Upload 9 files

README.md

@@ -1,12 +1,13 @@
 ---
-title: Peft Sd Realfill
-emoji: 🏆
-colorFrom: pink
-colorTo: red
+title: Peft Lora Sd Dreambooth
+emoji: 🎨
+colorFrom: purple
+colorTo: green
 sdk: gradio
-sdk_version: 3.48.0
+sdk_version: 3.16.2
 app_file: app.py
 pinned: false
+license: openrail
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,275 @@
+#!/usr/bin/env python
+"""
+Demo showcasing parameter-efficient fine-tuning of Stable Dissfusion via Dreambooth leveraging 🤗 PEFT (https://github.com/huggingface/peft)
+
+The code in this repo is partly adapted from the following repositories:
+https://huggingface.co/spaces/hysts/LoRA-SD-training
+https://huggingface.co/spaces/multimodalart/dreambooth-training
+"""
+from __future__ import annotations
+
+import os
+import pathlib
+
+import gradio as gr
+import torch
+from typing import List
+
+from inference import InferencePipeline
+from trainer import Trainer
+from uploader import upload
+
+
+TITLE = "# RealFill Training and Inference Demo 🎨"
+DESCRIPTION = "Demo showcasing parameter-efficient fine-tuning of Stable Diffusion Inpainting via RealFill leveraging 🤗 PEFT (https://github.com/huggingface/peft)."
+
+
+ORIGINAL_SPACE_ID = "thuanz123/peft-sd-realfill"
+
+SPACE_ID = os.getenv("SPACE_ID", ORIGINAL_SPACE_ID)
+SHARED_UI_WARNING = f"""# Attention - This Space doesn't work in this shared UI. You can duplicate and use it with a paid private T4 GPU.
+<center><a class="duplicate-button" style="display:inline-block" target="_blank" href="https://huggingface.co/spaces/{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></center>
+"""
+if os.getenv("SYSTEM") == "spaces" and SPACE_ID != ORIGINAL_SPACE_ID:
+    SETTINGS = f'<a href="https://huggingface.co/spaces/{SPACE_ID}/settings">Settings</a>'
+
+else:
+    SETTINGS = "Settings"
+CUDA_NOT_AVAILABLE_WARNING = f"""# Attention - Running on CPU.
+<center>
+You can assign a GPU in the {SETTINGS} tab if you are running this on HF Spaces.
+"T4 small" is sufficient to run this demo.
+</center>
+"""
+
+
+def show_warning(warning_text: str) -> gr.Blocks:
+    with gr.Blocks() as demo:
+        with gr.Box():
+            gr.Markdown(warning_text)
+    return demo
+
+
+def update_output_files() -> dict:
+    paths = sorted(pathlib.Path("results").glob("*.pt"))
+    config_paths = sorted(pathlib.Path("results").glob("*.json"))
+    paths = paths + config_paths
+    paths = [path.as_posix() for path in paths]  # type: ignore
+    return gr.update(value=paths or None)
+
+
+def create_training_demo(trainer: Trainer, pipe: InferencePipeline) -> gr.Blocks:
+    with gr.Blocks() as demo:
+        base_model = gr.Dropdown(
+            choices=[
+                "runwayml/stable-diffusion-inpainting",
+                "stabilityai/stable-diffusion-2-inpainting",
+            ],
+            value="stabilityai/stable-diffusion-2-inpainting",
+            label="Base Model",
+            visible=True,
+        )
+        resolution = gr.Dropdown(choices=["512"], value="512", label="Resolution", visible=False)
+
+        with gr.Row():
+            with gr.Box():
+                gr.Markdown("Training Data")
+                ref_images = gr.Files(label="Reference images")
+                target_image = gr.Files(label="Target image")
+                target_mask = gr.Files(label="Target mask")
+                gr.Markdown(
+                    """
+                    - Upload reference images of the scene you are planning on training on.
+                    - For a concept prompt, use a unique, made up word to avoid collisions.
+                    - Guidelines for getting good results:
+                        - 1-5 images of the object from different angles
+                        - 2000 iterations should be good enough. 
+                        - LoRA Rank for unet - 8
+                        - LoRA Alpha for unet - 16
+                        - lora dropout - 0.1
+                        - LoRA Bias for unet - `none`
+                        - Uncheck `FP16` and `8bit-Adam` only if you have VRAM at least 32GB
+                        - Experiment with various values for lora dropouts, enabling/disabling fp16 and 8bit-Adam
+                    """
+                )
+            with gr.Box():
+                gr.Markdown("Training Parameters")
+                num_training_steps = gr.Number(label="Number of Training Steps", value=2000, precision=0)
+                unet_learning_rate = gr.Number(label="Unet Learning Rate", value=2e-4)
+                text_encoder_learning_rate = gr.Number(label="Text Encoder Learning Rate", value=4e-5)
+                gradient_checkpointing = gr.Checkbox(label="Whether to use gradient checkpointing", value=True)
+                lora_rank = gr.Number(label="LoRA Rank for unet", value=8, precision=0)
+                lora_alpha = gr.Number(
+                    label="LoRA Alpha for unet. scaling factor = lora_alpha/lora_r", value=16, precision=0
+                )
+                lora_dropout = gr.Number(label="lora dropout", value=0.1)
+                lora_bias = gr.Dropdown(
+                    choices=["none", "all", "lora_only"],
+                    value="none",
+                    label="LoRA Bias for unet. This enables bias params to be trainable based on the bias type",
+                    visible=True,
+                )
+                gradient_accumulation = gr.Number(label="Number of Gradient Accumulation", value=1, precision=0)
+                fp16 = gr.Checkbox(label="FP16", value=True)
+                use_8bit_adam = gr.Checkbox(label="Use 8bit Adam", value=True)
+                gr.Markdown(
+                    """
+                    - It will take about 40-60 minutes to train for 2000 steps with a T4 GPU.
+                    - You may want to try a small number of steps first, like 1, to see if everything works fine in your environment.
+                    - Note that your trained models will be deleted when the second training is started. You can upload your trained model in the "Upload" tab.
+                    """
+                )
+
+        run_button = gr.Button("Start Training")
+        with gr.Box():
+            with gr.Row():
+                check_status_button = gr.Button("Check Training Status")
+                with gr.Column():
+                    with gr.Box():
+                        gr.Markdown("Message")
+                        training_status = gr.Markdown()
+                    output_files = gr.Files(label="Trained Model Files")
+
+        run_button.click(fn=pipe.clear)
+
+        run_button.click(
+            fn=trainer.run,
+            inputs=[
+                base_model,
+                resolution,
+                num_training_steps,
+                ref_images,
+                target_image,
+                target_mask,
+                unet_learning_rate,
+                text_encoder_learning_rate,
+                gradient_accumulation,
+                fp16,
+                use_8bit_adam,
+                gradient_checkpointing,
+                lora_rank,
+                lora_alpha,
+                lora_bias,
+                lora_dropout,
+            ],
+            outputs=[
+                training_status,
+                output_files,
+            ],
+            queue=False,
+        )
+        check_status_button.click(fn=trainer.check_if_running, inputs=None, outputs=training_status, queue=False)
+        check_status_button.click(fn=update_output_files, inputs=None, outputs=output_files, queue=False)
+    return demo
+
+
+def find_model_files() -> list[str]:
+    curr_dir = pathlib.Path(__file__).parent
+    paths = sorted(curr_dir.glob('*'))
+    paths = [
+        path for path in paths
+        if (path / 'model_index.json').exists()
+    ]
+    return [path.relative_to(curr_dir).as_posix() for path in paths]
+
+
+def reload_realfill_model_list() -> dict:
+    return gr.update(choices=find_model_files())
+
+
+def create_inference_demo(pipe: InferencePipeline) -> gr.Blocks:
+    with gr.Blocks() as demo:
+        with gr.Row():
+            with gr.Column():
+                reload_button = gr.Button("Reload Model List")
+                realfill_model = gr.Dropdown(
+                    choices=find_model_files(), label="RealFill Model File"
+                )
+                target_image = gr.Files(label="Target image")
+                target_mask = gr.Files(label="Target mask")
+                seed = gr.Slider(label="Seed", minimum=0, maximum=100000, step=1, value=1)
+                with gr.Accordion("Other Parameters", open=False):
+                    num_steps = gr.Slider(label="Number of Steps", minimum=0, maximum=1000, step=1, value=50)
+                    guidance_scale = gr.Slider(label="CFG Scale", minimum=0, maximum=50, step=0.1, value=7)
+
+                run_button = gr.Button("Generate")
+
+                gr.Markdown(
+                    """
+                    - After training, you can press "Reload Model List" button to load your trained model names.
+                    """
+                )
+            with gr.Column():
+                result = gr.Image(label="Result")
+
+        reload_button.click(fn=reload_realfill_model_list, inputs=None, outputs=realfill_model)
+        run_button.click(
+            fn=pipe.run,
+            inputs=[
+                realfill_model,
+                target_image,
+                target_mask,
+                seed,
+                num_steps,
+                guidance_scale,
+            ],
+            outputs=result,
+            queue=False,
+        )
+        seed.change(
+            fn=pipe.run,
+            inputs=[
+                realfill_model,
+                target_image,
+                target_mask,
+                seed,
+                num_steps,
+                guidance_scale,
+            ],
+            outputs=result,
+            queue=False,
+        )
+    return demo
+
+
+def create_upload_demo() -> gr.Blocks:
+    with gr.Blocks() as demo:
+        model_name = gr.Textbox(label="Model Name")
+        hf_token = gr.Textbox(label="Hugging Face Token (with write permission)")
+        upload_button = gr.Button("Upload")
+        with gr.Box():
+            gr.Markdown("Message")
+            result = gr.Markdown()
+        gr.Markdown(
+            """
+            - You can upload your trained model to your private Model repo (i.e. https://huggingface.co/{your_username}/{model_name}).
+            - You can find your Hugging Face token [here](https://huggingface.co/settings/tokens).
+            """
+        )
+
+    upload_button.click(fn=upload, inputs=[model_name, hf_token], outputs=result)
+
+    return demo
+
+
+pipe = InferencePipeline()
+trainer = Trainer()
+
+with gr.Blocks(css="style.css") as demo:
+    if os.getenv("IS_SHARED_UI"):
+        show_warning(SHARED_UI_WARNING)
+    if not torch.cuda.is_available():
+        show_warning(CUDA_NOT_AVAILABLE_WARNING)
+
+    gr.Markdown(TITLE)
+    gr.Markdown(DESCRIPTION)
+
+    with gr.Tabs():
+        with gr.TabItem("Train"):
+            create_training_demo(trainer, pipe)
+        with gr.TabItem("Test"):
+            create_inference_demo(pipe)
+        with gr.TabItem("Upload"):
+            create_upload_demo()
+
+demo.queue(default_enabled=False).launch(share=False)

