Follow node vNPU mode for pods without vnpu-mode annotation

README.md

@@ -125,7 +125,9 @@ To exclusively use an entire card or request multiple cards, you only need to se
 
 ### Usage in HAMi
 
-**How HAMi chooses soft vs legacy vNPU:** The device plugin applies **soft slicing** (`libvnpu` / `hami-vnpu-core` mounts and environment) **only** when the Pod sets `huawei.com/vnpu-mode: hami-core`. Pods **without** this annotation still follow the **original vNPU** path (virtualization templates and `ASCEND_VNPU_SPECS`). These two paths are different. If your cluster effectively has **only** soft-slicing–oriented Ascend capacity (for example every node is configured for `hami-vnpu-core` and workloads are expected to use soft slicing), Pods that **omit** `vnpu-mode=hami-core` may remain **Pending** because they still request the legacy vNPU allocation model, which may not match what those nodes expose or how the scheduler pairs Pods to nodes.
+**How HAMi chooses soft vs legacy vNPU:**
+- A Pod that sets `huawei.com/vnpu-mode: hami-core` always uses **soft slicing** and is scheduled only onto hami-core-capable nodes.
+- A Pod that **omits** the annotation is **mode-agnostic**: it may be scheduled onto either a template node or a hami-core node, and its effective mode **follows the node**. On a hami-core node the device plugin applies soft slicing (`libvnpu` / `hami-vnpu-core` mounts and environment); on a template node it uses the original vNPU path (`ASCEND_VNPU_SPECS`). The plugin resolves this from the node's `hami-vnpu-core` setting (`IsHamiVnpuCore()`), so annotation-less Pods no longer remain Pending on hami-core-only clusters.
 
 ```yaml
 ...

internal/server/allocate.go

@@ -66,8 +66,9 @@ func (ps *PluginServer) buildContainerAllocateResponse(pod *v1.Pod, ctrName stri
 	resp.Envs["ASCEND_VISIBLE_DEVICES"] = ascendVisibleDevices
 
 	vnpuMode := pod.Annotations[VNPUModeAnnotation]
-	klog.V(4).Infof("Pod %s vnpu mode: %s", pod.Name, vnpuMode)
-	if vnpuMode == VNPUModeHamiCore {
+	effectiveHamiCore := vnpuMode == VNPUModeHamiCore || (vnpuMode == "" && ps.mgr.IsHamiVnpuCore())
+	klog.V(4).Infof("Pod %s vnpu mode: %q, effectiveHamiCore: %v", pod.Name, vnpuMode, effectiveHamiCore)
+	if effectiveHamiCore {
 		// 1. Handle volume mount injection
 		var mounts []*v1beta1.Mount
 		// A.Huawei driver and SMI toolchain (Read-Only)

internal/server/util_test.go

@@ -422,6 +422,79 @@ func TestBuildContainerAllocateResponse(t *testing.T) {
 				},
 			},
 		},
+		{
+			name: "NoAnnotation_NodeHamiCore_FollowsNode",
+			setup: func() (*PluginServer, CleanupFunc) {
+				return &PluginServer{
+					mgr: &FakeManager{
+						GetDeviceByUUIDFunc: func(uuid string) *manager.Device {
+							return &manager.Device{UUID: "uuid1", PhyID: 3}
+						},
+						IsHamiVnpuCoreFunc: func() bool { return true },
+					},
+					allocAnno: allocAnno,
+				}, func() {}
+			},
+			args: buildContainerAllocateResponseArgs{
+				pod:           &v1.Pod{ObjectMeta: metav1.ObjectMeta{Annotations: map[string]string{}}},
+				containerDevs: device.ContainerDevices{cd("uuid1", "Ascend910", 1024, 4)},
+				rtInfoLookup: func() map[string]RuntimeInfo {
+					mem := int64(8738)
+					core := int32(4)
+					return map[string]RuntimeInfo{
+						"uuid1": {UUID: "uuid1", Temp: "vir08", Memory: &mem, Core: &core},
+					}
+				}(),
+			},
+			want: buildContainerAllocateResponseWant{
+				envs: map[string]string{
+					"ASCEND_VISIBLE_DEVICES": "3",
+					"NPU_MEM_QUOTA":          "8738",
+					"NPU_PRIORITY":           "4",
+					"NPU_GLOBAL_SHM_PATH":    "/hami-shared-region/3_global_registry",
+				},
+				mounts: []*v1beta1.Mount{
+					{HostPath: "/usr/local/bin/npu-smi", ContainerPath: "/usr/local/bin/npu-smi", ReadOnly: true},
+					{HostPath: "/etc/ascend_install.info", ContainerPath: "/etc/ascend_install.info", ReadOnly: true},
+					{HostPath: "/usr/local/Ascend/driver/lib64/driver", ContainerPath: "/usr/local/Ascend/driver/lib64/driver", ReadOnly: true},
+					{HostPath: "/usr/local/Ascend/driver/version.info", ContainerPath: "/usr/local/Ascend/driver/version.info", ReadOnly: true},
+					{HostPath: "/usr/local/hami-vnpu-core", ContainerPath: "/hami-vnpu-core", ReadOnly: true},
+					{HostPath: "/usr/local/hami-vnpu-core/ld.so.preload", ContainerPath: "/etc/ld.so.preload", ReadOnly: true},
+					{HostPath: "/usr/local/hami-shared-region", ContainerPath: "/hami-shared-region", ReadOnly: false},
+				},
+			},
+		},
+		{
+			name: "NoAnnotation_NodeTemplate_FollowsNode",
+			setup: func() (*PluginServer, CleanupFunc) {
+				return &PluginServer{
+					mgr: &FakeManager{
+						GetDeviceByUUIDFunc: func(uuid string) *manager.Device {
+							return &manager.Device{UUID: "uuid1", PhyID: 3}
+						},
+						IsHamiVnpuCoreFunc: func() bool { return false },
+					},
+					allocAnno: allocAnno,
+				}, func() {}
+			},
+			args: buildContainerAllocateResponseArgs{
+				pod:           &v1.Pod{ObjectMeta: metav1.ObjectMeta{Annotations: map[string]string{}}},
+				containerDevs: device.ContainerDevices{cd("uuid1", "Ascend910", 1024, 4)},
+				rtInfoLookup: func() map[string]RuntimeInfo {
+					mem := int64(8738)
+					core := int32(0)
+					return map[string]RuntimeInfo{
+						"uuid1": {UUID: "uuid1", Temp: "vir08", Memory: &mem, Core: &core},
+					}
+				}(),
+			},
+			want: buildContainerAllocateResponseWant{
+				envs: map[string]string{
+					"ASCEND_VISIBLE_DEVICES": "3",
+					"ASCEND_VNPU_SPECS":      "vir08",
+				},
+			},
+		},
 		{
 			name: "NonHamiCore_EmptyTemp",
 			setup: func() (*PluginServer, CleanupFunc) {