a photo of the cool {}",
+    "a close-up photo of a {}",
+    "a bright photo of the {}",
+    "a cropped photo of a {}",
+    "a photo of the {}",
+    "a good photo of the {}",
+    "a photo of one {}",
+    "a close-up photo of the {}",
+    "a rendition of the {}",
+    "a photo of the clean {}",
+    "a rendition of a {}",
+    "a photo of a nice {}",
+    "a good photo of a {}",
+    "a photo of the nice {}",
+    "a photo of the small {}",
+    "a photo of the weird {}",
+    "a photo of the large {}",
+    "a photo of a cool {}",
+    "a photo of a small {}",
+]
+
+imagenet_style_templates_small = [
+    "a painting in the style of {}",
+    "a rendering in the style of {}",
+    "a cropped painting in the style of {}",
+    "the painting in the style of {}",
+    "a clean painting in the style of {}",
+    "a dirty painting in the style of {}",
+    "a dark painting in the style of {}",
+    "a picture in the style of {}",
+    "a cool painting in the style of {}",
+    "a close-up painting in the style of {}",
+    "a bright painting in the style of {}",
+    "a cropped painting in the style of {}",
+    "a good painting in the style of {}",
+    "a close-up painting in the style of {}",
+    "a rendition in the style of {}",
+    "a nice painting in the style of {}",
+    "a small painting in the style of {}",
+    "a weird painting in the style of {}",
+    "a large painting in the style of {}",
+]
+
+
+class TextualInversionDataset(Dataset):
+    def __init__(
+        self,
+        data_root,
+        tokenizer,
+        learnable_property="object",  # [object, style]
+        size=512,
+        repeats=100,
+        interpolation="bicubic",
+        flip_p=0.5,
+        set="train",
+        placeholder_token="*",
+        center_crop=False,
+    ):
+        self.data_root = data_root
+        self.tokenizer = tokenizer
+        self.learnable_property = learnable_property
+        self.size = size
+        self.placeholder_token = placeholder_token
+        self.center_crop = center_crop
+        self.flip_p = flip_p
+
+        self.image_paths = [os.path.join(self.data_root, file_path) for file_path in os.listdir(self.data_root)]
+
+        self.num_images = len(self.image_paths)
+        self._length = self.num_images
+
+        if set == "train":
+            self._length = self.num_images * repeats
+
+        self.interpolation = {
+            "linear": PIL.Image.LINEAR,
+            "bilinear": PIL.Image.BILINEAR,
+            "bicubic": PIL.Image.BICUBIC,
+            "lanczos": PIL.Image.LANCZOS,
+        }[interpolation]
+
+        self.templates = imagenet_style_templates_small if learnable_property == "style" else imagenet_templates_small
+        self.flip_transform = transforms.RandomHorizontalFlip(p=self.flip_p)
+
+    def __len__(self):
+        return self._length
+
+    def __getitem__(self, i):
+        example = {}
+        image = Image.open(self.image_paths[i % self.num_images])
+
+        if not image.mode == "RGB":
+            image = image.convert("RGB")
+
+        placeholder_string = self.placeholder_token
+        text = random.choice(self.templates).format(placeholder_string)
+
+        example["input_ids"] = self.tokenizer(
+            text,
+            padding="max_length",
+            truncation=True,
+            max_length=self.tokenizer.model_max_length,
+            return_tensors="pt",
+        ).input_ids[0]
+
+        # default to score-sde preprocessing
+        img = np.array(image).astype(np.uint8)
+
+        if self.center_crop:
+            crop = min(img.shape[0], img.shape[1])
+            h, w, = (
+                img.shape[0],
+                img.shape[1],
+            )
+            img = img[(h - crop) // 2 : (h + crop) // 2, (w - crop) // 2 : (w + crop) // 2]
+
+        image = Image.fromarray(img)
+        image = image.resize((self.size, self.size), resample=self.interpolation)
+
+        image = self.flip_transform(image)
+        image = np.array(image).astype(np.uint8)
+        image = (image / 127.5 - 1.0).astype(np.float32)
+
+        example["pixel_values"] = torch.from_numpy(image).permute(2, 0, 1)
+        return example
+
+
+def get_full_repo_name(model_id: str, organization: Optional[str] = None, token: Optional[str] = None):
+    if token is None:
+        token = HfFolder.get_token()
+    if organization is None:
+        username = whoami(token)["name"]
+        return f"{username}/{model_id}"
+    else:
+        return f"{organization}/{model_id}"
+
+
+def freeze_params(params):
+    for param in params:
+        param.requires_grad = False
+
+
+def merge_two_dicts(starting_dict: dict, updater_dict: dict) -> dict:
+    """
+    Starts from base starting dict and then adds the remaining key values from updater replacing the values from
+    the first starting/base dict with the second updater dict.
