fix: 修复 GPU 选择索引错误及 CUDA 不可用问题，优化 GPU 硬件单选

- 修复 FineTuneCreateView GPU 选择使用 v-for idx 替代真实 gpu.id 导致
  多节点环境下 GPU 索引错误（如 gpu-node-02 仅 GPU 0 但请求 GPU 1）
- GPU 选择改为单选模式，已离线节点自动过滤不展示
- GPU 卡片增加节点编号展示
- 修复 CUDA_VISIBLE_DEVICES=all 无效值导致 torch.cuda.is_available() False
  改为 CUDA_VISIBLE_DEVICES=0
- .gitignore 新增 .claude/ CLAUDE.md 排除规则
- GpuInfo 类型增加 node_id/node_code/node_name 字段

Co-Authored-By: Claude <noreply@anthropic.com>

.gitignore

@@ -157,6 +157,8 @@ backend/config.yaml
 .codex-backups/
 .pnpm-store/
 .zcode/
+.claude/
+CLAUDE.md
 
 # Spyder project settings
 .spyderproject

docker/compute/.env

@@ -35,6 +35,6 @@ COMPUTE_GPU_MEMORY_GB=80
 COMPUTE_GPU_POWER_LIMIT_W=300
 
 LOG_DIR=/opt/yg-ft/logs/compute
-CUDA_VISIBLE_DEVICES=all
+CUDA_VISIBLE_DEVICES=0
 NVIDIA_VISIBLE_DEVICES=all
 NVIDIA_DRIVER_CAPABILITIES=compute,utility

frontend/src/types/index.ts

@@ -362,6 +362,9 @@ export interface GpuInfo {
   fan_speed?: number
   clock_mhz?: number
   driver_version?: string
+  node_id?: string
+  node_code?: string
+  node_name?: string
 }
 
 export interface SystemInfo {

frontend/src/views/fine-tune/FineTuneCreateView.vue

@@ -15,6 +15,7 @@ import {
 import { getModelList } from '@/api/modules/model'
 import { getDatasetList } from '@/api/modules/dataset'
 import { getSystemInfo } from '@/api/modules/system'
+import { getComputeNodes } from '@/api/modules/compute'
 import { TEMPLATE_GROUPS, LR_SCHEDULER_OPTIONS, QUANTIZATION_BIT_OPTIONS, QUANT_METHOD_OPTIONS, GGUF_FORMAT_OPTIONS } from '@/constants'
 import {
   DEFAULT_TRAINING_PARAMS,
@@ -35,7 +36,18 @@ const preflightResult = ref<FineTunePreflightResult | null>(null)
 const models = ref<ModelItem[]>([])
 const datasets = ref<DatasetItem[]>([])
 const gpus = ref<GpuInfo[]>([])
-const selectedGpus = ref<number[]>([])
+const computeNodes = ref<Array<{ id: string; scheduler_status?: string }>>([])
+const selectedGpuId = ref<number | null>(null)
+
+/** Only show GPUs from nodes that are online or draining */
+const availableGpus = computed(() => {
+  const onlineNodeIds = new Set(
+    computeNodes.value
+      .filter((n) => n.scheduler_status === 'online' || n.scheduler_status === 'draining')
+      .map((n) => n.id),
+  )
+  return gpus.value.filter((gpu) => !gpu.node_id || onlineNodeIds.has(gpu.node_id))
+})
 const modelDialogVisible = ref(false)
 
 const form = reactive(createDefaultFineTuneForm())
@@ -62,7 +74,8 @@ const selectedModel = computed(() => models.value.find((model) => model.id === f
 const modelDialogTitle = computed(() => selectedModel.value?.name || '')
 
 /** 训练命令与提交载荷共用同一份表单模型。 */
-const commandPreview = computed(() => buildFineTuneCommand(form, selectedGpus.value))
+const selectedGpuIds = computed(() => (selectedGpuId.value != null ? [selectedGpuId.value] : []))
+const commandPreview = computed(() => buildFineTuneCommand(form, selectedGpuIds.value))
 
 const remoteCommandPreview = computed(() => {
   const command = preflightResult.value?.preview?.command
@@ -70,11 +83,9 @@ const remoteCommandPreview = computed(() => {
   return preflightResult.value?.preview?.command_text || ''
 })
 
