File size: 2,440 Bytes
57b0265 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 |
---
library_name: diffusers
base_model: segmind/SSD-1B
tags:
- lora
- text-to-image
license: openrail++
inference: false
---
# Latent Consistency Model (LCM) LoRA: SSD-Tiny
Latent Consistency Model (LCM) LoRA was proposed in [LCM-LoRA: A universal Stable-Diffusion Acceleration Module](https://arxiv.org/abs/2311.05556)
by *Simian Luo, Yiqin Tan, Suraj Patil, Daniel Gu et al.*
It is a distilled consistency adapter for [`segmind/SSD-Tiny`]("https://huggingface.co/segmind/SSD-1B") that allows
to reduce the number of inference steps to only between **2 - 8 steps**.
| Model | Params / M |
|----------------------------------------------------------------------------|------------|
| [lcm-lora-sdv1-5](https://huggingface.co/latent-consistency/lcm-lora-sdv1-5) | 67.5 |
| [**lcm-lora-ssd-tiny**](https://huggingface.co/segmind/lcm-lora-ssd-tiny) | **62.7** |
| [lcm-lora-sdxl](https://huggingface.co/latent-consistency/lcm-lora-sdxl) | 197 |
## Usage
LCM-LoRA is supported in 🤗 Hugging Face Diffusers library from version v0.23.0 onwards. To run the model, first
install the latest version of the Diffusers library as well as `peft`, `accelerate` and `transformers`.
audio dataset from the Hugging Face Hub:
```bash
pip install --upgrade pip
pip install --upgrade diffusers transformers accelerate peft
```
### Text-to-Image
Let's load the base model `segmind/SSD-Tiny` first. Next, the scheduler needs to be changed to [`LCMScheduler`](https://huggingface.co/docs/diffusers/v0.22.3/en/api/schedulers/lcm#diffusers.LCMScheduler) and we can reduce the number of inference steps to just 2 to 8 steps.
Please make sure to either disable `guidance_scale` or use values between 1.0 and 2.0.
```python
import torch
from diffusers import LCMScheduler, AutoPipelineForText2Image
model_id = "segmind/SSD-Tiny"
adapter_id = "segmind/lcm-lora-ssd-tiny"
pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch.float16, variant="fp16")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
pipe.to("cuda")
# load and fuse lcm lora
pipe.load_lora_weights(adapter_id)
pipe.fuse_lora()
prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"
# disable guidance_scale by passing 0
image = pipe(prompt=prompt, num_inference_steps=4, guidance_scale=0).images[0]
```
![SSD-Tiny LCM LoRA Image]() |