Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add command line flag to specify device for inference #25

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Add command line flag to specify device for inference #25

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
acf52f0
Add support for specifying inference device
hvaara Nov 25, 2024
bf5fa1e
Fix lint and update error message
hvaara Nov 25, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 23 additions & 18 deletions inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,28 +29,26 @@
MAX_NUM_FRAMES = 257


def load_vae(vae_dir):
def load_vae(vae_dir, device):
vae_ckpt_path = vae_dir / "vae_diffusion_pytorch_model.safetensors"
vae_config_path = vae_dir / "config.json"
with open(vae_config_path, "r") as f:
vae_config = json.load(f)
vae = CausalVideoAutoencoder.from_config(vae_config)
vae_state_dict = safetensors.torch.load_file(vae_ckpt_path)
vae.load_state_dict(vae_state_dict)
if torch.cuda.is_available():
vae = vae.cuda()
vae = vae.to(device=device)
return vae.to(torch.bfloat16)


def load_unet(unet_dir):
def load_unet(unet_dir, device):
unet_ckpt_path = unet_dir / "unet_diffusion_pytorch_model.safetensors"
unet_config_path = unet_dir / "config.json"
transformer_config = Transformer3DModel.load_config(unet_config_path)
transformer = Transformer3DModel.from_config(transformer_config)
unet_state_dict = safetensors.torch.load_file(unet_ckpt_path)
transformer.load_state_dict(unet_state_dict, strict=True)
if torch.cuda.is_available():
transformer = transformer.cuda()
transformer = transformer.to(device=device)
return transformer


Expand Down Expand Up @@ -89,7 +87,6 @@ def load_image_to_tensor_with_resize_and_crop(
def calculate_padding(
source_height: int, source_width: int, target_height: int, target_width: int
) -> tuple[int, int, int, int]:

# Calculate total padding needed
pad_height = target_height - source_height
pad_width = target_width - source_width
Expand Down Expand Up @@ -157,8 +154,6 @@ def seed_everething(seed: int):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed(seed)


def main():
Expand Down Expand Up @@ -233,6 +228,13 @@ def main():
help="Denoise in bfloat16",
)

parser.add_argument(
"--device",
type=str,
default="cuda",
help="Device used for inference",
)

# Prompts
parser.add_argument(
"--prompt",
Expand All @@ -254,6 +256,13 @@ def main():

seed_everething(args.seed)

device = args.device
if device == "cuda" and not torch.cuda.is_available():
logger.warning(
"Device 'cuda' is not available. Falling back to 'cpu'. Use --device to specify device."
)
device = "cpu"

output_dir = (
Path(args.output_path)
if args.output_path
Expand Down Expand Up @@ -303,15 +312,14 @@ def main():
scheduler_dir = ckpt_dir / "scheduler"

# Load models
vae = load_vae(vae_dir)
unet = load_unet(unet_dir)
vae = load_vae(vae_dir, device)
unet = load_unet(unet_dir, device)
scheduler = load_scheduler(scheduler_dir)
patchifier = SymmetricPatchifier(patch_size=1)
text_encoder = T5EncoderModel.from_pretrained(
"PixArt-alpha/PixArt-XL-2-1024-MS", subfolder="text_encoder"
)
if torch.cuda.is_available():
text_encoder = text_encoder.to("cuda")
text_encoder = text_encoder.to(device=device)
tokenizer = T5Tokenizer.from_pretrained(
"PixArt-alpha/PixArt-XL-2-1024-MS", subfolder="tokenizer"
)
Expand All @@ -330,8 +338,7 @@ def main():
}

pipeline = LTXVideoPipeline(**submodel_dict)
if torch.cuda.is_available():
pipeline = pipeline.to("cuda")
pipeline = pipeline.to(device=device)

# Prepare input for the pipeline
sample = {
Expand All @@ -342,9 +349,7 @@ def main():
"media_items": media_items,
}

generator = torch.Generator(
device="cuda" if torch.cuda.is_available() else "cpu"
).manual_seed(args.seed)
generator = torch.Generator(device=device).manual_seed(args.seed)

images = pipeline(
num_inference_steps=args.num_inference_steps,
Expand Down