Files
test-repo/docs/.vitepress/theme/components/appendix/model-finetuning/FinetuningPipelineDemo.vue
T

334 lines
8.4 KiB
Vue
Raw Normal View History

<template>
<div class="finetuning-pipeline-demo">
<div class="pipeline-header">
<h4>微调流水线演示</h4>
<p class="subtitle">点击每个阶段了解微调的完整流程</p>
</div>
<div class="pipeline-steps">
<div
v-for="(step, index) in steps"
:key="step.id"
class="pipeline-step"
:class="{ active: activeStep === index, completed: index < activeStep }"
@click="setStep(index)"
>
<div class="step-icon">{{ step.icon }}</div>
<div class="step-label">{{ step.label }}</div>
<div v-if="index < steps.length - 1" class="step-arrow">
<svg width="24" height="24" viewBox="0 0 24 24" fill="none">
<path d="M5 12h14M13 6l6 6-6 6" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/>
</svg>
</div>
</div>
</div>
<div class="step-detail" v-if="activeStep >= 0">
<div class="detail-title">
{{ steps[activeStep].icon }} {{ steps[activeStep].label }}
</div>
<p class="detail-desc">{{ steps[activeStep].description }}</p>
<div class="detail-points">
<div v-for="(point, i) in steps[activeStep].points" :key="i" class="point-item">
<span class="point-bullet">{{ i + 1 }}</span>
<span>{{ point }}</span>
</div>
</div>
<div class="detail-example" v-if="steps[activeStep].example">
<div class="example-label">示例</div>
<code>{{ steps[activeStep].example }}</code>
</div>
</div>
<div class="pipeline-controls">
<button class="ctrl-btn" :disabled="activeStep <= 0" @click="prevStep">上一步</button>
<span class="step-indicator">{{ activeStep + 1 }} / {{ steps.length }}</span>
<button class="ctrl-btn primary" :disabled="activeStep >= steps.length - 1" @click="nextStep">下一步</button>
</div>
</div>
</template>
<script setup>
import { ref } from 'vue'
const activeStep = ref(0)
const steps = [
{
id: 'base',
icon: '🧠',
label: '选择基座模型',
description: '微调的第一步是选择一个合适的预训练基座模型。基座模型已经在海量数据上学习了通用的语言能力,我们要做的是在此基础上进行"专业化训练"。',
points: [
'根据任务需求选择模型规模(7B、13B、70B 等)',
'考虑开源许可证(Apache 2.0、Llama 许可等)',
'评估模型的基础能力是否匹配目标场景',
'常见选择:Llama、Qwen、Mistral、DeepSeek 等'
],
example: 'model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-7B")'
},
{
id: 'data',
icon: '📊',
label: '准备训练数据',
description: '高质量的训练数据是微调成功的关键。数据的质量远比数量重要——1000 条精心标注的数据,往往胜过 10 万条噪声数据。',
points: [
'收集与目标任务相关的数据样本',
'清洗数据:去重、过滤低质量内容',
'格式化为模型要求的输入格式(如 instruction-response 对)',
'划分训练集、验证集(通常 9:1)'
],
example: '{"instruction": "翻译成英文", "input": "你好世界", "output": "Hello World"}'
},
{
id: 'train',
icon: '⚙️',
label: '执行微调训练',
description: '使用准备好的数据对模型进行训练。现代微调通常采用参数高效方法(如 LoRA),只更新模型的一小部分参数,大幅降低计算成本。',
points: [
'配置训练超参数(学习率、批次大小、训练轮数)',
'选择微调策略(全量微调 / LoRA / QLoRA',
'监控训练损失曲线,防止过拟合',
'通常需要 1-4 个 GPU,训练数小时到数天'
],
example: 'trainer = SFTTrainer(model, train_dataset, peft_config=lora_config)'
},
{
id: 'eval',
icon: '📈',
label: '评估与测试',
description: '训练完成后,需要全面评估模型的表现。不仅要看自动化指标,更要进行人工评测,确保模型在真实场景中表现良好。',
points: [
'在验证集上计算损失和困惑度(Perplexity',
'使用任务特定指标(BLEU、ROUGE、准确率等)',
'人工评测:流畅度、准确性、安全性',
'与基座模型对比,确认微调带来了提升'
],
example: 'eval_results = trainer.evaluate(eval_dataset)'
},
{
id: 'deploy',
icon: '🚀',
label: '部署上线',
description: '将微调好的模型部署到生产环境,对外提供服务。部署前通常需要进行模型优化(量化、蒸馏等)以降低推理成本。',
points: [
'导出模型权重,合并 LoRA 适配器',
'应用量化技术压缩模型体积',
'选择部署方案(API 服务、边缘部署等)',
'配置监控和日志,持续跟踪线上表现'
],
example: 'model.merge_and_unload().save_pretrained("my-finetuned-model")'