colab.py

@@ -0,0 +1,371 @@
+#!/usr/bin/env python
+"""
+Demo showcasing parameter-efficient fine-tuning of Stable Dissfusion via Dreambooth leveraging 🤗 PEFT (https://github.com/huggingface/peft)
+
+The code in this repo is partly adapted from the following repositories:
+https://huggingface.co/spaces/hysts/LoRA-SD-training
+https://huggingface.co/spaces/multimodalart/dreambooth-training
+"""
+from __future__ import annotations
+
+import os
+import pathlib
+
+import gradio as gr
+import torch
+from typing import List
+
+from inference import InferencePipeline
+from trainer import Trainer
+from uploader import upload
+
+
+TITLE = "# LoRA + Dreambooth Training and Inference Demo 🎨"
+DESCRIPTION = "Demo showcasing parameter-efficient fine-tuning of Stable Dissfusion via Dreambooth leveraging 🤗 PEFT (https://github.com/huggingface/peft)."
+
+
+ORIGINAL_SPACE_ID = "smangrul/peft-lora-sd-dreambooth"
+
+SPACE_ID = os.getenv("SPACE_ID", ORIGINAL_SPACE_ID)
+SHARED_UI_WARNING = f"""# Attention - This Space doesn't work in this shared UI. You can duplicate and use it with a paid private T4 GPU.
+<center><a class="duplicate-button" style="display:inline-block" target="_blank" href="https://huggingface.co/spaces/{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></center>
+"""
+if os.getenv("SYSTEM") == "spaces" and SPACE_ID != ORIGINAL_SPACE_ID:
+    SETTINGS = f'<a href="https://huggingface.co/spaces/{SPACE_ID}/settings">Settings</a>'
+
+else:
+    SETTINGS = "Settings"
+CUDA_NOT_AVAILABLE_WARNING = f"""# Attention - Running on CPU.
+<center>
+You can assign a GPU in the {SETTINGS} tab if you are running this on HF Spaces.
+"T4 small" is sufficient to run this demo.
+</center>
+"""
+
+
+def show_warning(warning_text: str) -> gr.Blocks:
+    with gr.Blocks() as demo:
+        with gr.Box():
+            gr.Markdown(warning_text)
+    return demo
+
+
+def update_output_files() -> dict:
+    paths = sorted(pathlib.Path("results").glob("*.pt"))
+    config_paths = sorted(pathlib.Path("results").glob("*.json"))
+    paths = paths + config_paths
+    paths = [path.as_posix() for path in paths]  # type: ignore
+    return gr.update(value=paths or None)
+
+
+def create_training_demo(trainer: Trainer, pipe: InferencePipeline) -> gr.Blocks:
+    with gr.Blocks() as demo:
+        base_model = gr.Dropdown(
+            choices=[
+                "CompVis/stable-diffusion-v1-4",
+                "runwayml/stable-diffusion-v1-5",
+                "stabilityai/stable-diffusion-2-1-base",
+            ],
+            value="runwayml/stable-diffusion-v1-5",
+            label="Base Model",
+            visible=True,
+        )
+        resolution = gr.Dropdown(choices=["512"], value="512", label="Resolution", visible=False)
+
+        with gr.Row():
+            with gr.Box():
+                gr.Markdown("Training Data")
+                concept_images = gr.Files(label="Images for your concept")
+                concept_prompt = gr.Textbox(label="Concept Prompt", max_lines=1)
+                gr.Markdown(
+                    """
+                    - Upload images of the style you are planning on training on.
+                    - For a concept prompt, use a unique, made up word to avoid collisions.
+                    - Guidelines for getting good results:
+                        - Dreambooth for an `object` or `style`:
+                            - 5-10 images of the object from different angles
+                            - 500-800 iterations should be good enough. 
+                            - Prior preservation is recommended.
+                            - `class_prompt`:
+                                - `a photo of object`
+                                - `style`
+                            - `concept_prompt`:
+                                - `<concept prompt> object`
+                                - `<concept prompt> style`
+                                - `a photo of <concept prompt> object`
+                                - `a photo of <concept prompt> style`
+                        - Dreambooth for a `Person/Face`:
+                            - 15-50 images of the person from different angles, lighting, and expressions. 
+                            Have considerable photos with close up faces.
+                            - 800-1200 iterations should be good enough.
+                            - good defaults for hyperparams
+                                - Model - `runwayml/stable-diffusion-v1-5` or `stabilityai/stable-diffusion-2-1-base`
+                                - Use/check Prior preservation.
+                                - Number of class images to use - 200
+                                - Prior Loss Weight - 1
+                                - LoRA Rank for unet - 16
+                                - LoRA Alpha for unet - 20
+                                - lora dropout - 0
+                                - LoRA Bias for unet - `all`
+                                - LoRA Rank for CLIP - 16
+                                - LoRA Alpha for CLIP - 17
+                                - LoRA Bias for CLIP - `all`
+                                - lora dropout for CLIP - 0
+                                - Uncheck `FP16` and `8bit-Adam` (don't use them for faces)
+                            - `class_prompt`: Use the gender related word of the person
+                                - `man`
+                                - `woman`
+                                - `boy`
+                                - `girl`
+                            - `concept_prompt`: just the unique, made up word, e.g., `srm`
+                            - Choose `all` for `lora_bias` and `text_encode_lora_bias`
+                        - Dreambooth for a `Scene`:
+                            - 15-50 images of the scene from different angles, lighting, and expressions.
+                            - 800-1200 iterations should be good enough.
+                            - Prior preservation is recommended.
+                            - `class_prompt`:
+                                - `scene`
+                                - `landscape`
+                                - `city`
+                                - `beach`
+                                - `mountain`
+                            - `concept_prompt`:
+                                - `<concept prompt> scene`
+                                - `<concept prompt> landscape`
+                        - Experiment with various values for lora dropouts, enabling/disabling fp16 and 8bit-Adam
+                    """
+                )
+            with gr.Box():
+                gr.Markdown("Training Parameters")
+                num_training_steps = gr.Number(label="Number of Training Steps", value=1000, precision=0)
+                learning_rate = gr.Number(label="Learning Rate", value=0.0001)
+                gradient_checkpointing = gr.Checkbox(label="Whether to use gradient checkpointing", value=True)
+                train_text_encoder = gr.Checkbox(label="Train Text Encoder", value=True)
+                with_prior_preservation = gr.Checkbox(label="Prior Preservation", value=True)
+                class_prompt = gr.Textbox(
+                    label="Class Prompt", max_lines=1, placeholder='Example: "a photo of object"'
+                )
+                num_class_images = gr.Number(label="Number of class images to use", value=50, precision=0)
+                prior_loss_weight = gr.Number(label="Prior Loss Weight", value=1.0, precision=1)
+                # use_lora = gr.Checkbox(label="Whether to use LoRA", value=True)
+                lora_r = gr.Number(label="LoRA Rank for unet", value=4, precision=0)
+                lora_alpha = gr.Number(
+                    label="LoRA Alpha for unet. scaling factor = lora_r/lora_alpha", value=4, precision=0
+                )
+                lora_dropout = gr.Number(label="lora dropout", value=0.00)
+                lora_bias = gr.Dropdown(
+                    choices=["none", "all", "lora_only"],
+                    value="none",
+                    label="LoRA Bias for unet. This enables bias params to be trainable based on the bias type",