+
+    For later: how does d = {**d1, **d2} replace collision?
+
+    :param starting_dict:
+    :param updater_dict:
+    :return:
+    """
+    new_dict: dict = starting_dict.copy()   # start with keys and values of starting_dict
+    new_dict.update(updater_dict)    # modifies starting_dict with keys and values of updater_dict
+    return new_dict
+
+def merge_args(args1: argparse.Namespace, args2: argparse.Namespace) -> argparse.Namespace:
+    """
+
+    ref: https://stackoverflow.com/questions/56136549/how-can-i-merge-two-argparse-namespaces-in-python-2-x
+    :param args1:
+    :param args2:
+    :return:
+    """
+    # - the merged args
+    # The vars() function returns the __dict__ attribute to values of the given object e.g {field:value}.
+    merged_key_values_for_namespace: dict = merge_two_dicts(vars(args1), vars(args2))
+    args = argparse.Namespace(**merged_key_values_for_namespace)
+    return args
+
+def run_training(args_imported):
+    args_default = parse_args()
+    args = merge_args(args_default, args_imported)
+
+    print(args)
+    logging_dir = os.path.join(args.output_dir, args.logging_dir)
+
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with="tensorboard",
+        logging_dir=logging_dir,
+    )
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # Handle the repository creation
+    if accelerator.is_main_process:
+        if args.push_to_hub:
+            if args.hub_model_id is None:
+                repo_name = get_full_repo_name(Path(args.output_dir).name, token=args.hub_token)
+            else:
+                repo_name = args.hub_model_id
+            repo = Repository(args.output_dir, clone_from=repo_name)
+
+            with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
+                if "step_*" not in gitignore:
+                    gitignore.write("step_*\n")
+                if "epoch_*" not in gitignore:
+                    gitignore.write("epoch_*\n")
+        elif args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+    # Load the tokenizer and add the placeholder token as a additional special token
+    if args.tokenizer_name:
+        tokenizer = CLIPTokenizer.from_pretrained(args.tokenizer_name)
+    elif args.pretrained_model_name_or_path:
+        tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder="tokenizer")
+
+    # Add the placeholder token in tokenizer
+    num_added_tokens = tokenizer.add_tokens(args.placeholder_token)
+    if num_added_tokens == 0:
+        raise ValueError(
+            f"The tokenizer already contains the token {args.placeholder_token}. Please pass a different"
+            " `placeholder_token` that is not already in the tokenizer."
+        )
+
+    # Convert the initializer_token, placeholder_token to ids
+    token_ids = tokenizer.encode(args.initializer_token, add_special_tokens=False)
+    # Check if initializer_token is a single token or a sequence of tokens
+    if len(token_ids) > 1:
+        raise ValueError("The initializer token must be a single token.")