}
]
function setStep(index) {
activeStep.value = index
}
function prevStep() {
if (activeStep.value > 0) activeStep.value--
}
function nextStep() {
if (activeStep.value < steps.length - 1) activeStep.value++
}
</script>
<style scoped>
.finetuning-pipeline-demo {
border: 1px solid var(--vp-c-divider);
border-radius: 12px;
padding: 24px;
margin: 16px 0;
background: var(--vp-c-bg-soft);
}
.pipeline-header h4 {
margin: 0 0 4px;
font-size: 16px;
color: var(--vp-c-text-1);
}
.subtitle {
margin: 0 0 20px;
font-size: 13px;
color: var(--vp-c-text-3);
}
.pipeline-steps {
display: flex;
align-items: center;
justify-content: center;
gap: 4px;
flex-wrap: wrap;
margin-bottom: 20px;
}
.pipeline-step {
display: flex;
align-items: center;
gap: 4px;
cursor: pointer;
transition: all 0.2s;
}
.step-icon {
width: 48px;
height: 48px;
border-radius: 50%;
display: flex;
align-items: center;
justify-content: center;
font-size: 20px;
background: var(--vp-c-bg);
border: 2px solid var(--vp-c-divider);
transition: all 0.3s;
}
.pipeline-step.active .step-icon {
border-color: var(--vp-c-brand-1);
background: var(--vp-c-brand-soft);
transform: scale(1.1);
}
.pipeline-step.completed .step-icon {
border-color: #10b981;
background: #d1fae5;
}
.step-label {
font-size: 12px;
color: var(--vp-c-text-2);
max-width: 64px;
text-align: center;
line-height: 1.3;
}
.pipeline-step.active .step-label {
color: var(--vp-c-brand-1);
font-weight: 600;
}
.step-arrow {
color: var(--vp-c-text-3);
display: flex;
align-items: center;
margin: 0 2px;
}
.step-detail {
background: var(--vp-c-bg);
border-radius: 8px;
padding: 20px;
margin-bottom: 16px;
border: 1px solid var(--vp-c-divider);
}
.detail-title {
font-size: 16px;
font-weight: 600;
color: var(--vp-c-text-1);
margin-bottom: 8px;
}
.detail-desc {
font-size: 14px;
color: var(--vp-c-text-2);
line-height: 1.7;
margin: 0 0 12px;
}
.detail-points {
display: flex;
flex-direction: column;
gap: 8px;
margin-bottom: 12px;
}
.point-item {
display: flex;
align-items: flex-start;
gap: 8px;
font-size: 13px;
color: var(--vp-c-text-2);
}
.point-bullet {
min-width: 20px;
height: 20px;
border-radius: 50%;
background: var(--vp-c-brand-soft);
color: var(--vp-c-brand-1);
display: flex;
align-items: center;
justify-content: center;
font-size: 11px;
font-weight: 600;
}
.detail-example {
background: var(--vp-c-bg-soft);
border-radius: 6px;
padding: 12px;
}
.example-label {
font-size: 11px;
color: var(--vp-c-text-3);
margin-bottom: 6px;
text-transform: uppercase;
letter-spacing: 0.5px;
}
.detail-example code {
font-size: 12px;
color: var(--vp-c-brand-1);
word-break: break-all;
}
.pipeline-controls {
display: flex;
align-items: center;
justify-content: center;
gap: 16px;
}
.ctrl-btn {
padding: 6px 16px;
border-radius: 6px;
border: 1px solid var(--vp-c-divider);
background: var(--vp-c-bg);
color: var(--vp-c-text-2);
cursor: pointer;
font-size: 13px;
transition: all 0.2s;
}
.ctrl-btn:hover:not(:disabled) {
border-color: var(--vp-c-brand-1);
color: var(--vp-c-brand-1);
}
.ctrl-btn:disabled {
opacity: 0.4;
cursor: not-allowed;
}
.ctrl-btn.primary {
background: var(--vp-c-brand-1);
color: #fff;
border-color: var(--vp-c-brand-1);
}
.ctrl-btn.primary:hover:not(:disabled) {
opacity: 0.9;
}
.step-indicator {
font-size: 13px;
color: var(--vp-c-text-3);
}
</style>