rasmus/burn-stablediffusion-vibecode
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use wgpu by default and ndarray for convert

Browse Source
This commit is contained in:
Gadersd
2023-08-08 15:32:21 -04:00
parent b87273c2be
commit 1830756917
5 changed files with 20 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
README.md
View File

@@ -20,18 +20,19 @@ Start by downloading the SDv1-4.bin model provided on HuggingFace.
wget https://huggingface.co/Gadersd/Stable-Diffusion-Burn/resolve/main/V1/SDv1-4.bin
```
Next, set the appropriate CUDA version. It may be possible to run the model using wgpu without the need for torch in the future using `cargo run --features wgpu-backend...` but currently wgpu doesn't support buffer sizes large enough for Stable Diffusion.
```bash
export TORCH_CUDA_VERSION=cu113
```
### Step 2: Run the Sample Binary
Invoke the sample binary provided in the rust code, as shown below:
Invoke the sample binary provided in the rust code. By default, wgpu is used which requires a gpu with at least 10 GB of VRAM (will be lower in the future), but torch can be used with the `torch-backend` feature and can run on a 6 GB gpu.
```bash
# wgpu (NEEDS >= 10 GB VRAM)
# Arguments: <model_type(burn or dump)> <model> <unconditional_guidance_scale> <n_diffusion_steps> <prompt> <output_image>
cargo run --release --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img
# torch (at least 6 GB VRAM, possibly less)
export TORCH_CUDA_VERSION=cu113
# Arguments: <model_type(burn or dump)> <model> <unconditional_guidance_scale> <n_diffusion_steps> <prompt> <output_image>
cargo run --release --features torch-backend --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img
```
This command will generate an image according to the provided prompt, which will be saved as 'img0.png'.