+                    visible=True,
+                )
+                lora_text_encoder_r = gr.Number(label="LoRA Rank for CLIP", value=4, precision=0)
+                lora_text_encoder_alpha = gr.Number(
+                    label="LoRA Alpha for CLIP. scaling factor = lora_r/lora_alpha", value=4, precision=0
+                )
+                lora_text_encoder_dropout = gr.Number(label="lora dropout for CLIP", value=0.00)
+                lora_text_encoder_bias = gr.Dropdown(
+                    choices=["none", "all", "lora_only"],
+                    value="none",
+                    label="LoRA Bias for CLIP. This enables bias params to be trainable based on the bias type",
+                    visible=True,
+                )
+                gradient_accumulation = gr.Number(label="Number of Gradient Accumulation", value=1, precision=0)
+                fp16 = gr.Checkbox(label="FP16", value=True)
+                use_8bit_adam = gr.Checkbox(label="Use 8bit Adam", value=True)
+                gr.Markdown(
+                    """
+                    - It will take about 20-30 minutes to train for 1000 steps with a T4 GPU.
+                    - You may want to try a small number of steps first, like 1, to see if everything works fine in your environment.
+                    - Note that your trained models will be deleted when the second training is started. You can upload your trained model in the "Upload" tab.
+                    """
+                )
+
+        run_button = gr.Button("Start Training")
+        with gr.Box():
+            with gr.Row():
+                check_status_button = gr.Button("Check Training Status")
+                with gr.Column():
+                    with gr.Box():
+                        gr.Markdown("Message")
+                        training_status = gr.Markdown()
+                    output_files = gr.Files(label="Trained Weight Files and Configs")
+
+        run_button.click(fn=pipe.clear)
+
+        run_button.click(
+            fn=trainer.run,
+            inputs=[
+                base_model,
+                resolution,
+                num_training_steps,
+                concept_images,
+                concept_prompt,
+                learning_rate,
+                gradient_accumulation,
+                fp16,
+                use_8bit_adam,
+                gradient_checkpointing,
+                train_text_encoder,
+                with_prior_preservation,
+                prior_loss_weight,
+                class_prompt,
+                num_class_images,
+                lora_r,
+                lora_alpha,
+                lora_bias,
+                lora_dropout,
+                lora_text_encoder_r,
+                lora_text_encoder_alpha,
+                lora_text_encoder_bias,
+                lora_text_encoder_dropout,
+            ],
+            outputs=[
+                training_status,
+                output_files,
+            ],
+            queue=False,
+        )
+        check_status_button.click(fn=trainer.check_if_running, inputs=None, outputs=training_status, queue=False)
+        check_status_button.click(fn=update_output_files, inputs=None, outputs=output_files, queue=False)
+    return demo
+
+
+def find_weight_files() -> List[str]:
+    curr_dir = pathlib.Path(__file__).parent
+    paths = sorted(curr_dir.rglob("*.pt"))
+    return [path.relative_to(curr_dir).as_posix() for path in paths]
+
+
+def reload_lora_weight_list() -> dict:
+    return gr.update(choices=find_weight_files())
+
+
+def create_inference_demo(pipe: InferencePipeline) -> gr.Blocks:
+    with gr.Blocks() as demo:
+        with gr.Row():
+            with gr.Column():
+                base_model = gr.Dropdown(
+                    choices=[
+                        "CompVis/stable-diffusion-v1-4",
+                        "runwayml/stable-diffusion-v1-5",
+                        "stabilityai/stable-diffusion-2-1-base",
+                    ],
+                    value="runwayml/stable-diffusion-v1-5",
+                    label="Base Model",
+                    visible=True,
+                )
+                reload_button = gr.Button("Reload Weight List")
+                lora_weight_name = gr.Dropdown(
+                    choices=find_weight_files(), value="lora/lora_disney.pt", label="LoRA Weight File"
+                )
+                prompt = gr.Textbox(label="Prompt", max_lines=1, placeholder='Example: "style of sks, baby lion"')
+                negative_prompt = gr.Textbox(
+                    label="Negative Prompt", max_lines=1, placeholder='Example: "blurry, botched, low quality"'
+                )
+                seed = gr.Slider(label="Seed", minimum=0, maximum=100000, step=1, value=1)
+                with gr.Accordion("Other Parameters", open=False):
+                    num_steps = gr.Slider(label="Number of Steps", minimum=0, maximum=1000, step=1, value=50)
+                    guidance_scale = gr.Slider(label="CFG Scale", minimum=0, maximum=50, step=0.1, value=7)
+
+                run_button = gr.Button("Generate")
+
+                gr.Markdown(
+                    """
+                - After training, you can press "Reload Weight List" button to load your trained model names.
+                - Few repos to refer for ideas:
+                    - https://huggingface.co/smangrul/smangrul
+                    - https://huggingface.co/smangrul/painting-in-the-style-of-smangrul
+                    - https://huggingface.co/smangrul/erenyeager
+                """
+                )
+            with gr.Column():
+                result = gr.Image(label="Result")
+
+        reload_button.click(fn=reload_lora_weight_list, inputs=None, outputs=lora_weight_name)
+        prompt.submit(
+            fn=pipe.run,
+            inputs=[
+                base_model,
+                lora_weight_name,
+                prompt,
+                negative_prompt,
+                seed,
+                num_steps,
+                guidance_scale,
+            ],
+            outputs=result,
+            queue=False,
+        )
+        run_button.click(
+            fn=pipe.run,
+            inputs=[
+                base_model,
+                lora_weight_name,
+                prompt,
+                negative_prompt,
+                seed,
+                num_steps,
+                guidance_scale,
+            ],
+            outputs=result,
+            queue=False,
+        )
+        seed.change(
+            fn=pipe.run,
+            inputs=[
+                base_model,
+                lora_weight_name,
+                prompt,
+                negative_prompt,
+                seed,
+                num_steps,
+                guidance_scale,
+            ],
+            outputs=result,
+            queue=False,
+        )
+    return demo
+
+
+def create_upload_demo() -> gr.Blocks:
+    with gr.Blocks() as demo:
+        model_name = gr.Textbox(label="Model Name")
+        hf_token = gr.Textbox(label="Hugging Face Token (with write permission)")
+        upload_button = gr.Button("Upload")
+        with gr.Box():
+            gr.Markdown("Message")
+            result = gr.Markdown()
+        gr.Markdown(
+            """
+            - You can upload your trained model to your private Model repo (i.e. https://huggingface.co/{your_username}/{model_name}).
+            - You can find your Hugging Face token [here](https://huggingface.co/settings/tokens).
+            """
+        )
+
+    upload_button.click(fn=upload, inputs=[model_name, hf_token], outputs=result)
+
+    return demo
+
+
+pipe = InferencePipeline()
+trainer = Trainer()
+
+with gr.Blocks(css="style.css") as demo:
+    if os.getenv("IS_SHARED_UI"):
+        show_warning(SHARED_UI_WARNING)
+    if not torch.cuda.is_available():
+        show_warning(CUDA_NOT_AVAILABLE_WARNING)
+
+    gr.Markdown(TITLE)
+    gr.Markdown(DESCRIPTION)
+
+    with gr.Tabs():
+        with gr.TabItem("Train"):
+            create_training_demo(trainer, pipe)
+        with gr.TabItem("Test"):
+            create_inference_demo(pipe)
+        with gr.TabItem("Upload"):
+            create_upload_demo()
+
+demo.queue(default_enabled=False).launch(share=True)

