File size: 9,706 Bytes
03c97a6 67b96c4 03c97a6 182734b 86ead8f 7c006e5 182734b 80d2ebe 182734b a20b077 182734b ce1a0bf 182734b 391704c 182734b a20b077 182734b 80d2ebe 182734b a20b077 182734b a20b077 182734b 80d2ebe 182734b a20b077 ce1a0bf a20b077 ce1a0bf a20b077 80d2ebe a20b077 80d2ebe a20b077 80d2ebe a20b077 667e9cb a20b077 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 |
---
pipeline_tag: text-classification
tags:
- transformers
- sentence-transformers
- reranker
- cross-encoder
language:
- multilingual
license: cc-by-nc-4.0
---
<br><br>
<p align="center">
<img src="https://aeiljuispo.cloudimg.io/v7/https://cdn-uploads.huggingface.co/production/uploads/603763514de52ff951d89793/AFoybzd5lpBQXEBrQHuTt.png?w=200&h=200&f=face" alt="Finetuner logo: Finetuner helps you to create experiments in order to improve embeddings on search tasks. It accompanies you to deliver the last mile of performance-tuning for neural search applications." width="150px">
</p>
<p align="center">
<b>Trained by <a href="https://jina.ai/"><b>Jina AI</b></a>.</b>
</p>
# jina-reranker-v2-base-multilingual
## Intended Usage & Model Info
The **Jina Reranker v2** (`jina-reranker-v2-base-multilingual`) is a transformer-based model that has been fine-tuned for text reranking task, which is a crucial component in many information retrieval systems. It is a cross-encoder model that takes a query and a document pair as input and outputs a score indicating the relevance of the document to the query. The model is trained on a large dataset of query-document pairs and is capable of reranking documents in multiple languages with high accuracy.
Compared with the state-of-the-art reranker models, including the previous released `jina-reranker-v1-base-en`, the **Jina Reranker v2** model has demonstrated competitiveness across a series of benchmarks targeting for text retrieval, multilingual capability, function-calling-aware and text-to-SQL-aware reranking, and code retrieval tasks.
The `jina-reranker-v2-base-multilingual` model is capable of handling long texts with a context length of up to `1024` tokens, enabling the processing of extensive inputs. To enable the model to handle long texts that exceed 1024 tokens, the model uses a sliding window approach to chunk the input text into smaller pieces and rerank each chunk separately.
The model is also equipped with a flash attention mechanism, which significantly improves the model's performance.
# Usage
1. The easiest way to use `jina-reranker-v2-base-multilingual` is to call Jina AI's [Reranker API](https://jina.ai/reranker/).
```bash
curl https://api.jina.ai/v1/rerank \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "jina-reranker-v2-base-multilingual",
"query": "Organic skincare products for sensitive skin",
"documents": [
"Organic skincare for sensitive skin with aloe vera and chamomile.",
"New makeup trends focus on bold colors and innovative techniques",
"Bio-Hautpflege für empfindliche Haut mit Aloe Vera und Kamille",
"Neue Make-up-Trends setzen auf kräftige Farben und innovative Techniken",
"Cuidado de la piel orgánico para piel sensible con aloe vera y manzanilla",
"Las nuevas tendencias de maquillaje se centran en colores vivos y técnicas innovadoras",
"针对敏感肌专门设计的天然有机护肤产品",
"新的化妆趋势注重鲜艳的颜色和创新的技巧",
"敏感肌のために特別に設計された天然有機スキンケア製品",
"新しいメイクのトレンドは鮮やかな色と革新的な技術に焦点を当てています"
],
"top_n": 3
}'
