2026-07-10 16:45:55 +08:00
|
|
|
|
<script setup lang="ts">
|
2026-07-12 15:39:59 +08:00
|
|
|
|
import { onMounted, ref, watch } from 'vue'
|
2026-07-10 16:45:55 +08:00
|
|
|
|
import { useRouter } from 'vue-router'
|
2026-07-12 15:39:59 +08:00
|
|
|
|
import { ElMessage } from 'element-plus'
|
2026-07-10 16:45:55 +08:00
|
|
|
|
import PageCard from '@/components/PageCard.vue'
|
2026-07-12 15:39:59 +08:00
|
|
|
|
import EvalTaskSetupStep, { type EvalTaskSetupDraft } from './create/EvalTaskSetupStep.vue'
|
|
|
|
|
|
import EvalRuleSetupStep, { type EvalRuleSetupDraft } from './create/EvalRuleSetupStep.vue'
|
|
|
|
|
|
import { createDimension, getDimensionList, startEval } from '@/api/modules/eval'
|
|
|
|
|
|
import { getTrainedModels, getModelList } from '@/api/modules/model'
|
2026-07-10 16:45:55 +08:00
|
|
|
|
import { getDatasetList } from '@/api/modules/dataset'
|
|
|
|
|
|
import { getSystemInfo } from '@/api/modules/system'
|
2026-07-12 15:39:59 +08:00
|
|
|
|
import type { DatasetItem, Dimension, GpuInfo, ModelItem, TrainedModel } from '@/types'
|
|
|
|
|
|
|
|
|
|
|
|
type StepExposed = { validate: () => Promise<boolean> }
|
2026-07-10 16:45:55 +08:00
|
|
|
|
|
|
|
|
|
|
const router = useRouter()
|
2026-07-12 15:39:59 +08:00
|
|
|
|
const loading = ref(false)
|
2026-07-10 16:45:55 +08:00
|
|
|
|
const submitting = ref(false)
|
2026-07-12 15:39:59 +08:00
|
|
|
|
const taskStepRef = ref<StepExposed>()
|
|
|
|
|
|
const ruleStepRef = ref<StepExposed>()
|
2026-07-10 16:45:55 +08:00
|
|
|
|
|
|
|
|
|
|
const trainedModels = ref<TrainedModel[]>([])
|
|
|
|
|
|
const evalDatasets = ref<DatasetItem[]>([])
|
|
|
|
|
|
const dimensions = ref<Dimension[]>([])
|
2026-07-12 15:39:59 +08:00
|
|
|
|
const evalModels = ref<ModelItem[]>([])
|
2026-07-10 16:45:55 +08:00
|
|
|
|
const gpus = ref<GpuInfo[]>([])
|
2026-07-12 15:39:59 +08:00
|
|
|
|
const createdDimensionId = ref<string | number>('')
|
2026-07-10 16:45:55 +08:00
|
|
|
|
|
2026-07-12 15:39:59 +08:00
|
|
|
|
const taskForm = ref<EvalTaskSetupDraft>({
|
2026-07-10 16:45:55 +08:00
|
|
|
|
eval_task_name: '',
|
2026-07-12 15:39:59 +08:00
|
|
|
|
model_id: '',
|
2026-07-10 16:45:55 +08:00
|
|
|
|
gpu_id: '',
|
2026-07-12 15:39:59 +08:00
|
|
|
|
data_source: 'dataset',
|
|
|
|
|
|
dataset_id: '',
|
|
|
|
|
|
leaderboard: false,
|
2026-07-10 16:45:55 +08:00
|
|
|
|
})
|
|
|
|
|
|
|
2026-07-12 15:39:59 +08:00
|
|
|
|
const ruleForm = ref<EvalRuleSetupDraft>({
|
|
|
|
|
|
ruleMode: 'existing',
|
|
|
|
|
|
dimension_id: '',
|
|
|
|
|
|
newDimension: {
|
|
|
|
|
|
name: '',
|
|
|
|
|
|
type: '',
|
|
|
|
|
|
description: '',
|
|
|
|
|
|
eval_model: '',
|
|
|
|
|
|
eval_method: '',
|
|
|
|
|
|
eval_prompt: '',
|
|
|
|
|
|
is_active: true,
|
|
|
|
|
|
is_default: false,
|
|
|
|
|
|
bleu_n: 1,
|
|
|
|
|
|
output_precision: 3,
|
|
|
|
|
|
score_min: 0,
|
|
|
|
|
|
score_max: 5,
|
|
|
|
|
|
pass_threshold: 3,
|
|
|
|
|
|
},
|
|
|
|
|
|
})
|
2026-07-10 16:45:55 +08:00
|
|
|
|
|
2026-07-12 15:39:59 +08:00
|
|
|
|
watch(
|
|
|
|
|
|
() => ruleForm.value.newDimension,
|
|
|
|
|
|
() => {
|
|
|
|
|
|
createdDimensionId.value = ''
|
|
|
|
|
|
},
|
|
|
|
|
|
{ deep: true },
|
2026-07-10 16:45:55 +08:00
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
async function loadData() {
|
2026-07-12 15:39:59 +08:00
|
|
|
|
loading.value = true
|
|
|
|
|
|
const results = await Promise.allSettled([
|
|
|
|
|
|
getTrainedModels(),
