PereLluis13
commited on
Commit
β’
614f4b7
1
Parent(s):
8b11d07
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,181 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- en
|
4 |
+
---
|
5 |
+
|
6 |
+
<div align="center">
|
7 |
+
<img src="https://github.com/SapienzaNLP/relik/blob/main/relik.png?raw=true" height="150">
|
8 |
+
<img src="https://github.com/SapienzaNLP/relik/blob/main/Sapienza_Babelscape.png?raw=true" height="50">
|
9 |
+
</div>
|
10 |
+
|
11 |
+
<div align="center">
|
12 |
+
<h1>Retrieve, Read and LinK: Fast and Accurate Entity Linking and Relation Extraction on an Academic Budget</h1>
|
13 |
+
</div>
|
14 |
+
|
15 |
+
<div style="display:flex; justify-content: center; align-items: center; flex-direction: row;">
|
16 |
+
<a href="https://2024.aclweb.org/"><img src="http://img.shields.io/badge/ACL-2024-4b44ce.svg"></a>
|
17 |
+
<a href="https://aclanthology.org/"><img src="http://img.shields.io/badge/paper-ACL--anthology-B31B1B.svg"></a>
|
18 |
+
<a href="https://arxiv.org/abs/placeholder"><img src="https://img.shields.io/badge/arXiv-placeholder-b31b1b.svg"></a>
|
19 |
+
</div>
|
20 |
+
<div style="display:flex; justify-content: center; align-items: center; flex-direction: row;">
|
21 |
+
<a href="https://huggingface.co/collections/sapienzanlp/relik-retrieve-read-and-link-665d9e4a5c3ecba98c1bef19"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Collection-FCD21D"></a>
|
22 |
+
<a href="https://github.com/SapienzaNLP/relik"><img src="https://img.shields.io/badge/GitHub-Repo-121013?logo=github&logoColor=white"></a>
|
23 |
+
<a href="https://github.com/SapienzaNLP/relik/releases"><img src="https://img.shields.io/github/v/release/SapienzaNLP/relik"></a>
|
24 |
+
</div>
|
25 |
+
|
26 |
+
This model card is for a more lightweight index for the sapienzanlp/relik-retriever-e5-base-v2-aida-blink-encoder retriever. It contains the most popular 2M entities by their frequency across Wikipedia pages.
|
27 |
+
|
28 |
+
A blazing fast and lightweight Information Extraction model for **Entity Linking** and **Relation Extraction**.
|
29 |
+
|
30 |
+
**This repository contains the weights and the index for the Entity Linking ReLiK pipeline.**
|
31 |
+
|
32 |
+
## π οΈ Installation
|
33 |
+
|
34 |
+
Installation from PyPI
|
35 |
+
|
36 |
+
```bash
|
37 |
+
pip install relik
|
38 |
+
```
|
39 |
+
|
40 |
+
<details>
|
41 |
+
<summary>Other installation options</summary>
|
42 |
+
|
43 |
+
#### Install with optional dependencies
|
44 |
+
|
45 |
+
Install with all the optional dependencies.
|
46 |
+
|
47 |
+
```bash
|
48 |
+
pip install relik[all]
|
49 |
+
```
|
50 |
+
|
51 |
+
Install with optional dependencies for training and evaluation.
|
52 |
+
|
53 |
+
```bash
|
54 |
+
pip install relik[train]
|
55 |
+
```
|
56 |
+
|
57 |
+
Install with optional dependencies for [FAISS](https://github.com/facebookresearch/faiss)
|
58 |
+
|
59 |
+
FAISS PyPI package is only available for CPU. For GPU, install it from source or use the conda package.
|
60 |
+
|
61 |
+
For CPU:
|
62 |
+
|
63 |
+
```bash
|
64 |
+
pip install relik[faiss]
|
65 |
+
```
|
66 |
+
|
67 |
+
For GPU:
|
68 |
+
|
69 |
+
```bash
|
70 |
+
conda create -n relik python=3.10
|
71 |
+
conda activate relik
|
72 |
+
|
73 |
+
# install pytorch
|
74 |
+
conda install -y pytorch=2.1.0 pytorch-cuda=12.1 -c pytorch -c nvidia
|
75 |
+
|
76 |
+
# GPU
|
77 |
+
conda install -y -c pytorch -c nvidia faiss-gpu=1.8.0
|
78 |
+
# or GPU with NVIDIA RAFT
|
79 |
+
conda install -y -c pytorch -c nvidia -c rapidsai -c conda-forge faiss-gpu-raft=1.8.0
|
80 |
+
|
81 |
+
pip install relik
|
82 |
+
```
|
83 |
+
|
84 |
+
Install with optional dependencies for serving the models with
|
85 |
+
[FastAPI](https://fastapi.tiangolo.com/) and [Ray](https://docs.ray.io/en/latest/serve/quickstart.html).
|
86 |
+
|
87 |
+
```bash
|
88 |
+
pip install relik[serve]
|
89 |
+
```
|
90 |
+
|
91 |
+
#### Installation from source
|
92 |
+
|
93 |
+
```bash
|
94 |
+
git clone https://github.com/SapienzaNLP/relik.git
|
95 |
+
cd relik
|
96 |
+
pip install -e .[all]
|
97 |
+
```
|
98 |
+
|
99 |
+
</details>
|
100 |
+
|
101 |
+
## π Quick Start
|
102 |
+
|
103 |
+
[//]: # (Write a short description of the model and how to use it with the `from_pretrained` method.)
|
104 |
+
|
105 |
+
ReLiK is a lightweight and fast model for **Entity Linking** and **Relation Extraction**.
|
106 |
+
It is composed of two main components: a retriever and a reader.