inference.py

@@ -0,0 +1,63 @@
+from __future__ import annotations
+
+import gc
+import json
+import pathlib
+import sys
+
+import gradio as gr
+import PIL.Image
+import torch
+from diffusers import StableDiffusionInpaintingPipeline
+
+
+class InferencePipeline:
+    def __init__(self):
+        self.pipe = None
+        self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+
+    def clear(self) -> None:
+        del self.pipe
+        self.pipe = None
+        torch.cuda.empty_cache()
+        gc.collect()
+
+    def load_pipe(self, realfill_model: str) -> None:
+        pipe = StableDiffusionInpaintingPipeline.from_pretrained(
+            realfill_model, torch_dtype=torch.float16
+        ).to(self.device)
+        pipe = pipe.to(self.device)
+        self.pipe = pipe
+
+    def run(
+        self,
+        realfill_model: str,
+        target_image: PIL.Image,
+        target_mask: PIL.Image,
+        seed: int,
+        n_steps: int,
+        guidance_scale: float,
+    ) -> PIL.Image.Image:
+        if not torch.cuda.is_available():
+            raise gr.Error("CUDA is not available.")
+
+        self.load_pipe(realfill_model)
+
+        image = PIL.Image.open(target_image)
+        mask_image = PIL.Image.open(target_mask)
+
+        generator = torch.Generator(device=self.device).manual_seed(seed)
+        out = self.pipe(
+            "a photo of sks",
+            image=image, 
+            mask_image=mask_image, 
+            num_inference_steps=n_steps,
+            guidance_scale=guidance_scale,
+            generator=generator,
+        ).images[0]  # type: ignore
+
+        erode_kernel = PIL.ImageFilter.MaxFilter(3)
+        mask_image = mask_image.filter(erode_kernel)
+
+        result = PIL.Image.composite(result, out, mask_image)
+        return result

requirements.txt

@@ -0,0 +1,12 @@
+diffusers==0.20.1
+accelerate==0.23.0
+transformers==4.34.0
+peft==0.5.0
+torch==2.0.1
+torchvision==0.15.2
+ftfy==6.1.1
+tensorboard==2.14.0
+Jinja2==3.1.2
+Pillow==10.0.1
+bitsandbytes==0.41.1
+gradio==3.47.1

