Date: 30.08.2025

Stable Diffusion v1.5 ROCm PyTorch Test

Requirments

• AMD Mi50/MI100 32Gb VRAM
• Workstation 40 GB RAM, 200GB SSD, 750W Power supply
• Ubuntu 22.04
• Python 3.10

My test environment: HP Z440 + AMD Mi50

TESTs

Preapre python environment for ROCm + PyTorch + BitsAndBytes:

Prepare Compilation BitsAndBytes for ROCm 6.2

• Check PyTorch installation

python3 -c "import torch; print(torch.__version__); print(torch.cuda.is_available()); print(torch.version.hip);print(torch.cuda.get_device_name(0));"

• Check BitsAndBytes installation

python -m bitsandbytes

Get the StableDiffusion 1.5

git lfs install
git clone https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5 sd1.5

Create script test_rocm_sd1.5.py:

from diffusers import StableDiffusionPipeline, UNet2DConditionModel
from transformers import CLIPTextModel
import torch

print("GPU available:", torch.cuda.is_available())
print("GPU name:", torch.cuda.get_device_name(0))

model_dir = "/home/sysadmin/llm/sd1.5"

text_encoder = CLIPTextModel.from_pretrained(
    model_dir,
    subfolder="text_encoder",
    torch_dtype=torch.float16
)

unet = UNet2DConditionModel.from_pretrained(
    model_dir,
    subfolder="unet",
    torch_dtype=torch.float16,
)

pipe = StableDiffusionPipeline.from_pretrained(
    model_dir,
    text_encoder=text_encoder,
    unet=unet,
    torch_dtype=torch.float16,
    safety_checker=None,
    feature_extractor=None,
    use_safetensors=True,
    local_files_only=True,
)

pipe = pipe.to("cuda")

out = pipe(
    prompt="ford focus 3 on parking, high quality, 8k",
    height=512, width=512, guidance_scale=9, num_inference_steps=80)
out.images[0].save("test.png", format="PNG")

Create script test_rocm_bnb_sd1.5.py:

from diffusers import StableDiffusionPipeline, UNet2DConditionModel
from transformers import CLIPTextModel
from diffusers import BitsAndBytesConfig as DF_BNB
from transformers import BitsAndBytesConfig as TF_BNB
import torch

print("GPU available:", torch.cuda.is_available())
print("GPU name:", torch.cuda.get_device_name(0))

model_dir = "/home/sysadmin/llm/sd1.5"

te_qconf = TF_BNB(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=False, 
    bnb_4bit_compute_dtype=torch.float16, 
)
text_encoder = CLIPTextModel.from_pretrained(
    model_dir,
    subfolder="text_encoder",
    quantization_config=te_qconf,
    torch_dtype=torch.float16,
)

unet_qconf = DF_BNB(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=False, 
    bnb_4bit_compute_dtype=torch.float16,
)
unet = UNet2DConditionModel.from_pretrained(
    model_dir,
    subfolder="unet",
    quantization_config=unet_qconf,
    torch_dtype=torch.float16,
)

pipe = StableDiffusionPipeline.from_pretrained(
    model_dir,
    text_encoder=text_encoder,
    unet=unet, 
    torch_dtype=torch.float16,
    safety_checker=None,
    feature_extractor=None,
    use_safetensors=True,
    local_files_only=True,
)

pipe = pipe.to("cuda")

out = pipe(
    prompt="ford focus 3 on parking, high quality, 8k",
    height=512, width=512, guidance_scale=9, num_inference_steps=80)
out.images[0].save("test_bnb.png", format="PNG")

Run test

Check rocm-smi during each test while true; do rocm-smi; sleep 1; done

• Without quantization

python ./test_rocm_sd1.5.py

• With quantization

python ./test_rocm_bnb_sd1.5.py