modelscope-scepter/scepter/methods/studio/tuner_manager

frameworks, license, tasks

frameworks	license	tasks
Pytorch	apache-2.0	efficient-diffusion-tuning

{MODEL_NAME}

Model Introduction

{MODEL_DESCRIPTION}

Model Parameters

Base Model	Tuner Type	Training Parameters
		Batch Size	Epochs	Learning Rate	Resolution
{BASE_MODEL}	{TUNER_TYPE}	{TRAIN_BATCH_SIZE}	{TRAIN_EPOCH}	{LEARNING_RATE}	[{HEIGHT}, {WIDTH}]

Data Type	Data Space	Data Name	Data Subset
{DATA_TYPE}	{MS_DATA_SPACE}	{MS_DATA_NAME}	{MS_DATA_SUBNAME}

Model Performance

Given the input "{EVAL_PROMPT}," the following image may be generated:

Model Usage

Command Line Execution

• Run using Scepter's SDK, taking care to use different configuration files in accordance with the different base models, as per the corresponding relationships shown below

Base Model	LORA	SCE	TEXT_LORA	TEXT_SCE
	SD1.5	lora_cfg	sce_cfg	text_lora_cfg	text_sce_cfg
SD2.1		lora_cfg	sce_cfg	text_lora_cfg	text_sce_cfg
		SDXL	lora_cfg	sce_cfg	text_lora_cfg	text_sce_cfg

• Running from Source Code

git clone https://github.com/modelscope/scepter.git
cd scepter
pip install -r requirements/recommended.txt
PYTHONPATH=. python scepter/tools/run_inference.py
  --pretrained_model {this model folder}
  --cfg {lora_cfg} or {sce_cfg} or {text_lora_cfg} or {text_sce_cfg}
  --prompt '{EVAL_PROMPT}'
  --save_folder 'inference'

• Running after Installing Scepter (Recommended)

pip install scepter
python -m scepter/tools/run_inference.py
  --pretrained_model {this model folder}
  --cfg {lora_cfg} or {sce_cfg} or {text_lora_cfg} or {text_sce_cfg}
  --prompt '{EVAL_PROMPT}'
  --save_folder 'inference'

Running with Scepter Studio

pip install scepter
# Launch Scepter Studio
python -m scepter.tools.webui

• Refer to the following guides for model usage.

(video url)

Model Reference

If you wish to use this model for your own purposes, please cite it as follows.

@misc{{MODEL_NAME},
    title = {{MODEL_NAME}, {MODEL_URL}},
    author = {{USER_NAME}},
    year = {2024}
}

This model was trained using Scepter Studio; Scepter is an algorithm framework and toolbox developed by the Alibaba Tongyi Wanxiang Team. It provides a suite of tools and models for image generation, editing, fine-tuning, data processing, and more. If you find our work beneficial for your research, please cite as follows.

@misc{scepter,
    title = {SCEPTER, https://github.com/modelscope/scepter},
    author = {SCEPTER},
    year = {2023}
}