Files
YG-Datasets/frontend/src/views/project/EvalManage.vue

406 lines
11 KiB
Vue

<template>
<div class="eval-manage">
<!-- Page Header -->
<div class="page-header">
<div class="header-left">
<h2>评估管理</h2>
<p class="header-desc">评估数据集质量和模型性能</p>
</div>
<el-button type="primary" @click="showEvalDialog = true" class="create-btn">
<el-icon><Plus /></el-icon>
新建评估
</el-button>
</div>
<!-- Tabs -->
<el-tabs v-model="activeTab" class="eval-tabs">
<el-tab-pane label="评估数据集" name="eval">
<div class="eval-list" v-loading="loading">
<div v-if="!loading && evalDatasets.length === 0" class="empty-state">
<el-icon size="48"><DataAnalysis /></el-icon>
<h3>暂无评估数据集</h3>
<p>创建您的第一个评估数据集</p>
<el-button type="primary" @click="showEvalDialog = true">新建评估</el-button>
</div>
<div v-else class="evals-grid">
<div v-for="(evalSet, index) in evalDatasets" :key="evalSet.id" class="eval-card" :style="{ '--delay': index * 0.08 + 's' }">
<div class="eval-header">
<div class="eval-icon">
<el-icon size="24"><DataLine /></el-icon>
</div>
<div class="eval-actions">
<el-button type="primary" size="small" @click="runEval(evalSet)">运行</el-button>
<el-button size="small" @click="viewResults(evalSet)">结果</el-button>
</div>
</div>
<h3 class="eval-name">{{ evalSet.name }}</h3>
<div class="eval-meta">
<el-tag size="small" effect="dark">{{ getTypeName(evalSet.question_type) }}</el-tag>
<span class="meta-count">{{ evalSet.question_count || 0 }} </span>
</div>
<span class="eval-date">{{ formatDate(evalSet.created_at) }}</span>
</div>
</div>
</div>
</el-tab-pane>
<el-tab-pane label="盲测系统" name="blind-test">
<div class="blind-test" v-loading="loading">
<div v-if="!loading && blindTasks.length === 0" class="empty-state">
<el-icon size="48"><Aim /></el-icon>
<h3>暂无盲测任务</h3>
<p>创建盲测来对比模型表现</p>
<el-button type="primary" @click="showBlindTestDialog = true">创建测试</el-button>
</div>
<div v-else class="tasks-grid">
<div v-for="(task, index) in blindTasks" :key="task.id" class="task-card" :style="{ '--delay': index * 0.08 + 's' }">
<div class="task-header">
<div class="task-icon" :class="task.status">
<el-icon size="20"><component :is="getTaskIcon(task.status)" /></el-icon>
</div>
<el-tag :type="getStatusType(task.status)" size="small" effect="dark">{{ getStatusText(task.status) }}</el-tag>
</div>
<h3 class="task-name">{{ task.name }}</h3>
<el-progress :percentage="task.progress || 0" :stroke-width="6" />
<div class="task-meta">已完成: {{ task.completed || 0 }}/{{ task.total || 0 }}</div>
<div class="task-actions">
<el-button v-if="task.status === 'running'" type="danger" size="small" @click="stopTask(task)">停止</el-button>
<el-button type="primary" size="small" @click="continueTask(task)">{{ task.status === 'running' ? '查看' : '继续' }}</el-button>
</div>
</div>
</div>
</div>
</el-tab-pane>
</el-tabs>
<!-- Eval Dialog -->
<el-dialog v-model="showEvalDialog" title="新建评估数据集" width="520px" class="eval-dialog">
<el-form :model="evalForm" label-position="top">
<el-form-item label="数据集名称">
<el-input v-model="evalForm.name" placeholder="输入名称" size="large" />
</el-form-item>
<div class="form-row">
<el-form-item label="问题类型">
<el-select v-model="evalForm.question_type" placeholder="选择类型" size="large" style="width: 100%">
<el-option label="混合" value="mixed" />
<el-option label="事实性" value="fact" />
<el-option label="推理" value="reasoning" />
</el-select>
</el-form-item>
<el-form-item label="问题数量">
<el-input-number v-model="evalForm.count" :min="10" :max="500" size="large" style="width: 100%" />
</el-form-item>
</div>
</el-form>
<template #footer>
<el-button @click="showEvalDialog = false">取消</el-button>
<el-button type="primary" @click="createEval">创建</el-button>
</template>
</el-dialog>
<!-- Blind Test Dialog -->
<el-dialog v-model="showBlindTestDialog" title="创建盲测任务" width="520px" class="blind-dialog">
<el-form :model="blindForm" label-position="top">
<el-form-item label="任务名称">
<el-input v-model="blindForm.name" placeholder="输入任务名称" size="large" />
</el-form-item>
<el-form-item label="选择模型">
<el-checkbox-group v-model="blindForm.model_ids" class="model-check-group">
<el-checkbox-button label="gpt-4">GPT-4</el-checkbox-button>
<el-checkbox-button label="claude-3">Claude-3</el-checkbox-button>
<el-checkbox-button label="gemini-pro">Gemini Pro</el-checkbox-button>
<el-checkbox-button label="llama-3">Llama 3</el-checkbox-button>
</el-checkbox-group>
</el-form-item>
</el-form>
<template #footer>
<el-button @click="showBlindTestDialog = false">取消</el-button>
<el-button type="primary" @click="createBlindTest">创建</el-button>
</template>
</el-dialog>
</div>
</template>
<script setup>
import { ref, reactive, onMounted, computed } from 'vue'
import { useRoute } from 'vue-router'
import { ElMessage } from 'element-plus'
import { evalApi } from '@/api'
const route = useRoute()
