管理端：训练档位选择/排队状态/按档发布（双档体系 UI）

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

server_admin/src/views/DatasetDetail.vue

@@ -10,6 +10,36 @@ const router = useRouter()
 const datasetId = computed(() => Number(route.params.id))
 const dataset = ref(null)
 
+// 顶部状态汇总：各档位训练状态依次拼（双档 2026-09-03：s 高识别 / n 高性能 各一条最新记录）；
+// running 带进度，queued 提示排队；无训练记录回退数据集状态
+const variantLabel = { s: 's高识别', n: 'n高性能' }
+const trainBanner = computed(() => {
+  const d = dataset.value
+  if (!d) return { text: '', type: 'info' }
+  const trs = d.trains || []
+  if (!trs.length) {
+    const fallback = { building: ['建设中', 'info'], labeled: ['已标注', 'success'], synced: ['已同步', 'primary'] }
+    const [text, type] = fallback[d.status] || ['', 'info']
+    return { text, type }
+  }
+  const parts = []
+  for (const t of trs) {
+    const lab = variantLabel[t.variant] || t.variant
+    if (t.status === 'running') parts.push(`${lab} 训练中 ${t.currentEpoch || 0}/${t.totalEpochs || '-'} 轮`)
+    else if (t.status === 'queued') parts.push(`${lab} 排队中`)
+    else if (t.status === 'success') parts.push(`${lab} ${t.published ? '已训练·已发布' : '已训练'}`)
+    else if (t.status === 'failed') parts.push(`${lab} 训练失败`)
+  }
+  const type = trs.some((t) => t.status === 'running')
+    ? 'primary'
+    : trs.some((t) => t.status === 'queued')
+      ? 'warning'
+      : trs.some((t) => t.status === 'failed')
+        ? 'danger'
+        : 'success'
+  return { text: parts.join(' / ') || '—', type }
+})
+
 // ---------- 数据 ----------
 const detail = ref(null) // 工作台数据：images[{filename,url,width,height,boxes,labeled}]
 const imageMeta = ref({}) // filename -> {id, source, cleanExcluded}（来自 /datasets/images，行内展示用）
@@ -910,13 +940,7 @@ onBeforeUnmount(() => {
     <div class="detail-topbar">
       <el-button link @click="router.push('/datasets')">← 返回数据集</el-button>
       <span class="detail-title">{{ dataset?.name || '数据训练' }}</span>
-      <el-tag
-        v-if="dataset"
-        :type="dataset.trainingStatus === 'running' ? 'primary' : (dataset.trainingStatus === 'success' ? 'success' : (dataset.trainingStatus === 'failed' ? 'danger' : 'info'))"
-        size="small"
-      >
-        {{ dataset.trainingStatus === 'running' ? `训练中 ${dataset.trainingCurrentEpoch || 0}/${dataset.trainingTotalEpochs || '-'}` : (dataset.trainingStatus === 'success' ? '训练完成' : (dataset.trainingStatus === 'failed' ? '训练失败' : (dataset.status === 'labeled' ? '已标注' : '建设中'))) }}
-      </el-tag>
+      <el-tag v-if="dataset" :type="trainBanner.type" size="small">{{ trainBanner.text }}</el-tag>
     </div>
 
     <!-- 图片工具：未标注/已标注 tab 走勾选预标/补标/删除；已标注 tab 另有数据清洗入口；

server_admin/src/views/Datasets.vue

@@ -44,8 +44,22 @@ const trainingIds = reactive(new Set()) // 训练发起中的数据集 id（请
 
 const statusMap = { building: '建设中', labeled: '已标注', synced: '已同步' }
 const statusTag = { building: 'info', labeled: 'success', synced: 'primary' }
-const trainStatusMap = { running: '训练中', success: '已训练', failed: '训练失败' }
-const trainStatusTag = { running: 'primary', success: 'success', failed: 'danger' }
+const trainStatusMap = { queued: '排队中', running: '训练中', success: '已训练', failed: '训练失败' }
+const trainStatusTag = { queued: 'info', running: 'primary', success: 'success', failed: 'danger' }
+// 训练档位：s=高识别(yolov8s@1280 精度优先) | n=高性能(yolov8n@704 速度优先)
+const variantMeta = {
+  s: { label: 's 高识别', tag: 'primary' },
+  n: { label: 'n 高性能', tag: 'warning' },
+}
+
+// 卡片训练状态行：各档位（s/n）最新训练，无记录为空
+function trList(row) {
+  return row?.trains || []
+}
+function activeTrs(row) {
+  // queued/running 均为进行中（双档并存时最多两条，GPU 串行）
+  return trList(row).filter((t) => t.status === 'running' || t.status === 'queued')
+}
 
 function load() {
   loading.value = true
@@ -269,21 +283,44 @@ function regenPools() {
     })
 }
 