style.css

@@ -0,0 +1,3 @@
+h1 {
+    text-align: center;
+}

train_realfill.py

@@ -0,0 +1,952 @@
+import random
+import argparse
+import copy
+import itertools
+import logging
+import math
+import os
+import shutil
+from pathlib import Path
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+import torchvision.transforms.v2 as transforms_v2
+import transformers
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import 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, CLIPTextModel
+
+import diffusers
+from diffusers import (
+    AutoencoderKL,
+    DDPMScheduler,
+    StableDiffusionInpaintPipeline,
+    DPMSolverMultistepScheduler,
+    UNet2DConditionModel,
+)
+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
+
+from peft import PeftModel, LoraConfig, get_peft_model
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.20.1")
+
+logger = get_logger(__name__)
+
+def make_mask(images, resolution, times=30):
+    mask, times = torch.ones_like(images[0:1, :, :]), np.random.randint(1, times)
+    min_size, max_size, margin = np.array([0.03, 0.25, 0.01]) * resolution
+    max_size = min(max_size, resolution - margin * 2)
+
+    for _ in range(times):
+        width = np.random.randint(int(min_size), int(max_size))
+        height = np.random.randint(int(min_size), int(max_size))
+
+        x_start = np.random.randint(int(margin), resolution - int(margin) - width + 1)
+        y_start = np.random.randint(int(margin), resolution - int(margin) - height + 1)
+        mask[:, y_start:y_start + height, x_start:x_start + width] = 0
+
+    mask = 1 - mask if random.random() < 0.5 else mask
+    return mask
+
+def save_model_card(
+    repo_id: str,
+    images=None,
+    base_model=str,
+    repo_folder=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"""
+---
+license: creativeml-openrail-m
+base_model: {base_model}
+prompt: "a photo of sks"
+tags:
+- stable-diffusion-inpainting
+- stable-diffusion-inpainting-diffusers
+- text-to-image
+- diffusers
+- realfill
+inference: true
+---
+    """
+    model_card = f"""
+# RealFill - {repo_id}
+
+This is a realfill model derived from {base_model}. The weights were trained using [RealFill](https://realfill.github.io/).
+You can find some example images in the following. \n
+{img_str}
+"""
+    with open(os.path.join(repo_folder, "README.md"), "w") as f:
+        f.write(yaml + model_card)
+
+def log_validation(
+    text_encoder,
+    tokenizer,
+    unet,
+    args,
+    accelerator,
+    weight_dtype,
+    epoch,
+):
+    logger.info(
+        f"Running validation... \nGenerating {args.num_validation_images} images"
+    )
+
+    # create pipeline (note: unet and vae are loaded again in float32)
+    pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+        args.pretrained_model_name_or_path,
+        tokenizer=tokenizer,
+        revision=args.revision,
+        torch_dtype=weight_dtype,
+    )
+
+    # set `keep_fp32_wrapper` to True because we do not want to remove
+    # mixed precision hooks while we are still training
+    pipeline.unet = accelerator.unwrap_model(unet, keep_fp32_wrapper=True)
+    pipeline.text_encoder = accelerator.unwrap_model(text_encoder, keep_fp32_wrapper=True)
+    pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
+
+    pipeline = pipeline.to(accelerator.device)
+    pipeline.set_progress_bar_config(disable=True)
+
+    # run inference
+    generator = None if args.seed is None else torch.Generator(device=accelerator.device).manual_seed(args.seed)
+    
+    target_dir = Path(args.train_data_dir) / "target"
+    target_image, target_mask = target_dir / "target.png", target_dir / "mask.png"
+    image, mask_image = Image.open(target_image), Image.open(target_mask)
+
+    if image.mode != "RGB":
+        image = image.convert("RGB")
+
+    images = []
+    for _ in range(args.num_validation_images):
+        image = pipeline(
+            prompt="a photo of sks", image=image, mask_image=mask_image,
+            num_inference_steps=25, guidance_scale=5, generator=generator
+        ).images[0]
+        images.append(image)
+
+    for tracker in accelerator.trackers:
+        if tracker.name == "tensorboard":
+            np_images = np.stack([np.asarray(img) for img in images])
+            tracker.writer.add_images(f"validation", np_images, epoch, dataformats="NHWC")
+        if tracker.name == "wandb":
+            tracker.log(
+                {
+                    f"validation": [
+                        wandb.Image(image, caption=str(i)) for i, image in enumerate(images)
+                    ]
+                }
+            )
+
+    del pipeline
+    torch.cuda.empty_cache()
+
+    return images
+
+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(
+        "--revision",
+        type=str,
+        default=None,
+        required=False,
+        help="Revision of pretrained 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,
+        required=True,
+        help="A folder containing the training data of images.",
+    )
+    parser.add_argument(
+        "--num_validation_images",
+        type=int,
+        default=4,
+        help="Number of images that should be generated during validation with `validation_conditioning`.",
+    )
+    parser.add_argument(
+        "--validation_steps",
+        type=int,
+        default=100,
+        help=(
+            "Run realfill validation every X steps. RealFill validation consists of running the conditioning"
+            " `args.validation_conditioning` multiple times: `args.num_validation_images`."
+        ),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="realfill-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(
+        "--train_batch_size", type=int, default=4, help="Batch size (per device) for the training dataloader."
+    )
+    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(
+        "--unet_learning_rate",
+        type=float,
+        default=2e-4,
+        help="Learning rate to use for unet.",
+    )
+    parser.add_argument(
+        "--text_encoder_learning_rate",
+        type=float,
+        default=4e-5,
+        help="Learning rate to use for text encoder.",
+    )
+    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(
+        "--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(
+        "--wandb_key",
+        type=str,
+        default=None,
+        help=("If report to option is set to wandb, api-key for wandb used for login to wandb "),
+    )
+    parser.add_argument(
+        "--wandb_project_name",
+        type=str,
+        default=None,
+        help=("If report to option is set to wandb, project name in wandb for log tracking  "),
+    )
+    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(
+        "--set_grads_to_none",
+        action="store_true",
+        help=(
