fix(llm): require memory_mb for local_path HAMi devices (#25243)

local_path SKUs cannot estimate VRAM from InstantModel; validate HAMi devices set devices[].memory_mb on create/update and deployment.

locales/locales.go

@@ -1609,6 +1609,7 @@ var messageKeyToIndex = map[string]int{
 	"loadbalancer is already associated with eip":                                    780,
 	"loadbalancer is locked, cannot delete":                                          1366,
 	"local host storage is not empty":                                                1177,
+	"local_path SKU with HAMi requires per-GPU VRAM: set devices[].memory_mb on the LLM SKU": 2131,
 	"locally stored disks do not support detaching":                                  394,
 	"locally stored system disks do not support changing configuration":              391,
 	"login_account exceeds 32 characters":                                            820,
@@ -2172,7 +2173,7 @@ var messageKeyToIndex = map[string]int{
 	"zone_ids must be in the same cloudregion":                                         1582,
 }
 
-var en_USIndex = []uint32{ // 2132 elements
+var en_USIndex = []uint32{ // 2133 elements
 	// Entry 0 - 1F
 	0x00000000, 0x0000001b, 0x00000034, 0x00000049,
 	0x00000067, 0x0000007a, 0x00000099, 0x000000d4,
@@ -2773,9 +2774,10 @@ var en_USIndex = []uint32{ // 2132 elements
 	0x00011c60, 0x00011c78, 0x00011c8e, 0x00011c99,
 	0x00011ca7, 0x00011cba, 0x00011ccf, 0x00011ce4,
 	0x00011cf3, 0x00011d12, 0x00011d2c, 0x00011d41,
-} // Size: 8552 bytes
+	0x00011d98,
+} // Size: 8532 bytes
 
-const en_USData string = "" + // Size: 73025 bytes
+const en_USData string = "" + // Size: 73112 bytes
 	"\x02not found tenantId in body\x02failed to change project\x02get passwo" +
 	"rd in body\x02username or password is empty\x02missing credential\x02inc" +
 	"orrect username or password\x02The user has been locked, please contact " +
@@ -3881,9 +3883,10 @@ const en_USData string = "" + // Size: 73025 bytes
 	"vn controller configuration\x02unmarshal input fail %s\x02invalid charac" +
 	"ters %s\x02bad config\x02network error\x02ssh protocol error\x02invalid " +
 	"vpc provider\x02invalid ovn database\x02user not found\x02check name dup" +
-	"lication fail %s\x02getReferenceCount fail %s\x02policy is referenced"
+	"lication fail %s\x02getReferenceCount fail %s\x02policy is referenced" +
+	"\x02local_path SKU with HAMi requires per-GPU VRAM: set devices[].memory_mb on the LLM SKU"
 
-var zh_CNIndex = []uint32{ // 2132 elements
+var zh_CNIndex = []uint32{ // 2133 elements
 	// Entry 0 - 1F
 	0x00000000, 0x0000001b, 0x0000002e, 0x00000049,
 	0x00000062, 0x00000075, 0x00000094, 0x000000bf,
@@ -4484,9 +4487,10 @@ var zh_CNIndex = []uint32{ // 2132 elements
 	0x00010c9c, 0x00010cb1, 0x00010cc7, 0x00010cd7,
 	0x00010ce4, 0x00010cf4, 0x00010d0a, 0x00010d20,
 	0x00010d30, 0x00010d4e, 0x00010d71, 0x00010d8a,
-} // Size: 8552 bytes
+	0x00010e12,
+} // Size: 8532 bytes
 