-// ---------- 开始训练（一键发起：任务名自动生成，参数走 config.yml training 节点） ----------
+// ---------- 开始训练（弹窗选档位，默认双档 s+n 各建一条排队任务；任务名自动生成，
+//            参数走 config.yml training 节点；GPU 独占串行，忙时新任务自动排队 queued） ----------
 
-function startTrain(row) {
-  if (trainingIds.has(row.id)) return
+const startVisible = ref(false)
+const startRow = ref(null) // 当前发起训练的数据集
+const startSel = reactive({ s: true, n: true })
+
+function openStart(row) {
+  // 已有进行中档位置灰：同 (数据集,档位) 防重由后端校验，前端仅提示
+  const act = activeTrs(row).map((t) => t.variant)
+  startRow.value = row
+  startSel.s = !act.includes('s')
+  startSel.n = !act.includes('n')
+  startVisible.value = true
+}
+
+function startTrain() {
+  const variants = []
+  if (startSel.s) variants.push('s')
+  if (startSel.n) variants.push('n')
+  if (!variants.length) {
+    ElMessage.warning('请至少选择一个档位')
+    return
+  }
+  const row = startRow.value
+  if (!row || trainingIds.has(row.id)) return
   trainingIds.add(row.id)
   request
-    // 后端已异步化（请求只做校验+落记录），60s 仅作兜底
-    .post('/trainings', { datasetId: row.id }, { timeout: 60000 })
+    // 后端已异步化（请求只做校验+落 queued 记录），60s 仅作兜底
+    .post('/trainings', { datasetId: row.id, variants }, { timeout: 60000 })
     .then(() => {
-      ElMessage.success('训练任务已发起，完成后可发布为模型版本')
+      ElMessage.success('训练已加入队列，GPU 空闲后自动按顺序执行')
+      startVisible.value = false
     })
     .catch(() => {})
     .finally(() => {
       trainingIds.delete(row.id)
-      // 请求失败也可能已落 running 记录（如响应异常），一律刷新拿真实状态
+      // 请求失败也可能已落 queued 记录（如响应异常），一律刷新拿真实状态
       load()
     })
 }
@@ -320,18 +357,20 @@ function coverUrl(row) {
 }
 
 // 训练进度（百分比）：有总轮数才显示比例
-function trainPercent(row) {
-  if (row.trainingStatus !== 'running' || !row.trainingTotalEpochs) return 0
-  return Math.min(100, Math.round((row.trainingCurrentEpoch / row.trainingTotalEpochs) * 100))
+function trainPercent(t) {
+  if (t.status !== 'running' || !t.totalEpochs) return 0
+  return Math.min(100, Math.round((t.currentEpoch / t.totalEpochs) * 100))
 }
 
-function publishModel(row) {
+// 发布模型版本（按训练任务档位：s/n 各自的生效版本互不影响，版本号数据集内共用自增）
+function publishTrain(row, t) {
+  const v = variantMeta[t.variant]
   ElMessageBox.confirm(
-    `确定将「${row.name}」的最新训练结果发布为模型版本？将按数据集版本号自增（m<主>.<次>.<修订>），客户端下次启动热更新下载。`,
+    `确定将「${row.name}」的 ${v.label}（训练 #${t.trainingId}）结果发布为模型版本？版本号按数据集自增（m<主>.<次>.<修订>），客户端按识别档位热更新下载。`,
     '发布模型版本',
     { confirmButtonText: '发布', cancelButtonText: '取消', type: 'warning' },
   )
-    .then(() => request.post('/trainings/publish', { id: row.trainingId }))
+    .then(() => request.post('/trainings/publish', { id: t.trainingId }))
     .then((data) => {
       ElMessage.success(`已发布版本 ${data.version}`)
       load()
@@ -339,12 +378,31 @@ function publishModel(row) {
     .catch(() => {})
 }
 
