README.md

---
license: cc-by-nc-4.0
tags:
- feature-extraction
- sentence-similarity
- mteb
language:
  - multilingual
  - af
  - am
  - ar
  - as
  - az
  - be
  - bg
  - bn
  - br
  - bs
  - ca
  - cs
  - cy
  - da
  - de
  - el
  - en
  - eo
  - es
  - et
  - eu
  - fa
  - fi
  - fr
  - fy
  - ga
  - gd
  - gl
  - gu
  - ha
  - he
  - hi
  - hr
  - hu
  - hy
  - id
  - is
  - it
  - ja
  - jv
  - ka
  - kk
  - km
  - kn
  - ko
  - ku
  - ky
  - la
  - lo
  - lt
  - lv
  - mg
  - mk
  - ml
  - mn
  - mr
  - ms
  - my
  - ne
  - nl
  - no
  - om
  - or
  - pa
  - pl
  - ps
  - pt
  - ro
  - ru
  - sa
  - sd
  - si
  - sk
  - sl
  - so
  - sq
  - sr
  - su
  - sv
  - sw
  - ta
  - te
  - th
  - tl
  - tr
  - ug
  - uk
  - ur
  - uz
  - vi
  - xh
  - yi
  - zh
inference: false
library_name: transformers
---

<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>The embedding set trained by <a href="https://jina.ai/"><b>Jina AI</b></a>.</b>
</p>

<p align="center">
<b>Jina Embedding V3: A Multilingual Multi-Task Embedding Model</b>
</p>

## Quick Start

The easiest way to start using `jina-embeddings-v3` is with the [Jina Embedding API](https://jina.ai/embeddings/).


## Intended Usage & Model Info


`jina-embeddings-v3` is a **multilingual multi-task text embedding model** designed for a variety of NLP applications.
Based on the [XLM-RoBERTa architecture](https://huggingface.co/jinaai/xlm-roberta-flash-implementation), 
this model supports [Rotary Position Embeddings (RoPE)](https://arxiv.org/abs/2104.09864) to handle long input sequences up to **8192 tokens**.
Additionally, it features [LoRA](https://arxiv.org/abs/2106.09685) adapters to generate task-specific embeddings efficiently.

### Key Features:
- **Extended Sequence Length:** Supports up to 8192 tokens with RoPE.
- **Task-Specific Embedding:** Customize embeddings through the `task_type` argument with the following options:
    - `retrieval.query`: Used for query embeddings in asymmetric retrieval tasks
    - `retrieval.passage`: Used for passage embeddings in asymmetric retrieval tasks
    - `separation`: Used for embeddings in clustering and re-ranking applications
    - `classification`: Used for embeddings in classification tasks
    - `text-matching`: Used for embeddings in tasks that quantify similarity between two texts, such as STS or symmetric retrieval tasks
- **Matryoshka Embeddings**: Supports flexible embedding sizes (`32, 64, 128, 256, 512, 768, 1024`), allowing for truncating embeddings to fit your application.

### Model Lineage:

The `jina-embeddings-v3` model is an enhancement of the [FacebookAI/xlm-roberta-large](https://huggingface.co/FacebookAI/xlm-roberta-large) model, initially trained on 100 languages. This model's functionality has been extended through an additional pretraining phase using the [CulturaX](https://huggingface.co/datasets/uonlp/CulturaX) dataset. Additionally, LoRA was employed to increase the context length to 8192 tokens. For further optimization, contrastive fine-tuning was performed across 30 languages, improving its performance in both monolingual and cross-lingual embedding tasks.


### Supported Languages:
While the base model supports 100 languages, we've focused our tuning efforts on the following 30 languages: 
**Arabic, Bengali, Chinese, Danish, Dutch, English, Finnish, French, Georgian, German, Greek, 
Hindi, Indonesian, Italian, Japanese, Korean, Latvian, Norwegian, Polish, Portuguese, Romanian, 
Russian, Slovak, Spanish, Swedish, Thai, Turkish, Ukrainian, Urdu,** and **Vietnamese.**


## Data & Parameters

The data and training details are described in the technical report (coming soon).

