apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-30 03:22:05 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

qwen3_moe (#3084)

Browse Source
This commit is contained in:
bukejiyu
2025-08-06 14:45:27 +08:00
committed by GitHub
parent 91dc87f1c5
commit 20839abccf
30 changed files with 1361 additions and 1087 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
fastdeploy/model_executor/model_loader/__init__.py
View File

@@ -17,14 +17,16 @@
from fastdeploy.config import LoadChoices, LoadConfig
from fastdeploy.model_executor.model_loader.base_loader import BaseModelLoader
from fastdeploy.model_executor.model_loader.default_loader import DefaultModelLoader
from fastdeploy.model_executor.model_loader.new_loader import NewModelLoader
from fastdeploy.model_executor.model_loader.default_loader_v1 import (
DefaultModelLoaderV1,
)
def get_model_loader(load_config: LoadConfig) -> BaseModelLoader:
"""get_model_loader"""
if load_config.load_choices == LoadChoices.NEW_LOADER:
return NewModelLoader(load_config)
if load_config.load_choices == LoadChoices.DEFAULT_V1:
return DefaultModelLoaderV1(load_config)
return DefaultModelLoader(load_config)

7
fastdeploy/model_executor/model_loader/default_loader.py
View File

@@ -14,6 +14,8 @@
# limitations under the License.
"""
import contextlib
import paddle
from paddle import nn
from paddleformers.utils.log import logger
@@ -62,15 +64,16 @@ class DefaultModelLoader(BaseModelLoader):
self.clean_memory_fragments(state_dict)
def load_model(self, fd_config: FDConfig) -> nn.Layer:
context = paddle.LazyGuard()
architectures = fd_config.model_config.architectures[0]
logger.info(f"Starting to load model {architectures}")
if fd_config.load_config.dynamic_load_weight:
# register rl model
import fastdeploy.rl # noqa
architectures = architectures + "RL"
context = paddle.LazyGuard()
else:
context = contextlib.nullcontext()
with context:
model_cls = ModelRegistry.get_class(architectures)

14
fastdeploy/model_executor/model_loader/new_loader.py → fastdeploy/model_executor/model_loader/default_loader_v1.py
View File

@@ -14,6 +14,8 @@
# limitations under the License.
"""
import contextlib
import paddle
from paddle import nn
from paddleformers.utils.log import logger
@@ -29,7 +31,7 @@ from fastdeploy.model_executor.models.model_base import ModelRegistry
from fastdeploy.platforms import current_platform
class NewModelLoader(BaseModelLoader):
class DefaultModelLoaderV1(BaseModelLoader):
"""ModelLoader that can load registered models"""
def __init__(self, load_config: LoadConfig):
@@ -54,15 +56,19 @@ class NewModelLoader(BaseModelLoader):
def load_model(self, fd_config: FDConfig) -> nn.Layer:
architectures = fd_config.model_config.architectures[0]
logger.info(f"Starting to load model {architectures}")
if fd_config.load_config.dynamic_load_weight:
# register rl model
import fastdeploy.rl # noqa
architectures = architectures + "RL"
context = paddle.LazyGuard()
model_cls = ModelRegistry.get_class(architectures)
model = model_cls(fd_config)
else:
context = contextlib.nullcontext()
with context:
model_cls = ModelRegistry.get_class(architectures)
model = model_cls(fd_config)
model.eval()