[core] add torch compile for diffusion (#684)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
This commit is contained in:
1 parent
cddcc400d3
commit
c784b3e07d
15 files changed
+134
-25
No files matched your search
@@ -14,6 +14,7 @@ from vllm.logger import init_logger
|
||||
from vllm.utils.mem_utils import DeviceMemoryProfiler, GiB_bytes
|
||||
|
||||
from vllm_omni.diffusion.cache.selector import get_cache_backend
|
||||
from vllm_omni.diffusion.compile import regionally_compile
|
||||
from vllm_omni.diffusion.data import (
|
||||
DiffusionOutput,
|
||||
OmniDiffusionConfig,
|
||||
@@ -100,6 +101,16 @@ class GPUWorker:
|
||||
)
|
||||
logger.info(f"Worker {self.rank}: Model loaded successfully.")
|
||||
|
||||
if not self.od_config.enforce_eager:
|
||||
try:
|
||||
self.pipeline.transformer = regionally_compile(
|
||||
self.pipeline.transformer,
|
||||
dynamic=True,
|
||||
)
|
||||
logger.info(f"Worker {self.rank}: Model compiled with torch.compile.")
|
||||
except Exception as e:
|
||||
logger.warning(f"Worker {self.rank}: torch.compile failed with error: {e}. Using eager mode.")
|
||||
|
||||
# Setup cache backend based on type (both backends use enable()/reset() interface)
|
||||
self.cache_backend = get_cache_backend(self.od_config.cache_backend, self.od_config.cache_config)
|
||||
|
||||
|
||||
Reference in new issue
Block a user