-/** GPU 多选切换 */
-function toggleGpu(index: number) {
-  const idx = selectedGpus.value.indexOf(index)
-  if (idx === -1) selectedGpus.value.push(index)
-  else selectedGpus.value.splice(idx, 1)
+/** GPU 单选切换（每次只选中一张 GPU） */
+function toggleGpu(gpuId: number) {
+  selectedGpuId.value = selectedGpuId.value === gpuId ? null : gpuId
 }
 
 function gpuUsageWidth(percent: number) {
@@ -164,10 +175,11 @@ async function loadDatasets() {
 
 async function loadGpus() {
   try {
-    const sys = await getSystemInfo()
+    const [sys, nodes] = await Promise.all([getSystemInfo(), getComputeNodes().catch(() => [])])
     gpus.value = sys?.gpu || []
-    // 默认选中第一个
-    if (gpus.value.length > 0) selectedGpus.value = [0]
+    computeNodes.value = nodes || []
+    // Default select first available GPU
+    if (availableGpus.value.length > 0) selectedGpuId.value = availableGpus.value[0].id ?? null
   } catch {
     gpus.value = []
   }
@@ -177,8 +189,8 @@ async function handleSubmit() {
   if (!formRef.value) return
   await formRef.value.validate(async (valid) => {
     if (!valid) return
-    if (selectedGpus.value.length === 0) {
-      ElMessage.warning('请至少选择一个 GPU')
+    if (selectedGpuId.value == null) {
+      ElMessage.warning('请选择一个 GPU')
       return
     }
     submitting.value = true
@@ -195,7 +207,7 @@ async function handleSubmit() {
         return
       }
 
-      const payload = buildFineTunePayload(form, selectedGpus.value)
+      const payload = buildFineTunePayload(form, selectedGpuIds.value)
       const preflight = await runPreflight(payload)
       if (!preflight?.valid) {
         ElMessage.error('训练预检未通过，请先处理预检问题')
@@ -220,7 +232,7 @@ async function handleSubmit() {
   })
 }
 
-async function runPreflight(payload = buildFineTunePayload(form, selectedGpus.value)) {
+async function runPreflight(payload = buildFineTunePayload(form, selectedGpuIds.value)) {
   preflightLoading.value = true
   try {
     const result = await preflightFineTune(payload)
@@ -249,8 +261,8 @@ async function handlePreflightClick() {
   if (!formRef.value) return
   await formRef.value.validate(async (valid) => {
     if (!valid) return
-    if (selectedGpus.value.length === 0) {
-      ElMessage.warning('请至少选择一个 GPU')
+    if (selectedGpuId.value == null) {
+      ElMessage.warning('请选择一个 GPU')
       return
     }
     await runPreflight()
@@ -286,15 +298,18 @@ onMounted(() => {
         <el-form-item label="GPU 硬件">
           <div class="gpu-list">
             <div
-              v-for="(gpu, idx) in gpus"
-              :key="idx"
+              v-for="gpu in availableGpus"
+              :key="gpu.id"
               class="gpu-card"
-              :class="{ active: selectedGpus.includes(idx), 'is-busy': gpu.gpu_percent > 80 }"
-              @click="toggleGpu(idx)"
+              :class="{ active: selectedGpuId === gpu.id, 'is-busy': gpu.gpu_percent > 80 }"
+              @click="toggleGpu(gpu.id!)"
             >
               <div class="gpu-card-top">
                 <div class="gpu-title">
-                  <span class="gpu-index">GPU-{{ idx }}</span>
+                  <span class="gpu-index">
+                    GPU-{{ gpu.id }}
+                    <template v-if="gpu.node_code"> · {{ gpu.node_code }}</template>
+                  </span>
                   <span class="gpu-name">{{ gpu.name }}</span>
                 </div>
                 <span class="gpu-usage">{{ gpu.gpu_percent }}%</span>
@@ -308,7 +323,7 @@ onMounted(() => {
                 <span>{{ gpu.power_w }}W</span>
               </div>
             </div>
-            <div v-if="!gpus.length" class="gpu-empty">暂无 GPU 信息</div>
+            <div v-if="!availableGpus.length" class="gpu-empty">暂无可用 GPU（请检查算力节点是否在线）</div>
           </div>
         </el-form-item>