|
--- |
|
tags: |
|
- transformers |
|
language: |
|
- en |
|
license: cc-by-nc-4.0 |
|
--- |
|
<h1 align="center">Linq-AI-Research/Linq-Embed-Mistral</h1> |
|
|
|
**Linq-Embed-Mistral** |
|
|
|
Linq-Embed-Mistral has been developed by building upon the foundations of the [E5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct) and [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) models. We focus on improving text retrieval using advanced data refinement methods, including sophisticated data crafting, data filtering, and negative mining techniques. These methods are applied to both existing benchmark datasets and highly tailored synthetic datasets generated via LLMs. To enhance the quality of the synthetic data, we employ extensive prompt engineering and guidance from teacher models, ensuring these methods are specifically tailored to each task. Our efforts primarily aim to create high-quality triplet datasets (query, positive example, negative example), significantly improving text retrieval performance. |
|
|
|
Linq-Embed-Mistral performs well in the MTEB benchmarks. The model excels in retrieval tasks, ranking <ins>**`1st`**</ins> among all models listed on the MTEB leaderboard with a performance score of <ins>**`xxx`**</ins>. This outstanding performance underscores its superior capability in enhancing search precision and reliability. The model achieves an average score of <ins>**`yyy`**</ins> across 56 datasets in the MTEB benchmarks, making it the highest-ranking publicly accessible model and third overall. |
|
|
|
|
|
This project is for research purposes only. Third-party datasets may be subject to additional terms and conditions under their associated licenses. Please refer to specific papers for more details: |
|
|
|
- [MTEB benchmark](https://arxiv.org/abs/2210.07316) |
|
- [Mistral](https://arxiv.org/abs/2310.06825) |
|
- [E5-mistral-7b-instruct](https://arxiv.org/pdf/2401.00368.pdf) |
|
|
|
For more details, refer to [this blog post](https://getlinq.com/blog/linq-embed-mistral/). |
|
|
|
## How to use |
|
|
|
### Transformers |
|
|
|
Here is an example of how to encode queries and passages from the Mr.TyDi training dataset. |
|
|
|
```python |
|
import torch |
|
import torch.nn.functional as F |
|
from torch import Tensor |
|
from transformers import AutoTokenizer, AutoModel |
|
|
|
def last_token_pool(last_hidden_states: Tensor, |
|
attention_mask: Tensor) -> Tensor: |
|
left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0]) |
|
if left_padding: |
|
return last_hidden_states[:, -1] |
|
else: |
|
sequence_lengths = attention_mask.sum(dim=1) - 1 |
|
batch_size = last_hidden_states.shape[0] |
|
return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths] |
|
|
|
def get_detailed_instruct(task_description: str, query: str) -> str: |
|
return f'Instruct: {task_description}\nQuery: {query}' |
|
|
|
# Each query must come with a one-sentence instruction that describes the task |
|
task = 'Given a question, retrieve Wikipedia passages that answer the question' |
|
queries = [ |
|
get_detailed_instruct(task, '최초의 원자력 발전소는 무엇인가?'), |
|
get_detailed_instruct(task, 'Who invented Hangul?') |
|
] |
|
# No need to add instruction for retrieval documents |
|
passages = [ |
|
"현재 사용되는 핵분열 방식을 이용한 전력생산은 1948년 9월 미국 테네시주 오크리지에 설치된 X-10 흑연원자로에서 전구의 불을 밝히는 데 사용되면서 시작되었다. 그리고 1954년 6월에 구소련의 오브닌스크에 건설된 흑연감속 비등경수 압력관형 원자로를 사용한 오브닌스크 원자력 발전소가 시험적으로 전력생산을 시작하였고, 최초의 상업용 원자력 엉더이로를 사용한 영국 셀라필드 원자력 단지에 위치한 콜더 홀(Calder Hall) 원자력 발전소로, 1956년 10월 17일 상업 운전을 시작하였다.", |