+            "Save more memory by using setting grads to None instead of zero. Be aware, that this changes certain"
+            " behaviors, so disable this argument if it causes any problems. More info:"
+            " https://pytorch.org/docs/stable/generated/torch.optim.Optimizer.zero_grad.html"
+        ),
+    )
+    parser.add_argument(
+        "--lora_rank",
+        type=int,
+        default=16,
+        help=("The dimension of the LoRA update matrices."),
+    )
+    parser.add_argument(
+        "--lora_alpha",
+        type=int,
+        default=27,
+        help=("The alpha constant of the LoRA update matrices."),
+    )
+    parser.add_argument(
+        "--lora_dropout",
+        type=float,
+        default=0.1,
+        help="The dropout rate of the LoRA update matrices.",
+    )
+    parser.add_argument(
+        "--lora_bias",
+        type=str,
+        default="none",
+        help="The bias type of the Lora update matrices. Must be 'none', 'all' or 'lora_only'.",
+    )
+
+    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
+
+    return args
+
+class RealFillDataset(Dataset):
+    """
+    A dataset to prepare the training and conditioning images and
+    the masks with the dummy prompt for fine-tuning the model.
+    It pre-processes the images, masks and tokenizes the prompts.
+    """
+
+    def __init__(
+        self,
+        train_data_root,
+        tokenizer,
+        size=512,
+    ):
+        self.size = size
+        self.tokenizer = tokenizer
+
+        self.ref_data_root = Path(train_data_root) / "ref"
+        self.target_image = Path(train_data_root) / "target" / "target.png"
+        self.target_mask = Path(train_data_root) / "target" / "mask.png"
+        if not (self.ref_data_root.exists() and self.target_image.exists() and self.target_mask.exists()):
+            raise ValueError("Train images root doesn't exists.")
+
+        self.train_images_path = list(self.ref_data_root.iterdir()) + [self.target_image]
+        self.num_train_images = len(self.train_images_path)
+        self.train_prompt = "a photo of sks"
+
+        self.image_transforms = transforms.Compose(
+            [
+                transforms_v2.RandomResize(size, int(1.125 * size)),
+                transforms.RandomCrop(size),
+                transforms.ToTensor(),
+                transforms.Normalize([0.5], [0.5]),
+            ]
+        )
+
+    def __len__(self):
+        return self.num_train_images
+
+    def __getitem__(self, index):
+        example = {}
+
+        image = Image.open(self.train_images_path[index])
+        image = exif_transpose(image)
+
+        if not image.mode == "RGB":
+            image = image.convert("RGB")
+        example["images"] = self.image_transforms(image)
+
+        if random.random() < 0.1:
+            example["masks"] = torch.ones_like(example["images"][0:1, :, :])
+        else:
+            example["masks"] = make_mask(example["images"], self.size)
+
+        if index < len(self) - 1:
+            example["weightings"] = torch.ones_like(example["masks"])
+        else:
+            weighting = Image.open(self.target_mask)
+            weighting = exif_transpose(weighting)
+            
+            weightings = self.image_transforms(weighting)
+            example["weightings"] = weightings < 0.5
+
+        example["conditioning_images"] = example["images"] * (example["masks"] < 0.5)
+
+        train_prompt = "" if random.random() < 0.1 else self.train_prompt
+        example["prompt_ids"] = self.tokenizer(
+            train_prompt,
+            truncation=True,
+            padding="max_length",
+            max_length=self.tokenizer.model_max_length,
+            return_tensors="pt",
+        ).input_ids
+
+        return example
+
+def collate_fn(examples):
+    input_ids = [example["prompt_ids"] for example in examples]
+    images = [example["images"] for example in examples]
+
+    masks = [example["masks"] for example in examples]
+    weightings = [example["weightings"] for example in examples]
+    conditioning_images = [example["conditioning_images"] for example in examples]
+
+    images = torch.stack(images)
+    images = images.to(memory_format=torch.contiguous_format).float()
+
+    masks = torch.stack(masks)
+    masks = masks.to(memory_format=torch.contiguous_format).float()
+
+    weightings = torch.stack(weightings)
+    weightings = weightings.to(memory_format=torch.contiguous_format).float()
+
+    conditioning_images = torch.stack(conditioning_images)
+    conditioning_images = conditioning_images.to(memory_format=torch.contiguous_format).float()
+
+    input_ids = torch.cat(input_ids, dim=0)
+
+    batch = {
+        "input_ids": input_ids,
+        "images": images,
+        "masks": masks,
+        "weightings": weightings,
+        "conditioning_images": conditioning_images,
+    }
+    return batch
+
+def main(args):
+    logging_dir = Path(args.output_dir, args.logging_dir)
+
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.report_to,
+        project_dir=logging_dir,
+    )
+
+    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
+
+        wandb.login(key=args.wandb_key)
+        wandb.init(project=args.wandb_project_name)
+
+    # 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)
+
+    # 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 tokenizer
+    if args.tokenizer_name:
+        tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_name, revision=args.revision, use_fast=False)
+    elif args.pretrained_model_name_or_path:
+        tokenizer = AutoTokenizer.from_pretrained(
+            args.pretrained_model_name_or_path,
+            subfolder="tokenizer",
+            revision=args.revision,
+            use_fast=False,
+        )
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    text_encoder = CLIPTextModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision
+    )
+    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae", revision=args.revision)
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision
+    )
+
+    config = LoraConfig(
+        r=args.lora_rank,
+        lora_alpha=args.lora_alpha,
+        target_modules=["to_k", "to_q", "to_v", "key", "query", "value"],
+        lora_dropout=args.lora_dropout,
+        bias=args.lora_bias,
+    )
+    unet = get_peft_model(unet, config)
+
+    config = LoraConfig(
+        r=args.lora_rank,
+        lora_alpha=args.lora_alpha,
+        target_modules=["k_proj", "q_proj", "v_proj"],
+        lora_dropout=args.lora_dropout,
+        bias=args.lora_bias,
+    )
+    text_encoder = get_peft_model(text_encoder, config)
+
+    vae.requires_grad_(False)
+
+    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()
+        text_encoder.gradient_checkpointing_enable()
+
+    # 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:
