fix:test2

gpustack/routes/dashboard.py

@@ -17,6 +17,7 @@ from gpustack.schemas.dashboard import (
     SystemLoadSummary,
     SystemSummary,
     TimeSeriesData,
+    InstanceInfo,
 )
 from gpustack.schemas.model_usage import ModelUsage
 from gpustack.schemas.models import Model, ModelInstance
@@ -263,8 +264,27 @@ async def get_active_models(session: AsyncSession) -> List[ModelSummary]:
             ram=0,
             vram=0,
         )
+
+        instance_infos = []
         if result.id in model_instances_by_id:
             for model_instance in model_instances_by_id[result.id]:
+                vram = 0
+                for vram_e in (
+                    model_instance.computed_resource_claim.vram or {}
+                ).values():
+                    vram += vram_e
+                instance_infos.append(
+                    InstanceInfo(
+                        id=model_instance.id,
+                        name=model_instance.name,
+                        worker_name=model_instance.worker_name,
+                        worker_ip=model_instance.worker_ip,
+                        gpu_indexes=model_instance.gpu_indexes,
+                        gpu_addresses=model_instance.gpu_addresses,
+                        ram=model_instance.computed_resource_claim.ram or 0,
+                        vram=vram,
+                    )
+                )
                 aggregate_resource_claim(resource_claim, model_instance)
 
         model_summary.append(
@@ -274,6 +294,7 @@ async def get_active_models(session: AsyncSession) -> List[ModelSummary]:
                 categories=result.categories,
                 resource_claim=resource_claim,
                 instance_count=result.instance_count,
+                instance_infos=instance_infos,
                 token_count=(
                     result.total_token_count
                     if result.total_token_count is not None

gpustack/routes/gpu_devices.py

@@ -1,5 +1,9 @@
+from typing import List
+
 from fastapi import APIRouter
 from fastapi.responses import StreamingResponse
+
+from gpustack.schemas.models import ModelInstance
 from gpustack.server.deps import ListParamsDep, SessionDep, EngineDep
 from gpustack.schemas.gpu_devices import (
     GPUDevice,
@@ -28,13 +32,21 @@ async def get_gpus(
             media_type="text/event-stream",
         )
 
-    return await GPUDevice.paginated_by_query(
+    gpu_device_list = await GPUDevice.paginated_by_query(
         session=session,
         fuzzy_fields=fuzzy_fields,
         page=params.page,
         per_page=params.perPage,
     )
 
+    for gpu_device in gpu_device_list.items:
+        model_instances = await get_model_instances_by_gpu_and_worker(
+            session, gpu_device.index, gpu_device.worker_id
+        )
+        gpu_device.model_instances = model_instances
+
+    return gpu_device_list
+
 
 @router.get("/{id}", response_model=GPUDevicePublic)
 async def get_gpu(session: SessionDep, id: str):
@@ -43,3 +55,17 @@ async def get_gpu(session: SessionDep, id: str):
         raise NotFoundException(message="GPU device not found")
 
     return model
+
+
+async def get_model_instances_by_gpu_and_worker(
+    session: SessionDep, gpu_index: int, worker_id: int
+):
+    fields = {
+        "gpu_index": gpu_index,
+        "worker_id": worker_id,
+    }
+
+    model_instances: List[ModelInstance] = await ModelInstance.all_by_fields(
+        session, fields=fields
+    )
+    return model_instances

gpustack/routes/models.py

@@ -148,6 +148,7 @@ async def get_model_instances(
         )
 
     instances = model.instances
+
     count = len(instances)
     total_page = math.ceil(count / params.perPage)
     pagination = Pagination(

gpustack/routes/workers.py

@@ -1,5 +1,8 @@
+from typing import List
+
 from fastapi import APIRouter
 from fastapi.responses import StreamingResponse
+from gpustack.schemas.models import ModelInstance
 
 from gpustack.api.exceptions import (
     AlreadyExistsException,
@@ -44,7 +47,7 @@ async def get_workers(
             media_type="text/event-stream",
         )
 
-    return await Worker.paginated_by_query(
+    worker_list = await Worker.paginated_by_query(
         session=session,
         fields=fields,
         fuzzy_fields=fuzzy_fields,
@@ -52,6 +55,12 @@ async def get_workers(
         per_page=params.perPage,
     )
 
+    for worker in worker_list.items:
+        model_instances = await get_model_instances_by_worker(session, worker.id)
+        worker.model_instances = model_instances
+
+    return worker_list
+
 