|
|
|
|
|
|
getDimensionList(),
|
|
|
|
|
|
getDatasetList(),
|
|
|
|
|
|
getSystemInfo(),
|
|
|
|
|
|
getModelList(),
|
|
|
|
|
|
])
|
|
|
|
|
|
|
|
|
|
|
|
if (results[0].status === 'fulfilled') trainedModels.value = results[0].value?.models || []
|
|
|
|
|
|
if (results[1].status === 'fulfilled') dimensions.value = results[1].value || []
|
|
|
|
|
|
if (results[2].status === 'fulfilled') {
|
|
|
|
|
|
evalDatasets.value = (results[2].value || []).filter((dataset) => dataset.type === 'eval')
|
|
|
|
|
|
}
|
|
|
|
|
|
if (results[3].status === 'fulfilled') gpus.value = results[3].value?.gpu || []
|
|
|
|
|
|
if (results[4].status === 'fulfilled') {
|
|
|
|
|
|
evalModels.value = (results[4].value || []).filter((model) => model.purpose === 'evaluation')
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
const failedCount = results.filter((result) => result.status === 'rejected').length
|
|
|
|
|
|
if (failedCount > 0) {
|
|
|
|
|
|
ElMessage.error(`部分基础数据加载失败(${failedCount} 项),请刷新页面后重试`)
|
2026-07-10 16:45:55 +08:00
|
|
|
|
}
|
2026-07-12 15:39:59 +08:00
|
|
|
|
loading.value = false
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
function buildDimensionPayload() {
|
|
|
|
|
|
const draft = ruleForm.value.newDimension
|
|
|
|
|
|
const payload: Partial<Dimension> = {
|
|
|
|
|
|
name: draft.name,
|
|
|
|
|
|
type: draft.type,
|
|
|
|
|
|
description: draft.description,
|
|
|
|
|
|
eval_model: draft.type === 'text_similarity' ? undefined : draft.eval_model,
|
|
|
|
|
|
eval_method: draft.eval_method,
|
|
|
|
|
|
eval_prompt: draft.type === 'text_similarity' ? undefined : draft.eval_prompt,
|
|
|
|
|
|
is_active: draft.is_active,
|
|
|
|
|
|
is_default: draft.is_default,
|
|
|
|
|
|
create_time: new Date().toISOString(),
|
|
|
|
|
|
}
|
|
|
|
|
|
if (draft.type === 'text_similarity') {
|
|
|
|
|
|
payload.bleu_n = draft.bleu_n
|
|
|
|
|
|
payload.output_precision = draft.output_precision
|
|
|
|
|
|
}
|
|
|
|
|
|
if (draft.type === 'metric') {
|
|
|
|
|
|
payload.score_min = draft.score_min
|
|
|
|
|
|
payload.score_max = draft.score_max
|
|
|
|
|
|
payload.pass_threshold = draft.pass_threshold
|
|
|
|
|
|
}
|
|
|
|
|
|
return payload
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
async function resolveDimensionId() {
|
|
|
|
|
|
if (ruleForm.value.ruleMode === 'baseline') return ''
|
|
|
|
|
|
if (ruleForm.value.ruleMode === 'existing') return ruleForm.value.dimension_id
|
|
|
|
|
|
if (createdDimensionId.value !== '') return createdDimensionId.value
|
|
|
|
|
|
|
|
|
|
|
|
const created = await createDimension(buildDimensionPayload())
|
|
|
|
|
|
if (created?.id === undefined || created.id === null || created.id === '') {
|
|
|
|
|
|
throw new Error('评测维度已提交,但未返回维度 ID,无法启动评测任务')
|
|
|
|
|
|
}
|
|
|
|
|
|
createdDimensionId.value = created.id
|
|
|
|
|
|
return created.id
|
2026-07-10 16:45:55 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
async function handleSubmit() {
|
2026-07-12 15:39:59 +08:00
|
|
|
|
if (loading.value || submitting.value) return
|
|
|
|
|
|
const taskValid = await taskStepRef.value?.validate()
|
|
|
|
|
|
const ruleValid = await ruleStepRef.value?.validate()
|
|
|
|
|
|
if (!taskValid || !ruleValid) {
|
|
|
|
|
|
ElMessage.warning('请检查并完善所有必填项')
|
|
|
|
|
|
return
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
submitting.value = true
|
|
|
|
|
|
try {
|
|
|
|
|
|
const dimensionId = await resolveDimensionId()