+// 取消训练：running=终止进程；queued=直接取消排队
+function cancelTrain(row, t) {
+  const v = variantMeta[t.variant]
+  const isRun = t.status === 'running'
+  ElMessageBox.confirm(
+    isRun
+      ? `确定终止「${row.name}」${v.label} 的本次训练？进程将被终止，进度不保留。`
+      : `确定取消「${row.name}」${v.label} 的排队任务？`,
+    isRun ? '终止训练' : '取消排队训练',
+    { confirmButtonText: '确定', cancelButtonText: '再想想', type: 'warning' },
+  )
+    .then(() => request.post('/trainings/cancel', { id: t.trainingId }))
+    .then(() => {
+      ElMessage.success('已取消')
+      load()
+    })
+    .catch(() => {})
+}
+
 // 有训练中的任务时周期刷新，卡片进度条保持最新
 let trainPoll = null
 onMounted(() => {
   load()
   trainPoll = setInterval(() => {
-    if (list.value.some((r) => r.trainingStatus === 'running')) load()
+    if (list.value.some((r) => trList(r).some((t) => t.status === 'running'))) load()
   }, 10000)
 })
 
@@ -395,42 +453,45 @@ onBeforeUnmount(() => {
             size="small"
             :icon="VideoPlay"
             :loading="trainingIds.has(row.id)"
-            :disabled="trainingIds.has(row.id) || row.trainingStatus === 'running'"
-            @click="startTrain(row)"
+            :disabled="trainingIds.has(row.id) || activeTrs(row).length >= 2"
+            @click="openStart(row)"
           >
-            {{ row.trainingStatus === 'running' ? '训练中' : '开始训练' }}
+            {{ activeTrs(row).length >= 2 ? '训练中' : '开始训练' }}
           </el-button>
           <el-button size="small" :icon="EditPen" @click="openEdit(row)">编辑</el-button>
           <el-button size="small" type="danger" @click="removeDataset(row)">删除</el-button>
-          <!-- 训练成功且未发布过：发布按钮与其他操作同行（失败标签在下方状态行展示） -->
-          <el-button
-            v-if="row.trainingStatus === 'success' && row.trainingId && !row.trainingPublished"
-            size="small"
-            type="success"
-            @click="publishModel(row)"
-          >
-            发布模型
-          </el-button>
         </div>
 
-        <!-- 训练任务进度与状态（最新一条训练记录；成功态只留操作行按钮，不占独立行） -->
-        <div v-if="row.trainingStatus && row.trainingStatus !== 'success'" class="ds-train" @click.stop>
-          <template v-if="row.trainingStatus === 'running'">
-            <el-progress :percentage="trainPercent(row)" :stroke-width="6" :show-text="false" class="ds-train-bar" />
-            <div class="ds-train-text">
-              <span>训练中 {{ row.trainingCurrentEpoch || 0 }}/{{ row.trainingTotalEpochs || '-' }} 轮</span>
-            </div>
-          </template>
-          <template v-else>
-            <el-tooltip v-if="row.trainingStatus === 'failed' && row.trainingError" :content="row.trainingError" placement="top">
-              <el-tag :type="trainStatusTag[row.trainingStatus] || 'info'" size="small">
-                {{ trainStatusMap[row.trainingStatus] || row.trainingStatus }}
-              </el-tag>
+        <!-- 各档位（s/n）最新训练状态行；success 已发布只标记，未发布给发布入口 -->
+        <div v-if="trList(row).length" class="ds-train" @click.stop>
+          <div v-for="t in trList(row)" :key="t.variant" class="ds-train-line">
+            <el-tag :type="variantMeta[t.variant]?.tag || 'info'" size="small">{{ variantMeta[t.variant]?.label || t.variant }}</el-tag>
+            <template v-if="t.status === 'running'">
+              <el-progress :percentage="trainPercent(t)" :stroke-width="6" :show-text="false" class="ds-train-bar" />
+              <span class="ds-train-text">{{ t.currentEpoch || 0 }}/{{ t.totalEpochs || '-' }} 轮</span>
+            </template>
+            <el-tooltip v-else-if="t.status === 'failed' && t.error" :content="t.error" placement="top">
+              <el-tag :type="trainStatusTag[t.status] || 'info'" size="small">{{ trainStatusMap[t.status] }}</el-tag>
             </el-tooltip>
-            <el-tag v-else :type="trainStatusTag[row.trainingStatus] || 'info'" size="small">
-              {{ trainStatusMap[row.trainingStatus] || row.trainingStatus }}
+            <el-tag v-else-if="t.status !== 'success'" :type="trainStatusTag[t.status] || 'info'" size="small">
+              {{ trainStatusMap[t.status] || t.status }}
             </el-tag>
-          </template>
+            <template v-else>
+              <el-tag type="success" size="small">已训练</el-tag>
+              <el-button v-if="!t.published" size="small" type="success" class="ds-pub" @click="publishTrain(row, t)">发布模型</el-button>
+              <span v-else class="ds-train-text">已发布</span>
+            </template>
+            <el-button
+              v-if="t.status === 'running' || t.status === 'queued'"
+              size="small"
+              text
+              type="danger"
+              class="ds-cancel"
+              @click="cancelTrain(row, t)"
+            >
+              {{ t.status === 'queued' ? '取消排队' : '终止' }}
+            </el-button>
+          </div>
         </div>
       </div>
       <el-empty v-if="!loading && !list.length" description="暂无数据集，点击右上角新建" />
@@ -449,6 +510,28 @@ onBeforeUnmount(() => {
       />
     </div>
 