 @router.get("/{id}", response_model=WorkerPublic)
 async def get_worker(session: SessionDep, id: int):
@@ -101,3 +110,14 @@ async def delete_worker(session: SessionDep, id: int):
         await WorkerService(session).delete(worker)
     except Exception as e:
         raise InternalServerErrorException(message=f"Failed to delete worker: {e}")
+
+
+async def get_model_instances_by_worker(session: SessionDep, worker_id: int):
+    fields = {
+        "worker_id": worker_id,
+    }
+
+    model_instances: List[ModelInstance] = await ModelInstance.all_by_fields(
+        session, fields=fields
+    )
+    return model_instances

gpustack/schemas/dashboard.py

@@ -48,6 +48,17 @@ class ResourceClaim(BaseModel):
     vram: int  # in bytes
 
 
+class InstanceInfo(BaseModel):
+    id: int
+    name: str
+    worker_name: Optional[str] = None
+    worker_ip: Optional[str] = None
+    gpu_indexes: Optional[List[int]]
+    gpu_addresses: Optional[List[str]]
+    ram: Optional[int] = 0
+    vram: Optional[int] = 0
+
+
 class ModelSummary(BaseModel):
     id: int
     name: str
@@ -55,6 +66,7 @@ class ModelSummary(BaseModel):
     instance_count: int
     token_count: int
     categories: Optional[List[str]] = None
+    instance_infos: Optional[List[InstanceInfo]] = None
 
 
 class ResourceCounts(BaseModel):

hack/install.sh

@@ -25,15 +25,30 @@ function download_ui() {
   local ui_path="${ROOT_DIR}/gpustack/ui"
   local tmp_ui_path="${ui_path}/tmp"
   local tag="latest"
+  local local_ui_dir="/home/gpustack-ui/dist"  # 本地 UI 构建目录
 
   if [[ "${GIT_VERSION}" != "v0.0.0" ]]; then
     tag="${GIT_VERSION}"
   fi
 
+  # 删除旧的 ui 目录
   rm -rf "${ui_path}"
-  mkdir -p "${tmp_ui_path}/ui"
+  mkdir -p "${ui_path}"
+
+  gpustack::log::info "Checking for local UI build at ${local_ui_dir}"
+
+  # 优先检查本地构建目录是否存在且非空
+  if [[ -d "${local_ui_dir}" ]] && [[ -n "$(ls -A ${local_ui_dir} 2>/dev/null)" ]]; then
+    gpustack::log::info "Local UI found at ${local_ui_dir}, copying..."
+    cp -a "${local_ui_dir}/." "${ui_path}"
+    gpustack::log::info "Local UI copied successfully."
+    return 0
+  else
+    gpustack::log::info "No valid local UI found at ${local_ui_dir}, proceeding with download..."
+  fi
 
-  gpustack::log::info "downloading '${tag}' UI assets"
+  # 如果本地没有，则下载远程 UI
+  mkdir -p "${tmp_ui_path}/ui"
 
   if ! curl --retry 3 --retry-connrefused --retry-delay 3 -sSfL "https://gpustack-ui-1303613262.cos.accelerate.myqcloud.com/releases/${tag}.tar.gz" 2>/dev/null |
     tar -xzf - --directory "${tmp_ui_path}/ui" 2>/dev/null; then
@@ -48,8 +63,11 @@ function download_ui() {
       gpustack::log::fatal "failed to download '${default_tag}' ui archive"
     fi
   fi
+
+  # 复制解压后的内容
   cp -a "${tmp_ui_path}/ui/dist/." "${ui_path}"
 
+  # 清理临时目录
   rm -rf "${tmp_ui_path}"
 }
 

poetry.lock

@@ -10685,4 +10685,4 @@ vllm = ["bitsandbytes", "mistral_common", "timm", "vllm"]
 [metadata]
 lock-version = "2.0"
 python-versions = ">=3.10,<3.13"
-content-hash = "3b91cdb95388d8479acd383dfb3e585de2fce906b70709dff4a6a01e4d2819a7"
+content-hash = "4462e07255a2ae6b0f3af20a305c4044dade8edffbdf8e5dd605cea8a03d5d96"

pyproject.toml

@@ -86,6 +86,12 @@ twine = "^5.1.1"
 mike = "^2.1.3"
 mkdocs-glightbox = "^0.4.0"
 
+
+[[tool.poetry.source]]
+name = "tuna"
+url = "https://pypi.tuna.tsinghua.edu.cn/simple/"
+priority = "explicit"
+
 [build-system]
 requires = ["poetry-core"]
 build-backend = "poetry.core.masonry.api"