jina-reranker-v1-tiny-en

jina-reranker-v1-tiny-en项目介绍

项目概述

jina-reranker-v1-tiny-en是一个专为极快的重新排序而设计的模型，同时保持具有竞争力的性能。这个模型使用JinaBERT作为其基础。JinaBERT是BERT架构的一种独特变体，支持ALiBi的对称双向变体。这样的设计允许jina-reranker-v1-tiny-en处理比其他重新排序模型更长的文本序列，最多可达8,192个标记（tokens）。

为了实现速度的提升，jina-reranker-v1-tiny-en使用了一种称为知识蒸馏的技术。这种技术通过让一个复杂但较慢的模型（如原始的jina-reranker-v1-base-en）充当教师，把知识凝聚到一个较小但更快的学生模型中。这个学生模型保留了大部分教师模型的知识，使其能在极短时间内提供相似的准确性。

模型对比

以下是此项目提供的重新排序模型的对比：

从中可以看出，jina-reranker-v1-turbo-en采用了6层和37.8百万参数，为快速搜索和重新排序提供了平衡方案。而jina-reranker-v1-tiny-en则进一步强化了速度，凭借其4层和33.0百万参数结构，达到了最快的推理速度，适合于对绝对最高的准确性要求较低的场景。

使用方法

1. 通过Jina AI的Reranker API是最简单的使用方式。

curl https://api.jina.ai/v1/rerank \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
  "model": "jina-reranker-v1-tiny-en",
  "query": "Organic skincare products for sensitive skin",
  "documents": [
    "Eco-friendly kitchenware for modern homes",
    "Biodegradable cleaning supplies for eco-conscious consumers",
    "Organic cotton baby clothes for sensitive skin",
    "Natural organic skincare range for sensitive skin",
    "Tech gadgets for smart homes: 2024 edition",
    "Sustainable gardening tools and compost solutions",
    "Sensitive skin-friendly facial cleansers and toners",
    "Organic food wraps and storage solutions",
    "All-natural pet food for dogs with allergies",
    "Yoga mats made from recycled materials"
  ],
  "top_n": 3
}'

2. 或者，可以使用最新版的sentence-transformers>=0.27.0库，通过pip安装：

pip install -U sentence-transformers

然后使用以下代码与模型交互：

from sentence_transformers import CrossEncoder

model = CrossEncoder("jinaai/jina-reranker-v1-tiny-en", trust_remote_code=True)
query = "Organic skincare products for sensitive skin"
documents = [
    "Eco-friendly kitchenware for modern homes",
    "Biodegradable cleaning supplies for eco-conscious consumers",
    "Organic cotton baby clothes for sensitive skin",
    "Natural organic skincare range for sensitive skin",
    "Tech gadgets for smart homes: 2024 edition",
    "Sustainable gardening tools and compost solutions",
    "Sensitive skin-friendly facial cleansers and toners",
    "Organic food wraps and storage solutions",
    "All-natural pet food for dogs with allergies",
    "Yoga mats made from recycled materials"
]

results = model.rank(query, documents, return_documents=True, top_k=3)

3. 也可以使用transformers库通过编程方式与模型交互。

!pip install transformers
from transformers import AutoModelForSequenceClassification

model = AutoModelForSequenceClassification.from_pretrained(
    'jinaai/jina-reranker-v1-tiny-en', num_labels=1, trust_remote_code=True
)

query = "Organic skincare products for sensitive skin"
documents = [
    "Eco-friendly kitchenware for modern homes",
    "Biodegradable cleaning supplies for eco-conscious consumers",
    "Organic cotton baby clothes for sensitive skin",
    "Natural organic skincare range for sensitive skin",
    "Tech gadgets for smart homes: 2024 edition",
    "Sustainable gardening tools and compost solutions",
    "Sensitive skin-friendly facial cleansers and toners",
    "Organic food wraps and storage solutions",
    "All-natural pet food for dogs with allergies",
    "Yoga mats made from recycled materials"
]

sentence_pairs = [[query, doc] for doc in documents]
scores = model.compute_score(sentence_pairs)

4. 还可以使用transformers.js库直接在JavaScript（浏览器、Node.js、Deno等）中运行模型。

通过NPM安装Transformers.js JavaScript库：

npm i @xenova/transformers

然后可以使用以下代码与模型互动：

import { AutoTokenizer, AutoModelForSequenceClassification } from '@xenova/transformers';

const model_id = 'jinaai/jina-reranker-v1-tiny-en';
const model = await AutoModelForSequenceClassification.from_pretrained(model_id, { quantized: false });
const tokenizer = await AutoTokenizer.from_pretrained(model_id);

async function rank(query, documents, {
    top_k = undefined,
    return_documents = false,
} = {}) {
    const inputs = tokenizer(
        new Array(documents.length).fill(query),
        { text_pair: documents, padding: true, truncation: true }
    )
    const { logits } = await model(inputs);
    return logits.sigmoid().tolist()
        .map(([score], i) => ({
            corpus_id: i,
            score,
            ...(return_documents ? { text: documents[i] } : {})
        })).sort((a, b) => b.score - a.score).slice(0, top_k);
}

const query = "Organic skincare products for sensitive skin"
const documents = [
    "Eco-friendly kitchenware for modern homes",
    "Biodegradable cleaning supplies for eco-conscious consumers",
    "Organic cotton baby clothes for sensitive skin",
    "Natural organic skincare range for sensitive skin",
    "Tech gadgets for smart homes: 2024 edition",
    "Sustainable gardening tools and compost solutions",
    "Sensitive skin-friendly facial cleansers and toners",
    "Organic food wraps and storage solutions",
    "All-natural pet food for dogs with allergies",
    "Yoga mats made from recycled materials",
]

const results = await rank(query, documents, { return_documents: true, top_k: 3 });
console.log(results);

性能评估

jina-reranker在3个关键基准测试中进行评估，以确保顶级性能和搜索相关性。

NDCG@10是衡量排名质量的一种标准，分数越高表示搜索结果越好，而命中率则衡量相关文档出现在前10个搜索结果中的百分比。对于其他模型，由于不支持超过512个标记的长文档，因此没有LoCo数据集的结果。

更多详情请参考我们的基准测试表。

联系

欢迎加入我们的Discord社区，与其他社区成员交流想法。