|
107 |
+
The retriever is responsible for retrieving relevant documents from a large collection,
|
108 |
+
while the reader is responsible for extracting entities and relations from the retrieved documents.
|
109 |
+
ReLiK can be used with the `from_pretrained` method to load a pre-trained pipeline.
|
110 |
+
|
111 |
+
Here is an example of how to use ReLiK for **Entity Linking**:
|
112 |
+
|
113 |
+
```python
|
114 |
+
from relik import Relik
|
115 |
+
from relik.inference.data.objects import RelikOutput
|
116 |
+
|
117 |
+
relik = Relik.from_pretrained("sapienzanlp/relik-entity-linking-large")
|
118 |
+
relik_out: RelikOutput = relik("Michael Jordan was one of the best players in the NBA.")
|
119 |
+
```
|
120 |
+
|
121 |
+
RelikOutput(
|
122 |
+
text="Michael Jordan was one of the best players in the NBA.",
|
123 |
+
tokens=['Michael', 'Jordan', 'was', 'one', 'of', 'the', 'best', 'players', 'in', 'the', 'NBA', '.'],
|
124 |
+
id=0,
|
125 |
+
spans=[
|
126 |
+
Span(start=0, end=14, label="Michael Jordan", text="Michael Jordan"),
|
127 |
+
Span(start=50, end=53, label="National Basketball Association", text="NBA"),
|
128 |
+
],
|
129 |
+
triples=[],
|
130 |
+
candidates=Candidates(
|
131 |
+
span=[
|
132 |
+
[
|
133 |
+
[
|
134 |
+
{"text": "Michael Jordan", "id": 4484083},
|
135 |
+
{"text": "National Basketball Association", "id": 5209815},
|
136 |
+
{"text": "Walter Jordan", "id": 2340190},
|
137 |
+
{"text": "Jordan", "id": 3486773},
|
138 |
+
{"text": "50 Greatest Players in NBA History", "id": 1742909},
|
139 |
+
...
|
140 |
+
]
|
141 |
+
]
|
142 |
+
]
|
143 |
+
),
|
144 |
+
)
|
145 |
+
|
146 |
+
## π Performance
|
147 |
+
|
148 |
+
We evaluate the performance of ReLiK on Entity Linking using [GERBIL](http://gerbil-qa.aksw.org/gerbil/). The following table shows the results (InKB Micro F1) of ReLiK Large and Base:
|
149 |
+
|
150 |
+
| Model | AIDA | MSNBC | Der | K50 | R128 | R500 | O15 | O16 | Tot | OOD | AIT (m:s) |
|
151 |
+
|------------------------------------------|------|-------|------|------|------|------|------|------|------|------|------------|
|
152 |
+
| GENRE | 83.7 | 73.7 | 54.1 | 60.7 | 46.7 | 40.3 | 56.1 | 50.0 | 58.2 | 54.5 | 38:00 |
|
153 |
+
| EntQA | 85.8 | 72.1 | 52.9 | 64.5 | **54.1** | 41.9 | 61.1 | 51.3 | 60.5 | 56.4 | 20:00 |
|
154 |
+
| [ReLiK<sub>Base<sub>](https://huggingface.co/sapienzanlp/relik-entity-linking-base) | 85.3 | 72.3 | 55.6 | 68.0 | 48.1 | 41.6 | 62.5 | 52.3 | 60.7 | 57.2 | 00:29 |
|
155 |
+
| β‘οΈ [ReLiK<sub>Large<sub>](https://huggingface.co/sapienzanlp/relik-entity-linking-large) | **86.4** | **75.0** | **56.3** | **72.8** | 51.7 | **43.0** | **65.1** | **57.2** | **63.4** | **60.2** | 01:46 |
|
156 |
+
|
157 |
+
Comparison systems' evaluation (InKB Micro F1) on the *in-domain* AIDA test set and *out-of-domain* MSNBC (MSN), Derczynski (Der), KORE50 (K50), N3-Reuters-128 (R128),
|
158 |
+
N3-RSS-500 (R500), OKE-15 (O15), and OKE-16 (O16) test sets. **Bold** indicates the best model.
|
159 |
+
GENRE uses mention dictionaries.
|
160 |
+
The AIT column shows the time in minutes and seconds (m:s) that the systems need to process the whole AIDA test set using an NVIDIA RTX 4090,
|
161 |
+
except for EntQA which does not fit in 24GB of RAM and for which an A100 is used.
|
162 |
+
|
163 |
+
## π€ Models
|
164 |
+
|
165 |
+
Models can be found on [π€ Hugging Face](https://huggingface.co/collections/sapienzanlp/relik-retrieve-read-and-link-665d9e4a5c3ecba98c1bef19).
|
166 |
+
|
167 |
+
## π½ Cite this work
|
168 |
+
|
169 |
+
If you use any part of this work, please consider citing the paper as follows:
|
170 |
+
|
171 |
+
```bibtex
|
172 |
+
@inproceedings{orlando-etal-2024-relik,
|
173 |
+
title = "Retrieve, Read and LinK: Fast and Accurate Entity Linking and Relation Extraction on an Academic Budget",
|
174 |
+
author = "Orlando, Riccardo and Huguet Cabot, Pere-Llu{\'\i}s and Barba, Edoardo and Navigli, Roberto",
|
175 |
+
booktitle = "Findings of the Association for Computational Linguistics: ACL 2024",
|
176 |
+
month = aug,
|
177 |
+
year = "2024",
|
178 |
+
address = "Bangkok, Thailand",
|
179 |
+
publisher = "Association for Computational Linguistics",
|
180 |
+
}
|
181 |
+
```
|