const projectId = computed(() => route.params.id)
const loading = ref(false)
const activeTab = ref('eval')
const evalDatasets = ref([])
const blindTasks = ref([])
const showEvalDialog = ref(false)
const showBlindTestDialog = ref(false)
const evalForm = reactive({
name: '',
question_type: 'mixed',
count: 50
})
const blindForm = reactive({
name: '',
model_ids: []
})
const fetchEvalDatasets = async () => {
loading.value = true
try {
const res = await evalApi.list(projectId.value)
evalDatasets.value = res.data.datasets || []
} catch (error) {
evalDatasets.value = []
} finally {
loading.value = false
}
}
const createEval = async () => {
try {
await evalApi.create(projectId.value, evalForm)
ElMessage.success('创建成功')
showEvalDialog.value = false
fetchEvalDatasets()
} catch (error) {
ElMessage.error('创建失败')
}
}
const runEval = async (evalSet) => {
try {
await evalApi.run(projectId.value, evalSet.id)
ElMessage.success('评估已启动')
} catch (error) {
ElMessage.error('启动失败')
}
}
const viewResults = (evalSet) => {
ElMessage.info('结果查看功能开发中')
}
const createBlindTest = () => {
ElMessage.info('盲测功能开发中')
showBlindTestDialog.value = false
}
const stopTask = (task) => ElMessage.info('停止功能开发中')
const continueTask = (task) => ElMessage.info('继续功能开发中')
const getTypeName = (type) => {
const map = { mixed: '混合', fact: '事实性', reasoning: '推理' }
return map[type] || type
}
const getStatusType = (status) => {
const map = { pending: 'info', running: 'warning', completed: 'success', failed: 'danger' }
return map[status] || 'info'
}
const getStatusText = (status) => {
const map = { pending: '等待中', running: '运行中', completed: '已完成', failed: '失败' }
return map[status] || '未知'
}
const getTaskIcon = (status) => {
const map = { pending: 'Clock', running: 'Loading', completed: 'CircleCheck', failed: 'CircleClose' }
return map[status] || 'Clock'
}
const formatDate = (date) => {
if (!date) return ''
const d = new Date(date)
return `${d.getMonth() + 1}/${d.getDate()}`
}
onMounted(() => fetchEvalDatasets())
</script>
<style scoped>
.eval-manage {
padding: 32px;
max-width: 1200px;
}
.page-header {
display: flex;
justify-content: space-between;
align-items: center;
margin-bottom: 28px;
}
.header-left h2 {
font-size: 24px;
font-weight: 600;
margin-bottom: 4px;
}
.header-desc {
font-size: 14px;
color: var(--text-tertiary);
}
.create-btn {
padding: 10px 20px;
font-weight: 500;
border-radius: var(--radius-md);
}
.eval-tabs {
background: var(--bg-secondary);
border: 1px solid var(--border-subtle);
border-radius: var(--radius-lg);
padding: 20px;
}
.evals-grid, .tasks-grid {
display: grid;
grid-template-columns: repeat(auto-fill, minmax(300px, 1fr));
gap: 16px;
}
.eval-card, .task-card {
background: var(--bg-tertiary);
border: 1px solid var(--border-subtle);
border-radius: var(--radius-lg);
padding: 20px;
transition: all var(--transition-base);
animation: cardIn 0.4s ease backwards;
animation-delay: var(--delay);
}
@keyframes cardIn {
from { opacity: 0; transform: translateY(15px); }
}
.eval-card:hover, .task-card:hover {
border-color: var(--accent-primary);
transform: translateY(-3px);
}
.eval-header, .task-header {
display: flex;
justify-content: space-between;
align-items: flex-start;
margin-bottom: 16px;
}
.eval-icon {
width: 48px;
height: 48px;
display: flex;
align-items: center;
justify-content: center;
background: var(--accent-primary-muted);
border-radius: var(--radius-md);
color: var(--accent-primary);
}
.task-icon {
width: 40px;
height: 40px;
display: flex;
align-items: center;
justify-content: center;
border-radius: var(--radius-md);
background: var(--bg-secondary);
}
.task-icon.running {
background: rgba(251, 191, 36, 0.15);
color: var(--warning);
animation: pulse 2s infinite;
}
.task-icon.completed {
background: var(--success-muted);
color: var(--success);
}
@keyframes pulse {
0%, 100% { opacity: 1; }
50% { opacity: 0.7; }
}
.eval-actions {
display: flex;
gap: 8px;
}
.eval-name, .task-name {
font-size: 16px;
font-weight: 600;
margin-bottom: 12px;
}
.eval-meta {
display: flex;
align-items: center;
gap: 12px;
font-size: 13px;
color: var(--text-tertiary);
}
.meta-count {
color: var(--text-muted);
}
.eval-date {
display: block;
margin-top: 12px;
font-size: 12px;
color: var(--text-muted);
}
.task-meta {
font-size: 13px;
color: var(--text-secondary);
margin: 12px 0;
}
.task-actions {
display: flex;
gap: 8px;
justify-content: flex-end;
}
.empty-state {
text-align: center;
padding: 60px 20px;
}
.empty-state .el-icon {
color: var(--text-muted);
margin-bottom: 16px;
}
.empty-state h3 {
font-size: 18px;
margin-bottom: 8px;
}
.empty-state p {
color: var(--text-tertiary);
margin-bottom: 20px;
}
.form-row {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 16px;
}
.model-check-group {
display: flex;
flex-wrap: wrap;
gap: 8px;
}
</style>