## Usage

**<details><summary>Apply mean pooling when integrating the model.</summary>**
<p>

### Why Use Mean Pooling?

Mean pooling takes all token embeddings from the model's output and averages them at the sentence or paragraph level. 
This approach has been shown to produce high-quality sentence embeddings.

We provide an `encode` function that handles this for you automatically.

However, if you're working with the model directly, outside of the `encode` function, 
you'll need to apply mean pooling manually. Here's how you can do it:


```python
import torch
import torch.nn.functional as F
from transformers import AutoTokenizer, AutoModel


def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0]
    input_mask_expanded = (
        attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    )
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
        input_mask_expanded.sum(1), min=1e-9
    )


sentences = ["How is the weather today?", "What is the current weather like today?"]

tokenizer = AutoTokenizer.from_pretrained("jinaai/jina-embeddings-v3")
model = AutoModel.from_pretrained("jinaai/jina-embeddings-v3", trust_remote_code=True)

encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors="pt")

with torch.no_grad():
    model_output = model(**encoded_input)

embeddings = mean_pooling(model_output, encoded_input["attention_mask"])
embeddings = F.normalize(embeddings, p=2, dim=1)

```

</p>
</details>

The easiest way to start using `jina-embeddings-v3` is with the [Jina Embedding API](https://jina.ai/embeddings/).

Alternatively, you can use `jina-embeddings-v3` directly via Transformers package:
```bash
!pip install transformers torch einops
!pip install 'numpy<2'
```
If you run it on a GPU that support [FlashAttention-2](https://github.com/Dao-AILab/flash-attention). By 2024.9.12, it supports Ampere, Ada, or Hopper GPUs (e.g., A100, RTX 3090, RTX 4090, H100),

```bash
!pip install flash-attn --no-build-isolation
```

```python
from transformers import AutoModel

# Initialize the model
model = AutoModel.from_pretrained("jinaai/jina-embeddings-v3", trust_remote_code=True)

texts = [
    "Follow the white rabbit.",  # English
    "Sigue al conejo blanco.",  # Spanish
    "Suis le lapin blanc.",  # French
    "跟着白兔走。",  # Chinese
    "اتبع الأرنب الأبيض.",  # Arabic
    "Folge dem weißen Kaninchen.",  # German
]

# When calling the `encode` function, you can choose a `task_type` based on the use case:
# 'retrieval.query', 'retrieval.passage', 'separation', 'classification', 'text-matching'
# Alternatively, you can choose not to pass a `task_type`, and no specific LoRA adapter will be used.
embeddings = model.encode(texts, task_type="text-matching")

# Compute similarities
print(embeddings[0] @ embeddings[1].T)
```

By default, the model supports a maximum sequence length of 8192 tokens. 
However, if you want to truncate your input texts to a shorter length, you can pass the `max_length` parameter to the `encode` function:
```python
embeddings = model.encode(["Very long ... document"], max_length=2048)

```

In case you want to use **Matryoshka embeddings** and switch to a different dimension, 
you can adjust it by passing the `truncate_dim` parameter to the `encode` function:
```python
embeddings = model.encode(['Sample text'], truncate_dim=256)
```


Note that the `truncate_dim` could be any integer between 1 and 1024 for the `separation`, `classification`, and `text-matching` tasks. As for the `retrieval.passage` and `retrieval.query` tasks, the value must be larger than the length of the instruction prompt. By default, the value must be larger than 9 for the `retrieval.passage` task and larger than 12 for the `retrieval.query` task.