-const zh_CNData string = "" + // Size: 69002 bytes
+const zh_CNData string = "" + // Size: 69138 bytes
 	"\x02获取tenantId参数失败\x02切换项目失败\x02获取password参数失败\x02用户名或密码为空\x02缺少认证信息\x02" +
 	"用户名或者密码不正确\x02用户已经被锁定，请联系管理员\x02用户已经被禁用，请联系管理员\x02用户已经被禁用或处于不可用状态\x02无" +
 	"效的认证信息\x02获取请求的JSON内容失败：%v\x02此用户禁止从Web控制台登录\x02请求内容为空\x02缺少id参数\x02fe" +
@@ -5070,6 +5074,7 @@ const zh_CNData string = "" + // Size: 69002 bytes
 	"告警条件类型: %s\x02解析URL %q出错：%v\x02无效的IP\x02停止dap搜索\x02ovn controller\x02o" +
 	"vn controller配置\x02解析input失败 %s\x02不支持的字符 %s\x02无效的配置\x02网络错误\x02ssh协议错误" +
 	"\x02无效的vpc provider\x02无效的ovn database\x02找不到用户\x02检查重复名称失败：%s\x02getRef" +
-	"erenceCount调用出错：%s\x02权限定义正被使用"
+	"erenceCount调用出错：%s\x02权限定义正被使用" +
+	"\x02主机本地模型（local_path）推理模板使用 HAMi 时必须设置每卡显存：请到推理模板中填写 GPU 显存（MiB）"
 
 	// Total table size 159131 bytes (155KiB); checksum: A1EC1008

locales/zh-CN/messages.gotext.json

@@ -9951,6 +9951,11 @@
             "message": "auto_gpu_memory_utilization requires GPU devices: configure GPU on the LLM SKU",
             "translation": "开启 auto_gpu_memory_utilization 需要先在推理模板中配置 GPU 设备"
         },
+        {
+            "id": "local_path SKU with HAMi requires per-GPU VRAM: set devices[].memory_mb on the LLM SKU",
+            "message": "local_path SKU with HAMi requires per-GPU VRAM: set devices[].memory_mb on the LLM SKU",
+            "translation": "主机本地模型（local_path）推理模板使用 HAMi 时必须设置每卡显存：请到推理模板中填写 GPU 显存（MiB）"
+        },
         {
             "id": "LLM is not running",
             "message": "LLM is not running",

pkg/llm/models/llm_deployment.go

@@ -153,6 +153,11 @@ func (man *SLLMDeploymentManager) ValidateCreateData(
 		if err := validateLocalPathDeploymentPreferHosts(ctx, userCred, input, lSku); err != nil {
 			return input, err
 		}
+		if SkuHasLocalHostPathModel(lSku) {
+			if err := ValidateLocalPathHamiDevicesRequireMemoryMb(lSku.Devices); err != nil {
+				return input, err
+			}
+		}
 		if err := ValidateDeploymentDevices(lSku.LLMType, lSku); err != nil {
 			return input, err
 		}
@@ -190,6 +195,11 @@ func (man *SLLMDeploymentManager) ValidateCreateData(
 		if err := validateLocalPathDeploymentPreferHosts(ctx, userCred, input, skuInput); err != nil {
 			return input, err
 		}
+		if SkuHasLocalHostPathModel(skuInput) {
+			if err := ValidateLocalPathHamiDevicesRequireMemoryMb(skuInput.Devices); err != nil {
+				return input, err
+			}
+		}
 		if err := ValidateDeploymentDevices(input.SkuSpec.LLMType, skuInput); err != nil {
 			return input, err
 		}

pkg/llm/models/llm_sku.go

@@ -249,6 +249,9 @@ func (man *SLLMSkuManager) ValidateCreateData(ctx context.Context, userCred mccl
 		if err := ValidateLocalPathSkuCreate(input); err != nil {
 			return input, err
 		}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(input.Devices); err != nil {
+			return input, err
+		}
 		resolved, err := resolvePreferHosts(ctx, userCred, input.PreferHosts)
 		if err != nil {
 			return input, err
@@ -437,6 +440,12 @@ func (sku *SLLMSku) ValidateUpdateData(ctx context.Context, userCred mcclient.To
 		return input, errors.Wrap(err, "validate LLMSkuBaseUpdateInput")
 	}
 
+	if SkuHasLocalHostPathModel(sku) && input.Devices != nil {
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(input.Devices); err != nil {
+			return input, err
+		}
+	}
+
 	if sku.LLMSpec == nil {
 		return input, nil
 	}

pkg/llm/models/llm_sku_local_path.go

@@ -6,6 +6,7 @@ import (
 
 	"yunion.io/x/pkg/errors"
 
