[Hardware] Support platforms and plugin system (#774)

Signed-off-by: gcanlin <canlinguosdu@gmail.com>
This commit is contained in:
Canlin Guo authored and GitHub committed 2026-01-28 22:12:52 +08:00
1 parent b11d436e19
commit c9392dff01
103 files changed
+1770 -1218

No files matched your search

@@ -96,8 +96,12 @@ Key arguments:
- `--num_inference_steps`: diffusion sampling steps (more steps = higher quality, slower).
- `--height/--width`: output resolution (defaults 1024x1024).
- `--output`: path to save the generated PNG.
- `--vae_use_slicing`: enable VAE slicing for memory optimization.
- `--vae_use_tiling`: enable VAE tiling for memory optimization.
- `--enable-cpu-offload`: enable CPU offloading for diffusion models.
> ℹ️ If you encounter OOM errors, try using `--vae_use_slicing` and `--vae_use_tiling` to reduce memory usage.
> ℹ️ Qwen-Image currently publishes best-effort presets at `1328x1328`, `1664x928`, `928x1664`, `1472x1140`, `1140x1472`, `1584x1056`, and `1056x1584`. Adjust `--height/--width` accordingly for the most reliable outcomes.
## Web UI Demo
@@ -7,7 +7,7 @@ import torch
from vllm_omni.entrypoints.omni import Omni
from vllm_omni.inputs.data import OmniDiffusionSamplingParams
from vllm_omni.outputs import OmniRequestOutput
from vllm_omni.utils.platform_utils import detect_device_type, is_npu
from vllm_omni.platforms import current_omni_platform
ASPECT_RATIOS: dict[str, tuple[int, int]] = {
"1:1": (1328, 1328),
@@ -62,8 +62,8 @@ def parse_args() -> argparse.Namespace:
@lru_cache(maxsize=1)
def get_omni(model_name: str) -> Omni:
# Enable VAE memory optimizations on NPU
vae_use_slicing = is_npu()
vae_use_tiling = is_npu()
vae_use_slicing = current_omni_platform.is_npu()
vae_use_tiling = current_omni_platform.is_npu()
return Omni(
model=model_name,
vae_use_slicing=vae_use_slicing,
@@ -72,7 +72,6 @@ def get_omni(model_name: str) -> Omni:
def build_demo(args: argparse.Namespace) -> gr.Blocks:
device = detect_device_type()
omni = get_omni(args.model)
def run_inference(
@@ -99,7 +98,7 @@ def build_demo(args: argparse.Namespace) -> gr.Blocks:
raise gr.Error("Inference steps must be a positive integer.")
if num_images not in {1, 2, 3, 4}:
raise gr.Error("Number of images must be 1, 2, 3, or 4.")
generator = torch.Generator(device=device).manual_seed(seed)
generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(seed)
outputs = omni.generate(
prompt.strip(),
OmniDiffusionSamplingParams(
@@ -12,7 +12,7 @@ from vllm_omni.diffusion.data import DiffusionParallelConfig, logger
from vllm_omni.entrypoints.omni import Omni
from vllm_omni.inputs.data import OmniDiffusionSamplingParams
from vllm_omni.outputs import OmniRequestOutput
from vllm_omni.utils.platform_utils import detect_device_type, is_npu
from vllm_omni.platforms import current_omni_platform
def parse_args() -> argparse.Namespace:
@@ -113,17 +113,22 @@ def parse_args() -> argparse.Namespace:
default=1,
help="Number of GPUs used for tensor parallelism (TP) inside the DiT.",
)
parser.add_argument(
"--vae_use_slicing",
action="store_true",
help="Enable VAE slicing for memory optimization.",
)
parser.add_argument(
"--vae_use_tiling",
action="store_true",
help="Enable VAE tiling for memory optimization.",
)
return parser.parse_args()
def main():
args = parse_args()
device = detect_device_type()
generator = torch.Generator(device=device).manual_seed(args.seed)
# Enable VAE memory optimizations on NPU
vae_use_slicing = is_npu()
vae_use_tiling = is_npu()
generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(args.seed)
# Configure cache based on backend type
cache_config = None
@@ -167,8 +172,8 @@ def main():
omni = Omni(
model=args.model,
vae_use_slicing=vae_use_slicing,
vae_use_tiling=vae_use_tiling,
vae_use_slicing=args.vae_use_slicing,
vae_use_tiling=args.vae_use_tiling,
cache_backend=args.cache_backend,
cache_config=cache_config,
enable_cache_dit_summary=args.enable_cache_dit_summary,