```
2. You can also use the `transformers` library to interact with the model programmatically.
Before you start, install the `transformers` and `einops` libraries:
```bash
pip install transformers einops
```
And then:
```python
from transformers import AutoModelForSequenceClassification
model = AutoModelForSequenceClassification.from_pretrained(
'jinaai/jina-reranker-v2-base-multilingual',
torch_dtype="auto",
trust_remote_code=True,
)
model.to('cuda') # or 'cpu' if no GPU is available
model.eval()
# Example query and documents
query = "Organic skincare products for sensitive skin"
documents = [
"Organic skincare for sensitive skin with aloe vera and chamomile.",
"New makeup trends focus on bold colors and innovative techniques",
"Bio-Hautpflege für empfindliche Haut mit Aloe Vera und Kamille",
"Neue Make-up-Trends setzen auf kräftige Farben und innovative Techniken",
"Cuidado de la piel orgánico para piel sensible con aloe vera y manzanilla",
"Las nuevas tendencias de maquillaje se centran en colores vivos y técnicas innovadoras",
"针对敏感肌专门设计的天然有机护肤产品",
"新的化妆趋势注重鲜艳的颜色和创新的技巧",
"敏感肌のために特別に設計された天然有機スキンケア製品",
"新しいメイクのトレンドは鮮やかな色と革新的な技術に焦点を当てています",
]
# construct sentence pairs
sentence_pairs = [[query, doc] for doc in documents]
scores = model.compute_score(sentence_pairs, max_length=1024)
```
The scores will be a list of floats, where each float represents the relevance score of the corresponding document to the query. Higher scores indicate higher relevance.
For instance the returning scores in this case will be:
```bash
[0.8311430811882019, 0.09401018172502518,
0.6334102749824524, 0.08269733935594559,
0.7620701193809509, 0.09947021305561066,
0.9263036847114563, 0.05834583938121796,
0.8418256044387817, 0.11124119907617569]
```
The model gives high relevance scores to the documents that are most relevant to the query regardless of the language of the document.
Note that by default, the `jina-reranker-v2-base-multilingual` model uses [flash attention](https://github.com/Dao-AILab/flash-attention), which requires certain types of GPU hardware to run.
If you encounter any issues, you can try call `AutoModelForSequenceClassification.from_pretrained()` with `use_flash_attn=False`.
This will use the standard attention mechanism instead of flash attention.
If you want to use flash attention for fast inference, you need to install the following packages:
```bash
pip install ninja # required for flash attention
pip install flash-attn --no-build-isolation
```
Enjoy the 3x-6x speedup with flash attention! ⚡️⚡️⚡️
That's it! You can now use the `jina-reranker-v2-base-multilingual` model in your projects.
In addition to the `compute_score()` function, the `jina-reranker-v2-base-multilingual` model also provides a `model.rerank()` function that can be used to rerank documents based on a query. You can use it as follows:
```python
result = model.rerank(
query,
documents,
max_query_length=512,
max_length=1024,
top_n=3
)
```
Inside the `result` object, you will find the reranked documents along with their scores. You can use this information to further process the documents as needed.
The `rerank()` function will automatically chunk the input documents into smaller pieces if they exceed the model's maximum input length. This allows you to rerank long documents without running into memory issues.
Specifically, the `rerank()` function will split the documents into chunks of size `max_length` and rerank each chunk separately. The scores from all the chunks are then combined to produce the final reranking results. You can control the query length and document length in each chunk by setting the `max_query_length` and `max_length` parameters. The `rerank()` function also supports the `overlap` parameter (default is `80`) which determines how much overlap there is between adjacent chunks. This can be useful when reranking long documents to ensure that the model has enough context to make accurate predictions.
# Evaluation
We evaluated Jina Reranker v2 on multiple benchmarks to ensure top-tier performance and search relevance.
| Model Name | Model Size | Miracl(nDCG@10, 18 langs) | MKQA(nDCG@10, 26 langs) | BEIR(nDCG@10, 17 datasets) | MLDR(recall@10, 13 langs) | CodeSearchNet (MRR@10, 3 tasks) | AirBench (nDCG@10, zh/en) | ToolBench (recall@3, 3 tasks) | TableSearch (recall@3) |
| -----------------------------: | :-------------------------: | :-------------------------: | :-------------------------: | :----------------------------: | :----------------------------: | :----------------------------: | :----------------------------: | :----------------------------: | :----------------------------: |
| jina-reranker-v2-multilingual | 278M | 62.14 | 54.83 | 53.17 | 68.95 | 71.36 | 61.33 | 77.75 | 93.31 |
| bge-reranker-v2-m3 | 568M | 63.43 | 54.17 | 53.65 | 59.73 | 62.86 | 61.28 | 78.46 | 74.86 |
| mmarco-mMiniLMv2-L12-H384-v1 | 118M | 59.71 | 53.37 | 45.40 | 28.91 | 51.78 | 56.46 | 58.39 | 53.60 |
| jina-reranker-v1-base-en | 137M | - | - | 52.45 | - | - | - | 74.13 | 72.89 |
Note:
- NDCG@10 and MRR@10 measure ranking quality, with higher scores indicating better search results
- recall@3 measures the proportion of relevant documents retrieved, with higher scores indicating better search results
|