+            for model in models:
+                sub_dir = "unet" if isinstance(model.base_model.model, type(accelerator.unwrap_model(unet.base_model.model))) else "text_encoder"
+                model.save_pretrained(os.path.join(output_dir, sub_dir))
+
+                # make sure to pop weight so that corresponding model is not saved again
+                weights.pop()
+
+    def load_model_hook(models, input_dir):
+        while len(models) > 0:
+            # pop models so that they are not loaded again
+            model = models.pop()
+
+            sub_dir = "unet" if isinstance(model.base_model.model, type(accelerator.unwrap_model(unet.base_model.model))) else "text_encoder"
+            model_cls = UNet2DConditionModel if isinstance(model.base_model.model, type(accelerator.unwrap_model(unet.base_model.model))) else CLIPTextModel
+
+            load_model = model_cls.from_pretrained(args.pretrained_model_name_or_path, subfolder=sub_dir)
+            load_model = PeftModel.from_pretrained(load_model, input_dir, subfolder=sub_dir)
+
+            model.load_state_dict(load_model.state_dict())
+            del load_model
+
+    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.unet_learning_rate = (
+            args.unet_learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
+        )
+
+        args.text_encoder_learning_rate = (
+            args.text_encoder_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
+    optimizer = optimizer_class(
+        [
+            {"params": unet.parameters(), "lr": args.unet_learning_rate},
+            {"params": text_encoder.parameters(), "lr": args.text_encoder_learning_rate}
+        ],
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    # Dataset and DataLoaders creation:
+    train_dataset = RealFillDataset(
+        train_data_root=args.train_data_dir,
+        tokenizer=tokenizer,
+        size=args.resolution,
+    )
+
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset,
+        batch_size=args.train_batch_size,
+        shuffle=True,
+        collate_fn=collate_fn,
+        num_workers=1,
+    )
+
+    # 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,
+        num_cycles=args.lr_num_cycles,
+        power=args.lr_power,
+    )
+
+    # Prepare everything with our `accelerator`.
+    unet, text_encoder, optimizer, train_dataloader = accelerator.prepare(
+        unet, text_encoder, optimizer, train_dataloader
+    )
+
+    # 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 vae to device and cast to weight_dtype
+    vae.to(accelerator.device, dtype=weight_dtype)
+
+    # 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:
+        tracker_config = vars(copy.deepcopy(args))
+        accelerator.init_trackers("realfill", config=tracker_config)
+
+    # 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):
+        unet.train()
+        text_encoder.train()
+
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet, text_encoder):
+                # Convert images to latent space
+                latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
+                latents = latents * 0.18215
+
+                conditionings = vae.encode(batch["conditioning_images"].to(dtype=weight_dtype)).latent_dist.sample()
+                conditionings = conditionings * 0.18215
+
+                masks, size = batch["masks"].to(dtype=weight_dtype), latents.shape[2]
+                masks = F.interpolate(masks, size=size)
+
+                weightings = batch["weightings"].to(dtype=weight_dtype)
+                weightings = F.interpolate(weightings, size=size)
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents)
+                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
+                )
+                timesteps = timesteps.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)
+
+                # Concatenate noisy latents, masks and conditionings to get inputs to unet
+                inputs = torch.cat([noisy_latents, masks, conditionings], dim=1)
+
+                # Get the text embedding for conditioning
+                encoder_hidden_states = text_encoder(batch["input_ids"])[0]
+
+                # Predict the noise residual
+                model_pred = unet(inputs, timesteps, encoder_hidden_states).sample
+
+                # Compute the diffusion loss
+                assert noise_scheduler.config.prediction_type == "epsilon"
+                loss = (weightings * F.mse_loss(model_pred.float(), noise.float(), reduction="none")).mean()
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    params_to_clip = itertools.chain(
+                        unet.parameters(), text_encoder.parameters()
+                    )
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad(set_to_none=args.set_grads_to_none)
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                if args.report_to == "wandb":
+                    accelerator.print(progress_bar)
+                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}")
+
+                    if global_step % args.validation_steps == 0:
+                        log_validation(
+                            text_encoder,
+                            tokenizer,
+                            unet,
+                            args,
+                            accelerator,
+                            weight_dtype,
+                            global_step,
+                        )
+
+            logs = {"loss": loss.detach().item()}
+            progress_bar.set_postfix(**logs)
+            accelerator.log(logs, step=global_step)
+
+            if global_step >= args.max_train_steps:
+                break
+
+    # Save the lora layers
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            unet=accelerator.unwrap_model(unet.merge_and_unload(), keep_fp32_wrapper=True),
+            text_encoder=accelerator.unwrap_model(text_encoder.merge_and_unload(), keep_fp32_wrapper=True),
+            revision=args.revision,
+        )
+
+        pipeline.save_pretrained(args.output_dir)
+
+        # Final inference
+        images = log_validation(
+            text_encoder,
+            tokenizer,
+            unet,
+            args,
+            accelerator,
+            weight_dtype,
+            global_step,
+        )
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id,
+                images=images,
+                base_model=args.pretrained_model_name_or_path,
+                repo_folder=args.output_dir,
+            )
+            upload_folder(
+                repo_id=repo_id,
+                folder_path=args.output_dir,
+                commit_message="End of training",
+                ignore_patterns=["step_*", "epoch_*"],
+            )
+
+    accelerator.end_training()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