+	computeapi "yunion.io/x/onecloud/pkg/apis/compute"
 	api "yunion.io/x/onecloud/pkg/apis/llm"
 	"yunion.io/x/onecloud/pkg/httperrors"
 )
@@ -86,3 +87,25 @@ func SkuHasLocalHostPathModel(sku *SLLMSku) bool {
 	}
 	return hostPathsHasContainerMount(*sku.HostPaths, 0)
 }
+
+// ValidateLocalPathHamiDevicesRequireMemoryMb requires every HAMi device to set
+// memory_mb for local_path SKUs (no InstantModel VRAM estimate available).
+// Devices are normalized first so empty SharingMode (default HAMi) is treated
+// the same as pod create.
+func ValidateLocalPathHamiDevicesRequireMemoryMb(devices *api.Devices) error {
+	if devices == nil || len(*devices) == 0 {
+		return nil
+	}
+	for i := range *devices {
+		dev := (*devices)[i]
+		normalizeLLMSkuDevice(&dev)
+		if strings.TrimSpace(dev.SharingMode) != computeapi.DEVICE_SHARING_MODE_HAMI {
+			continue
+		}
+		if dev.MemoryMb <= 0 {
+			return httperrors.NewInputParameterError(
+				"local_path SKU with HAMi requires per-GPU VRAM: set devices[].memory_mb on the LLM SKU")
+		}
+	}
+	return nil
+}

pkg/llm/models/llm_sku_local_path_test.go

@@ -119,6 +119,50 @@ func TestSkuHasLocalHostPathModel(t *testing.T) {
 	}
 }
 
+func TestValidateLocalPathHamiDevicesRequireMemoryMb(t *testing.T) {
+	t.Run("hami without memory_mb fails", func(t *testing.T) {
+		devs := llm.Devices{{Model: "A100", SharingMode: "HAMI"}}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(&devs); err == nil {
+			t.Fatal("expected error when HAMi memory_mb is missing")
+		}
+	})
+	t.Run("empty sharing_mode defaults to hami and requires memory_mb", func(t *testing.T) {
+		devs := llm.Devices{{Model: "A100"}}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(&devs); err == nil {
+			t.Fatal("expected error when default HAMi memory_mb is missing")
+		}
+	})
+	t.Run("legacy NVIDIA_HAMI without memory_mb fails", func(t *testing.T) {
+		devs := llm.Devices{{Model: "A100", DevType: "NVIDIA_HAMI"}}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(&devs); err == nil {
+			t.Fatal("expected error when legacy NVIDIA_HAMI memory_mb is missing")
+		}
+	})
+	t.Run("hami with memory_mb passes", func(t *testing.T) {
+		devs := llm.Devices{{Model: "A100", SharingMode: "HAMI", MemoryMb: 8192}}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(&devs); err != nil {
+			t.Fatalf("expected valid HAMi devices, got %v", err)
+		}
+	})
+	t.Run("empty sharing_mode with memory_mb passes", func(t *testing.T) {
+		devs := llm.Devices{{Model: "A100", MemoryMb: 8192}}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(&devs); err != nil {
+			t.Fatalf("expected default HAMi with memory_mb to pass, got %v", err)
+		}
+	})
+	t.Run("exclusive without memory_mb passes", func(t *testing.T) {
+		devs := llm.Devices{{Model: "A100", SharingMode: "EXCLUSIVE"}}
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(&devs); err != nil {
+			t.Fatalf("expected non-HAMi devices without memory_mb to pass, got %v", err)
+		}
+	})
+	t.Run("nil devices passes", func(t *testing.T) {
+		if err := ValidateLocalPathHamiDevicesRequireMemoryMb(nil); err != nil {
+			t.Fatalf("expected nil devices to pass, got %v", err)
+		}
+	})
+}
+
 func TestValidateRequireMountedModelsSkipsLocalPathSku(t *testing.T) {
 	hostPaths := llm.HostPaths{
 		{