+    <!-- 发起训练：档位选择（默认双档 s+n；训练机 GPU 独占串行，有任务时新任务排队 queued 逐个执行） -->
+    <el-dialog v-model="startVisible" title="发起训练" width="min(440px, 94vw)">
+      <div class="start-dataset">数据集：{{ startRow?.name }}</div>
+      <div class="start-opts">
+        <div class="start-opt" :class="{ 'is-act': activeTrs(startRow).some((t) => t.variant === 's') }">
+          <el-checkbox v-model="startSel.s" :disabled="activeTrs(startRow).some((t) => t.variant === 's')">s 高识别</el-checkbox>
+          <span class="start-tip">yolov8s @1280 · 精度优先，默认档</span>
+        </div>
+        <div class="start-opt" :class="{ 'is-act': activeTrs(startRow).some((t) => t.variant === 'n') }">
+          <el-checkbox v-model="startSel.n" :disabled="activeTrs(startRow).some((t) => t.variant === 'n')">n 高性能</el-checkbox>
+          <span class="start-tip">yolov8n @704 · 速度优先</span>
+        </div>
+      </div>
+      <p class="start-note">
+        训练机 GPU 独占：已有任务时本次训练自动排队（queued），按发起顺序逐个执行；该档位已有进行中任务时会提示错误。排队的任务可在卡片上取消。
+      </p>
+      <template #footer>
+        <el-button @click="startVisible = false">取消</el-button>
+        <el-button type="primary" :loading="trainingIds.has(startRow?.id)" @click="startTrain">开始训练</el-button>
+      </template>
+    </el-dialog>
+
     <!-- 新建 / 编辑数据集（同一弹窗，编辑模式回填原值且允许改名） -->
     <el-dialog v-model="formVisible" :title="dlgForm.mode === 'create' ? '新建数据集' : '编辑数据集'" width="min(720px, 94vw)">
       <el-form label-width="90px">
@@ -605,11 +688,17 @@ onBeforeUnmount(() => {
 }
 .ds-train {
   display: flex;
-  align-items: center;
-  gap: 10px;
+  flex-direction: column;
+  gap: 6px;
   padding: 8px 12px 12px;
   border-top: 1px dashed #ebeef5;
 }
+.ds-train-line {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  min-width: 0;
+}
 .ds-train-bar {
   flex: 1;
 }
@@ -618,6 +707,43 @@ onBeforeUnmount(() => {
   color: #606266;
   white-space: nowrap;
 }
+.ds-pub {
+  margin-left: auto;
+}
+.ds-cancel {
+  margin-left: auto;
+}
+.start-dataset {
+  font-size: 14px;
+  font-weight: 600;
+  margin-bottom: 12px;
+}
+.start-opts {
+  display: flex;
+  flex-direction: column;
+  gap: 10px;
+}
+.start-opt {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  padding: 8px 10px;
+  border: 1px solid #ebeef5;
+  border-radius: 6px;
+}
+.start-opt.is-act {
+  background: #f5f7fa;
+}
+.start-tip {
+  font-size: 12px;
+  color: #909399;
+}
+.start-note {
+  font-size: 12px;
+  color: #909399;
+  line-height: 1.6;
+  margin: 12px 0 0;
+}
 .ds-body {
   padding: 10px 12px 6px;
 }