feat: enhance model loading with prefetching and retry logic

- Introduced `from_pretrained_with_prefetch` to handle racy cache scenarios by re-prefetching and retrying on failures.
- Updated various model pipelines to utilize the new prefetching mechanism, ensuring robust loading of model components.
- Prefetch logic added to multiple models, including HunyuanVideo, LongCatImage, and QwenImage, to mitigate issues with incomplete cache states.
This commit is contained in:
tzhouam committed 2026-06-02 09:01:47 +00:00
1 parent b550709bba
commit 617c6ca55f
20 files changed
+678 -194

No files matched your search

@@ -27,6 +27,7 @@ from vllm_omni.diffusion.data import DiffusionOutput, OmniDiffusionConfig
from vllm_omni.diffusion.distributed.cfg_parallel import CFGParallelMixin
from vllm_omni.diffusion.distributed.utils import get_local_device
from vllm_omni.diffusion.model_loader.diffusers_loader import DiffusersPipelineLoader
from vllm_omni.diffusion.model_loader.hub_prefetch import from_pretrained_with_prefetch, prefetch_subfolders
from vllm_omni.diffusion.models.longcat_image.longcat_image_transformer import LongCatImageTransformer2DModel
from vllm_omni.diffusion.profiler.diffusion_pipeline_profiler import DiffusionPipelineProfilerMixin
from vllm_omni.diffusion.request import OmniDiffusionRequest
@@ -223,19 +224,32 @@ class LongCatImagePipeline(nn.Module, CFGParallelMixin, DiffusionPipelineProfile
model = od_config.model
local_files_only = os.path.exists(model)
# See ``hub_prefetch.py`` for the transformers v5 multi-worker subfolder
# race; prefetch the whole component set before any from_pretrained.
longcat_subfolders = ["scheduler", "text_encoder", "tokenizer", "vae"]
prefetch_subfolders(model, longcat_subfolders, local_files_only=local_files_only)
self.scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(
model, subfolder="scheduler", local_files_only=local_files_only
)
self.text_encoder = Qwen2_5_VLForConditionalGeneration.from_pretrained(
model, subfolder="text_encoder", local_files_only=local_files_only
self.text_encoder = from_pretrained_with_prefetch(
Qwen2_5_VLForConditionalGeneration.from_pretrained,
model,
subfolder="text_encoder",
prefetch_list=longcat_subfolders,
local_files_only=local_files_only,
)
self.text_processor = Qwen2VLProcessor.from_pretrained(
model, subfolder="tokenizer", local_files_only=local_files_only
)
self.vae = AutoencoderKL.from_pretrained(model, subfolder="vae", local_files_only=local_files_only).to(
self.device
)
self.vae = from_pretrained_with_prefetch(
AutoencoderKL.from_pretrained,
model,
subfolder="vae",
prefetch_list=longcat_subfolders,
local_files_only=local_files_only,
).to(self.device)
self.transformer = LongCatImageTransformer2DModel(od_config=od_config)
self.tokenizer = AutoTokenizer.from_pretrained(model, subfolder="tokenizer", local_files_only=local_files_only)