---
title: slam-sd1.5
canonical_url: "https://www.modelscope.cn/models/alimama-creative/slam-sd1.5"
md_url: "https://www.modelscope.cn/models/alimama-creative/slam-sd1.5.md"
repository: alimama-creative/slam-sd1.5
last_updated: 2024-09-24
license: apache-2.0
parameters: 1.1B
tensor_type:
  - F16
  - F32
library_name:
  - pytorch
  - lora
  - safetensors
  - diffusers
downloads: 2119
stars: 1
tags:
  - text-to-image
---

# slam-sd1.5

> slam-sd1.5 - alimama-creative 在 ModelScope 开源的模型。Sub-path Linear Approximation Model (SLAM): SD1.5 Paper: https://arxiv.org/abs/2404.13903 Project Page: https://subpath-linear-approx-model.github.io/ The checkpoint is a distilled from…

alimama-creative/slam-sd1.5 是 ModelScope 魔搭社区上的 1.1B 参数机器学习模型，采用 apache-2.0 许可。

- **Repository**: alimama-creative/slam-sd1.5
- **License**: apache-2.0
- **Parameters**: 1.1B
- **Tags**: text-to-image
- **Downloads**: 2119
- **Stars**: 1
- **Last updated**: 2024-09-24

Source: https://www.modelscope.cn/models/alimama-creative/slam-sd1.5

---

# Sub-path Linear Approximation Model (SLAM): SD1.5
Paper: [https://arxiv.org/abs/2404.13903](https://arxiv.org/abs/2404.13903)</br>
Project Page: [https://subpath-linear-approx-model.github.io/](https://subpath-linear-approx-model.github.io/)</br>
The checkpoint is a distilled from [runwayml/stable-diffusion-v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5) with our proposed Sub-path Linear Approximation Model, which reduces the number of inference steps to only between 2-4 steps.
## Usage
First, install the latest version of the Diffusers library as well as peft, accelerate and transformers.
```bash
pip install --upgrade pip
pip install --upgrade diffusers transformers accelerate peft
```

We implement SLAM to be compatible with [LCMScheduler](https://huggingface.co/docs/diffusers/v0.22.3/en/api/schedulers/lcm#diffusers.LCMScheduler). You can use SLAM just like you use LCM, with guidance_scale set to 1 constantly.
```python
from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained("alimama-creative/slam-sd1.5")

# To save GPU memory, torch.float16 can be used, but it may compromise image quality.
pipe.to(torch_device="cuda", torch_dtype=torch.float16)

prompt = "a painting of a majestic kingdom with towering castles, lush gardens, ice and snow world"

num_inference_steps = 2

images = pipe(prompt=prompt, num_inference_steps=num_inference_steps, guidance_scale=1, lcm_origin_steps=50, output_type="pil").images

```
![castle2_slam_step4.png](https://intranetproxy.alipay.com/skylark/lark/0/2024/png/102756509/1714305791356-5ba636a5-8435-4c90-84f3-f06163ebab51.png#clientId=uaea4a13b-3c46-4&from=ui&height=389&id=ue56980e1&originHeight=512&originWidth=512&originalType=binary&ratio=2&rotation=0&showTitle=false&size=518096&status=done&style=none&taskId=ue11b546c-420c-4d47-b87e-13084b19902&title=&width=389)