Add triton_models to whl package (#163)

* defer symlink

* fix lint

MANIFEST.in

@@ -3,3 +3,5 @@ include lmdeploy/lib/*.so
 include lmdeploy/lib/*.so*
 include lmdeploy/lib/*.dll
 include lmdeploy/lib/*.pyd
+include lmdeploy/serve/turbomind/service_docker_up.sh
+recursive-include lmdeploy/serve/turbomind/triton_models *

lmdeploy/serve/turbomind/deploy.py

@@ -497,6 +497,15 @@ def pack_model_repository(workspace_path: str):
     Args:
         workspace_path: the path of workspace
     """
+    os.symlink(src='../../tokenizer',
+               dst=osp.join(workspace_path, 'triton_models', 'preprocessing',
+                            '1', 'tokenizer'))
+    os.symlink(src='../../tokenizer',
+               dst=osp.join(workspace_path, 'triton_models', 'postprocessing',
+                            '1', 'tokenizer'))
+    os.symlink(src='../../weights',
+               dst=osp.join(workspace_path, 'triton_models', 'interactive',
+                            '1', 'weights'))
     model_repo_dir = osp.join(workspace_path, 'model_repository')
     os.makedirs(model_repo_dir, exist_ok=True)
     os.symlink(src=osp.join('../triton_models/interactive'),

lmdeploy/serve/turbomind/triton_models/interactive/1/weights

-../../weights
\ No newline at end of file

lmdeploy/serve/turbomind/triton_models/postprocessing/1/tokenizer

-../../tokenizer/
\ No newline at end of file

lmdeploy/serve/turbomind/triton_models/preprocessing/1/tokenizer

-../../tokenizer
\ No newline at end of file

setup.py

@@ -103,8 +103,7 @@ if __name__ == '__main__':
           long_description_content_type='text/markdown',
           author='OpenMMLab',
           author_email='openmmlab@gmail.com',
-          packages=find_packages(
-              exclude=('lmdeploy/serve/turbomind/triton_models', )),
+          packages=find_packages(exclude=()),
           include_package_data=True,
           install_requires=parse_requirements('requirements.txt'),
           classifiers=[