trainer.py

@@ -0,0 +1,145 @@
+from __future__ import annotations
+
+import os
+import pathlib
+import shlex
+import shutil
+import subprocess
+
+import gradio as gr
+import PIL.Image
+import torch
+
+
+def pad_image(image: PIL.Image.Image) -> PIL.Image.Image:
+    w, h = image.size
+    if w == h:
+        return image
+    elif w > h:
+        new_image = PIL.Image.new(image.mode, (w, w), (0, 0, 0))
+        new_image.paste(image, (0, (w - h) // 2))
+        return new_image
+    else:
+        new_image = PIL.Image.new(image.mode, (h, h), (0, 0, 0))
+        new_image.paste(image, ((h - w) // 2, 0))
+        return new_image
+
+
+class Trainer:
+    def __init__(self):
+        self.is_running = False
+        self.is_running_message = "Another training is in progress."
+
+        self.output_dir = pathlib.Path("results")
+        self.data_dir = pathlib.Path("data")
+        
+        self.ref_data_dir = self.data_dir / "ref"
+        self.target_data_dir = self.data_dir / "target"
+
+    def check_if_running(self) -> dict:
+        if self.is_running:
+            return gr.update(value=self.is_running_message)
+        else:
+            return gr.update(value="No training is running.")
+
+    def cleanup_dirs(self) -> None:
+        shutil.rmtree(self.output_dir, ignore_errors=True)
+
+    def prepare_dataset(
+        self, 
+        ref_images: list, 
+        target_image: PIL.Image, 
+        target_mask: PIL.Image, 
+        resolution: int
+    ) -> None:
+        self.ref_data_dir.mkdir(parents=True)
+        self.target_data_dir.mkdir(parents=True)
+        
+        for i, temp_path in enumerate(ref_images):
+            image = PIL.Image.open(temp_path.name)
+            image = pad_image(image)
+            image = image.resize((resolution, resolution))
+            image = image.convert("RGB")
+            out_path = self.ref_data_dir / f"{i:03d}.jpg"
+            image.save(out_path, format="JPEG", quality=100)
+
+        target_image.save(self.target_data_dir / "target.jpg", format="JPEG", quality=100)
+        target_mask.save(self.target_data_dir / "mask.jpg", format="JPEG", quality=100)
+
+    def run(
+        self,
+        base_model: str,
+        resolution_s: str,
+        n_steps: int,
+        ref_images: list | None,
+        target_image: PIL.Image,
+        target_mask: PIL.Image,
+        unet_learning_rate: float,
+        text_encoder_learning_rate: float,
+        gradient_accumulation: int,
+        fp16: bool,
+        use_8bit_adam: bool,
+        gradient_checkpointing: bool,
+        lora_rank: int,
+        lora_alpha: int,
+        lora_bias: str,
+        lora_dropout: float,
+    ) -> tuple[dict, list[pathlib.Path]]:
+        if not torch.cuda.is_available():
+            raise gr.Error("CUDA is not available.")
+
+        if self.is_running:
+            return gr.update(value=self.is_running_message), []
+
+        if ref_images is None:
+            raise gr.Error("You need to upload reference images.")
+        if target_image is None:
+            raise gr.Error("You need to upload target image.")
+        if target_mask is None:
+            raise gr.Error("You need to upload target mask.")
+
+        resolution = int(resolution_s)
+
+        self.cleanup_dirs()
+        self.prepare_dataset(ref_images, target_image, target_mask, resolution)
+
+        command = f"""
+        accelerate launch train_dreambooth.py \
+            --pretrained_model_name_or_path={base_model}  \
+            --train_data_dir={self.data_dir} \
+            --output_dir={self.output_dir} \
+            --resolution={resolution} \
+            --gradient_accumulation_steps={gradient_accumulation} \
+            --unet_learning_rate={unet_learning_rate} \
+            --text_encoder_learning_rate={text_encoder_learning_rate} \
+            --max_train_steps={n_steps} \
+            --train_batch_size=16 \
+            --lr_scheduler=constant \
+            --lr_warmup_steps=100 \
+            --lora_r={lora_rank} \
+            --lora_alpha={lora_alpha} \
+            --lora_bias={lora_bias} \
+            --lora_dropout={lora_dropout} \
+        """
+
+        if fp16:
+            command += " --mixed_precision fp16"
+        if use_8bit_adam:
+            command += " --use_8bit_adam"
+        if gradient_checkpointing:
+            command += " --gradient_checkpointing"
+
+        with open(self.output_dir / "train.sh", "w") as f:
+            command_s = " ".join(command.split())
+            f.write(command_s)
+
+        self.is_running = True
+        res = subprocess.run(shlex.split(command))
+        self.is_running = False
+
+        if res.returncode == 0:
+            result_message = "Training Completed!"
+        else:
+            result_message = "Training Failed!"
+        model_paths = sorted(self.output_dir.glob("*"))
+        return gr.update(value=result_message), model_paths

uploader.py

@@ -0,0 +1,17 @@
+import gradio as gr
+from huggingface_hub import HfApi
+
+
+def upload(model_name: str, hf_token: str) -> None:
+    api = HfApi(token=hf_token)
+    user_name = api.whoami()["name"]
+    model_id = f"{user_name}/{model_name}"
+    try:
+        api.create_repo(model_id, repo_type="model", private=True)
+        api.upload_folder(repo_id=model_id, folder_path="results", path_in_repo="results", repo_type="model")
+        url = f"https://huggingface.co/{model_id}"
+        message = f"Your model was successfully uploaded to [{url}]({url})."
+    except Exception as e:
+        message = str(e)
+
+    return gr.update(value=message, visible=True)