---
license: openrail++
---

This Repo contains a diffusers format version of the PixArt-Sigma Repos 
PixArt-alpha/pixart_sigma_sdxlvae_T5_diffusers
PixArt-alpha/PixArt-Sigma-XL-2-2K-MS
with the models loaded and saved in fp16 and bf16 formats, roughly halfing their sizes.
It can be used where download bandwith, memory or diskspace are relatively low, a T4 Colab instance for example.

To use in a diffusers script you currently(15/04/2024) need to use a Source distribution of Diffusers
and an extra 'patch' from the PixArt-Alpha's teams Sigma Github repo

**NOTE: This Model has been converted but not successfully tested, during the memory effecient attention
it generates 16Gb buffer, this appears break an MPS limitation, but it may also mean if requires more than 16Gb even
with the 16 bit model**

The diffusers script below assumes those with more memory on none MPS GPU's have more luck running it!

a Diffusers script looks like this, **currently (25th April 2024) you need will to install diffusers from source**.


```py
import random
import sys
import torch
from diffusers from PixArtSigmaPipeline

device = 'mps'
weight_dtype = torch.bfloat16

pipe = PixArtSigmaPipeline.from_pretrained(
    "Vargol/PixArt-Sigma_2k_16bit",
    torch_dtype=weight_dtype,
    variant="fp16",
    use_safetensors=True,
)

# Enable memory optimizations.
# pipe.enable_model_cpu_offload()
pipe.to(device)

prompt = "Cinematic science fiction film still.A cybernetic demon awaits her friend in a bar selling flaming oil drinks.  The barman is a huge tree being, towering over the demon"

for i in range(4):

    seed = random.randint(0, sys.maxsize)
    generator =  torch.Generator("mps").manual_seed(seed);

    image = pipe(prompt, generator=generator, num_iferencenum_inference_steps=40).images[0]
    image.save(f"pas_{seed}.png")a
~~

```