The latest version (3.1.0) of [SentenceTransformers](https://github.com/UKPLab/sentence-transformers) also supports `jina-embeddings-v3`:

```bash
!pip install -U sentence-transformers
```

```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("jinaai/jina-embeddings-v3", trust_remote_code=True)

task_type = "retrieval.query"
embeddings = model.encode(
    ["What is the weather like in Berlin today?"],
    task_type=task_type,
    prompt_name=task_type,
)
```



## Performance

### English MTEB
| Model                          |  Dimension  |  Average  | Classification   | Clustering   | Pair Classification   | Reranking   | Retrieval   | STS  | Summarization   |
|:------------------------------:|:-----------:|:---------:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|
| jina-embeddings-v3             |    1024     | **65.60** | **82.58**| 45.27| 84.01| 58.13| 53.87| **85.8** | 30.98|
| jina-embeddings-v2-en          |     768     |   58.12   |   68.82    | 40.08| 84.44| 55.09| 45.64| 80.00| 30.56|
| text-embedding-3-large         |    3072     |   62.03   |   75.45    | 49.01| 84.22| 59.16| 55.44| 81.04| 29.92|
| multilingual-e5-large-instruct |    4096     |   64.41   |   77.56   | 47.1 | 86.19| 58.58| 52.47| 84.78| 30.39|
| Cohere-embed-multilingual-v3.0 | 4096 |   60.08   |   64.01   | 46.6 | 86.15| 57.86| 53.84| 83.15| 30.99|

### Multilingual MTEB

|             Model              | Dimension |  Average  | Classification | Clustering | Pair Classification | Reranking | Retrieval |    STS    | Summarization |
|:------------------------------:|:---------:|:---------:|:--------------:|:----------:|:-------------------:|:---------:|:---------:|:---------:|:-------------:|
|       jina-embeddings-v3       |   1024    | **64.44** |   **71.46**    |   46.71    |        76.91        |   63.98   |   57.98   | **69.83** |       -       |
|     multilingual-e5-large      |   4096    |   59.58   |     65.22      |   42.12    |        76.95        |   63.4    |   52.37   |   64.65   |       -       |
| multilingual-e5-large-instruct |   4096    |   64.25   |     67.45      | **52.12**  |        77.79        | **69.02** | **58.38** |   68.77   |       -       |


### Long Context Tasks (LongEmbed)

|         Model          | Dimension |  Average  | NarrativeQA |  Needle   |  Passkey   |   QMSum   | SummScreen |  WikiQA   |
|:----------------------:|:---------:|:---------:|:-----------:|:---------:|:----------:|:---------:|:----------:|:---------:|
|  jina-embeddings-v3*   |   1024    | **70.39** |    33.32    | **84.00** | **100.00** | **39.75** |   92.78    |   72.46   |
|   jina-embeddings-v2   |    768    |   58.12   |    37.89    |   54.25   |   50.25    |   38.87   |   93.48    |   73.99   |
| text-embedding-3-large |   3072    |   51.30   |    44.09    |   29.25   |   63.00    |   32.49   |   84.80    |   54.16   |
|      baai-bge-m3       |   1024    |   56.56   |  **45.76**  |   40.25   |   46.00    |   35.54   | **94.09**  | **77.73** |

Notes: `*`, use the text-matching adapter


#### Matryoshka Embeddings

|  Dimension  |  Retrieval  |   STS   |
|:-----------:|:-----------:|:-------:|
|     32      |    52.54    |  76.35  |
|     64      |    58.54    |  77.03  |
|     128     |    61.64    |  77.43  |
|     256     |    62.72    |  77.56  |
|     512     |    63.16    |  77.59  |
|     768     |    63.3     |  77.59  |
|    1024     |    63.35    |  77.58  |

For a comprehensive evaluation and detailed metrics, please refer to the full paper available here (coming soon).

## Contact

Join our [Discord community](https://discord.jina.ai) and chat with other community members about ideas.

## Citation

If you find `jina-embeddings-v3` useful in your research, please cite the following paper:

```bibtex

```