|
"Hangul was personally created and promulgated by the fourth king of the Joseon dynasty, Sejong the Great.[1][2] Sejong's scholarly institute, the Hall of Worthies, is often credited with the work, and at least one of its scholars was heavily involved in its creation, but it appears to have also been a personal project of Sejong." |
|
] |
|
|
|
# Load model and tokenizer |
|
tokenizer = AutoTokenizer.from_pretrained('Linq-AI-Research/Linq-Embed-Mistral') |
|
model = AutoModel.from_pretrained('Linq-AI-Research/Linq-Embed-Mistral') |
|
|
|
max_length = 4096 |
|
input_texts = [*queries, *passages] |
|
# Tokenize the input texts |
|
batch_dict = tokenizer(input_texts, max_length=max_length, padding=True, truncation=True, return_tensors="pt") |
|
outputs = model(**batch_dict) |
|
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask']) |
|
|
|
# Normalize embeddings |
|
embeddings = F.normalize(embeddings, p=2, dim=1) |
|
scores = (embeddings[:2] @ embeddings[2:].T) * 100 |
|
print(scores.tolist()) |
|
# [[73.72909545898438, 30.122783660888672], [29.155078887939453, 79.25374603271484]] |
|
``` |
|
|
|
### MTEB Benchmark Evaluation |
|
|
|
Check out [unilm/e5](https://github.com/microsoft/unilm/tree/master/e5) to reproduce evaluation results on the [BEIR](https://arxiv.org/abs/2104.08663) and [MTEB](https://arxiv.org/abs/2210.07316) benchmark. |
|
|
|
## Evaluation Result |
|
|
|
### MTEB |
|
|
|
| Model Name | Retrieval (15) | Average (56) | |
|
| :------------------------------------------------------------------------------: | :------------: | :----------: | |
|
| [Linq-Embed-Mistral](https://huggingface.co/Linq-AI-Research/Linq-Embed-Mistral) | 60.0 | 68.1 | |
|
| [NV-Embed-v1](https://huggingface.co/nvidia/NV-Embed-v1) | 59.4 | 69.3 | |
|
| [SFR-Embedding-Mistral](https://huggingface.co/Salesforce/SFR-Embedding-Mistral) | 59.0 | 67.6 | |
|
| [voyage-large-2-instruct](https://docs.voyageai.com/docs/embeddings) | 58.3 | 68.3 | |
|
| [GritLM-7B](https://huggingface.co/GritLM/GritLM-7B) | 57.4 | 66.8 | |
|
| [voyage-lite-02-instruct](https://docs.voyageai.com/docs/embeddings) | 56.6 | 67.1 | |
|
|[gte-Qwen1.5-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct)| 56.2 | 67.3 | |
|
| [e5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct) | 56.9 | 66.6 | |
|
|[google-gecko.text-embedding-preview-0409](https://cloud.google.com/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings?hl=ko#latest_models)| 55.7 | 66.3 | |
|
|[text-embedding-3-large](https://openai.com/index/new-embedding-models-and-api-updates/)| 55.4 | 64.6 | |
|
|[Cohere-embed-english-v3.0](https://huggingface.co/Cohere/Cohere-embed-english-v3.0)| 55.0 | 64.5 | |
|
|
|
Linq Research Team. |
|
|
|
- Junseong Kim |
|
- Seolhwa Lee |
|
- Jihoon Kwon |
|
- Sangmo Gu |
|
- Yejin Kim |
|
- Minkyung Cho |
|
- Jy-yong Sohn |
|
- Chanyeol Choi |
|
|
|
### Citation |
|
|
|
```bibtex |
|
@misc{LinqAIResearch2024, |
|
title={Linq-Embed-Mistral:Elevating Text Retrieval with Improved GPT Data Through Task-Specific Control and Quality Refinement}, |
|
author={Junseong Kim, Seolhwa Lee, Jihoon Kwon, Sangmo Gu, Yejin Kim, Minkyung Cho, Jy-yong Sohn, Chanyeol Choi}, |
|
howpublished={Linq AI Research Blog}, |
|
year={2024}, |
|
url={https://getlinq.com/blog/linq-embed-mistral/} |
|
} |
|
``` |
|
|