+
+    initializer_token_id = token_ids[0]
+    placeholder_token_id = tokenizer.convert_tokens_to_ids(args.placeholder_token)
+
+    # Load models and create wrapper for stable diffusion
+    text_encoder = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="text_encoder")
+    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae")
+    unet = UNet2DConditionModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="unet")
+
+    # Resize the token embeddings as we are adding new special tokens to the tokenizer
+    text_encoder.resize_token_embeddings(len(tokenizer))
+
+    # Initialise the newly added placeholder token with the embeddings of the initializer token
+    token_embeds = text_encoder.get_input_embeddings().weight.data
+    token_embeds[placeholder_token_id] = token_embeds[initializer_token_id]
+
+    # Freeze vae and unet
+    freeze_params(vae.parameters())
+    freeze_params(unet.parameters())
+    # Freeze all parameters except for the token embeddings in text encoder
+    params_to_freeze = itertools.chain(
+        text_encoder.text_model.encoder.parameters(),
+        text_encoder.text_model.final_layer_norm.parameters(),
+        text_encoder.text_model.embeddings.position_embedding.parameters(),
+    )
+    freeze_params(params_to_freeze)
+
+    if args.scale_lr:
+        args.learning_rate = (
+            args.learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
+        )
+
+    # Initialize the optimizer
+    optimizer = torch.optim.AdamW(
+        text_encoder.get_input_embeddings().parameters(),  # only optimize the embeddings
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    # TODO (patil-suraj): load scheduler using args
+    noise_scheduler = DDPMScheduler(
+        beta_start=0.00085,
+        beta_end=0.012,
+        beta_schedule="scaled_linear",
+        num_train_timesteps=1000,
+    )
+
+    train_dataset = TextualInversionDataset(
+        data_root=args.train_data_dir,
+        tokenizer=tokenizer,
+        size=args.resolution,
+        placeholder_token=args.placeholder_token,
+        repeats=args.repeats,
+        learnable_property=args.learnable_property,
+        center_crop=args.center_crop,
+        set="train",
+    )
+    train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=args.train_batch_size, shuffle=True)
+
+    # 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 * args.gradient_accumulation_steps,
+        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+    )
+
+    text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        text_encoder, optimizer, train_dataloader, lr_scheduler
+    )
+
+    # Move vae and unet to device
+    vae.to(accelerator.device)
+    unet.to(accelerator.device)
+
+    # Keep vae and unet in eval model as we don't train these
+    vae.eval()
+    unet.eval()
+
+    # 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("textual_inversion", 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 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}")
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
+    global_step = 0
+
+    for epoch in range(args.num_train_epochs):
+        text_encoder.train()
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(text_encoder):
+                # Convert images to latent space
+                latents = vae.encode(batch["pixel_values"]).latent_dist.sample().detach()
+                latents = latents * 0.18215
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn(latents.shape).to(latents.device)
+                bsz = latents.shape[0]
+                # Sample a random timestep for each image
+                timesteps = torch.randint(
+                    0, noise_scheduler.config.num_train_timesteps, (bsz,), device=latents.device
+                ).long()
+
+                # Add noise to the latents according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+
+                # Get the text embedding for conditioning
+                encoder_hidden_states = text_encoder(batch["input_ids"])[0]
+
+                # Predict the noise residual
+                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                loss = F.mse_loss(noise_pred, noise, reduction="none").mean([1, 2, 3]).mean()
+                accelerator.backward(loss)
+
+                # Zero out the gradients for all token embeddings except the newly added
+                # embeddings for the concept, as we only want to optimize the concept embeddings
+                if accelerator.num_processes > 1:
+                    grads = text_encoder.module.get_input_embeddings().weight.grad
+                else:
+                    grads = text_encoder.get_input_embeddings().weight.grad
+                # Get the index for tokens that we want to zero the grads for
+                index_grads_to_zero = torch.arange(len(tokenizer)) != placeholder_token_id
+                grads.data[index_grads_to_zero, :] = grads.data[index_grads_to_zero, :].fill_(0)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+                if global_step % args.save_steps == 0:
+                    save_progress(text_encoder, placeholder_token_id, accelerator, args)
+
+            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
+
+        accelerator.wait_for_everyone()
+
+    # Create the pipeline using using the trained modules and save it.
+    if accelerator.is_main_process:
+        pipeline = StableDiffusionPipeline(
+            text_encoder=accelerator.unwrap_model(text_encoder),
+            vae=vae,
+            unet=unet,
+            tokenizer=tokenizer,
+            scheduler=PNDMScheduler(
+                beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", skip_prk_steps=True
+            ),
+            safety_checker=StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker"),
+            feature_extractor=CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32"),
+        )
+        pipeline.save_pretrained(args.output_dir)
+        # Also save the newly trained embeddings
+        save_progress(text_encoder, placeholder_token_id, accelerator, args)
+
+        if args.push_to_hub:
+            repo.push_to_hub(commit_message="End of training", blocking=False, auto_lfs_prune=True)
+
+    accelerator.end_training()
+    torch.cuda.empty_cache()
+    gc.collect()
+
+
+if __name__ == "__main__":
+    main()