|
|
|
|
|
|
await startEval({
|
|
|
|
|
|
eval_task_name: taskForm.value.eval_task_name,
|
|
|
|
|
|
eval_type: ruleForm.value.ruleMode === 'baseline' ? 'baseline' : 'custom',
|
|
|
|
|
|
model_id: taskForm.value.model_id,
|
|
|
|
|
|
gpu_id: taskForm.value.gpu_id,
|
|
|
|
|
|
dataset_id: taskForm.value.data_source === 'dataset' ? taskForm.value.dataset_id : '',
|
|
|
|
|
|
dimension_id: dimensionId,
|
|
|
|
|
|
data_source: taskForm.value.data_source,
|
|
|
|
|
|
leaderboard: taskForm.value.leaderboard,
|
|
|
|
|
|
})
|
|
|
|
|
|
ElMessage.success('评测任务已创建并启动')
|
|
|
|
|
|
router.push('/model-eval')
|
|
|
|
|
|
} catch (error) {
|
|
|
|
|
|
const message = error instanceof Error ? error.message : '创建评测任务失败,请稍后重试'
|
|
|
|
|
|
ElMessage.error(message)
|
|
|
|
|
|
} finally {
|
|
|
|
|
|
submitting.value = false
|
|
|
|
|
|
}
|
2026-07-10 16:45:55 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
function handleCancel() {
|
|
|
|
|
|
router.back()
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
onMounted(loadData)
|
|
|
|
|
|
</script>
|
|
|
|
|
|
|
|
|
|
|
|
<template>
|
2026-07-12 15:39:59 +08:00
|
|
|
|
<PageCard title="新建评测任务" subtitle="完成任务配置与评测规则设置后启动评测">
|
|
|
|
|
|
<div v-loading="loading" class="form-container">
|
|
|
|
|
|
<div class="section-card">
|
|
|
|
|
|
<div class="section-title">任务基本配置</div>
|
|
|
|
|
|
<EvalTaskSetupStep
|
|
|
|
|
|
ref="taskStepRef"
|
|
|
|
|
|
v-model="taskForm"
|
|
|
|
|
|
:trained-models="trainedModels"
|
|
|
|
|
|
:eval-datasets="evalDatasets"
|
|
|
|
|
|
:gpus="gpus"
|
|
|
|
|
|
:loading="loading"
|
|
|
|
|
|
:disabled="submitting"
|
|
|
|
|
|
/>
|
|
|
|
|
|
</div>
|
|
|
|
|
|
|
|
|
|
|
|
<div class="section-card">
|
|
|
|
|
|
<div class="section-title">评测规则设置</div>
|
|
|
|
|
|
<EvalRuleSetupStep
|
|
|
|
|
|
ref="ruleStepRef"
|
|
|
|
|
|
v-model="ruleForm"
|
|
|
|
|
|
:dimensions="dimensions"
|
|
|
|
|
|
:eval-models="evalModels"
|
|
|
|
|
|
:loading="loading"
|
|
|
|
|
|
:disabled="submitting"
|
|
|
|
|
|
/>
|
|
|
|
|
|
</div>
|
|
|
|
|
|
</div>
|
|
|
|
|
|
|
|
|
|
|
|
<div class="form-footer">
|
|
|
|
|
|
<el-button :disabled="submitting" @click="handleCancel">取消</el-button>
|
|
|
|
|
|
<el-button
|
|
|
|
|
|
type="primary"
|
|
|
|
|
|
:loading="submitting"
|
|
|
|
|
|
:disabled="loading"
|
|
|
|
|
|
@click="handleSubmit"
|
|
|
|
|
|
>
|
|
|
|
|
|
创建并启动评测
|
|
|
|
|
|
</el-button>
|
|
|
|
|
|
</div>
|
2026-07-10 16:45:55 +08:00
|
|
|
|
</PageCard>
|
|
|
|
|
|
</template>
|
2026-07-12 15:39:59 +08:00
|
|
|
|
|
|
|
|
|
|
<style scoped>
|
|
|
|
|
|
.form-container {
|
|
|
|
|
|
display: flex;
|
|
|
|
|
|
flex-direction: column;
|
|
|
|
|
|
gap: 24px;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
.section-card {
|
|
|
|
|
|
padding: 24px;
|
|
|
|
|
|
border-radius: 8px;
|
|
|
|
|
|
background: #f9fafb;
|
|
|
|
|
|
border: 1px solid #e5e7eb;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
.section-title {
|
|
|
|
|
|
font-size: 16px;
|
|
|
|
|
|
font-weight: 600;
|
|
|
|
|
|
color: #111827;
|
|
|
|
|
|
margin-bottom: 24px;
|
|
|
|
|
|
padding-bottom: 12px;
|
|
|
|
|
|
border-bottom: 1px solid #e5e7eb;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
.form-footer {
|
|
|
|
|
|
display: flex;
|
|
|
|
|
|
justify-content: flex-end;
|
|
|
|
|
|
gap: 12px;
|
|
|
|
|
|
margin-top: 32px;
|
|
|
|
|
|
padding-top: 24px;
|
|
|
|
|
|
border-top: 1px solid #e5e7eb;
|
|
|
|
|
|
}
|
|
|
|
|
|
</style>
|