github/DB-GPT
1
0
Fork 0
mirror of https://github.com/csunny/DB-GPT.git synced 2025-09-01 09:06:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat:llm manage

Browse Source
This commit is contained in:
aries_ckt
2023-09-21 22:35:05 +08:00
parent ef2884d1ad
commit 96b7fb38b1
1 changed files with 49 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

86
pilot/server/llm_manage/api.py
View File

@@ -5,7 +5,7 @@ from fastapi import APIRouter
from pilot.component import ComponentType from pilot.component import ComponentType
from pilot.configs.config import Config from pilot.configs.config import Config
from pilot.model.cluster import WorkerStartupRequest from pilot.model.cluster import WorkerStartupRequest, WorkerManagerFactory
from pilot.openapi.api_view_model import Result from pilot.openapi.api_view_model import Result
from pilot.server.llm_manage.request.request import ModelResponse from pilot.server.llm_manage.request.request import ModelResponse
@@ -77,29 +77,35 @@ async def model_list():
@router.post("/v1/worker/model/stop") @router.post("/v1/worker/model/stop")
async def model_start(request: WorkerStartupRequest): async def model_stop(request: WorkerStartupRequest):
print(f"/v1/worker/model/stop:") print(f"/v1/worker/model/stop:")
try: try:
from pilot.model.cluster.controller.controller import BaseModelController from pilot.model.cluster.controller.controller import BaseModelController
controller = CFG.SYSTEM_APP.get_component( # controller = CFG.SYSTEM_APP.get_component(
ComponentType.MODEL_CONTROLLER, BaseModelController # ComponentType.MODEL_CONTROLLER, BaseModelController
) # )
instances = await controller.get_all_instances( # instances = await controller.get_all_instances(
model_name="WorkerManager@service", healthy_only=True # model_name="WorkerManager@service", healthy_only=True
) # )
request.params = {} worker_manager = CFG.SYSTEM_APP.get_component(
worker_instance = None ComponentType.WORKER_MANAGER_FACTORY, WorkerManagerFactory
for instance in instances: ).create()
if instance.host == request.host and instance.port == request.port: if not worker_manager:
from pilot.model.cluster import ModelRegistryClient
from pilot.model.cluster import RemoteWorkerManager
registry = ModelRegistryClient(f"http://{request.host}:{request.port}")
worker_manager = RemoteWorkerManager(registry)
return Result.succ(await worker_manager.model_shutdown(request))
if not worker_instance:
return Result.faild(code="E000X", msg=f"can not find worker manager") return Result.faild(code="E000X", msg=f"can not find worker manager")
request.params = {}
return Result.succ(await worker_manager.model_shutdown(request))
# worker_instance = None
# for instance in instances:
# if instance.host == request.host and instance.port == request.port:
# from pilot.model.cluster import ModelRegistryClient
# from pilot.model.cluster import RemoteWorkerManager
#
# registry = ModelRegistryClient(f"http://{request.host}:{request.port}")
# worker_manager = RemoteWorkerManager(registry)
# return Result.succ(await worker_manager.model_shutdown(request))
# if not worker_instance:
# return Result.faild(code="E000X", msg=f"can not find worker manager")
except Exception as e: except Exception as e:
return Result.faild(code="E000X", msg=f"model stop failed {e}") return Result.faild(code="E000X", msg=f"model stop failed {e}")
@@ -108,24 +114,30 @@ async def model_start(request: WorkerStartupRequest):
async def model_start(request: WorkerStartupRequest): async def model_start(request: WorkerStartupRequest):
print(f"/v1/worker/model/start:") print(f"/v1/worker/model/start:")
try: try:
from pilot.model.cluster.controller.controller import BaseModelController # from pilot.model.cluster.controller.controller import BaseModelController
#
controller = CFG.SYSTEM_APP.get_component( # controller = CFG.SYSTEM_APP.get_component(
ComponentType.MODEL_CONTROLLER, BaseModelController # ComponentType.MODEL_CONTROLLER, BaseModelController
) # )
instances = await controller.get_all_instances( # instances = await controller.get_all_instances(
model_name="WorkerManager@service", healthy_only=True # model_name="WorkerManager@service", healthy_only=True
) # )
worker_instance = None worker_manager = CFG.SYSTEM_APP.get_component(
for instance in instances: ComponentType.WORKER_MANAGER_FACTORY, WorkerManagerFactory
if instance.host == request.host and instance.port == request.port: ).create()
from pilot.model.cluster import ModelRegistryClient if not worker_manager:
from pilot.model.cluster import RemoteWorkerManager
registry = ModelRegistryClient(f"http://{request.host}:{request.port}")
worker_manager = RemoteWorkerManager(registry)
return Result.succ(await worker_manager.model_startup(request))
if not worker_instance:
return Result.faild(code="E000X", msg=f"can not find worker manager") return Result.faild(code="E000X", msg=f"can not find worker manager")
return Result.succ(await worker_manager.model_startup(request))
# worker_instance = None
# for instance in instances:
# if instance.host == request.host and instance.port == request.port:
# from pilot.model.cluster import ModelRegistryClient
# from pilot.model.cluster import RemoteWorkerManager
#
# registry = ModelRegistryClient(f"http://{request.host}:{request.port}")
# worker_manager = RemoteWorkerManager(registry)
# return Result.succ(await worker_manager.model_startup(request))
# if not worker_instance:
# return Result.faild(code="E000X", msg=f"can not find worker manager")
except Exception as e: except Exception as e:
return Result.faild(code="E000X", msg=f"model start failed {e}") return Result.faild(code="E000X", msg=f"model start failed {e}")