File size: 1,257 Bytes
cf0deb8
 
f65d917
 
e0add86
 
 
f65d917
 
 
 
 
 
c53fdeb
2d1fc2c
 
 
 
 
e0add86
c53fdeb
 
 
 
cc5883f
c53fdeb
 
 
ecbe1f0
a6cf465
c53fdeb
 
 
 
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
---
license: cc0-1.0
language:
- en



tags:
- arxiv
- science
pretty_name: dataclysm-arxiv
size_categories:
- 1M<n<10M
---

# EVERY SINGLE PAPER ON ARXIV’S TITLE AND ABSTRACT EMBEDDED
# WITH BAAI/bge-small-en-v1.5, STORED AS PARQUET FILES, AND ONLY
# 7.4 GB UNCOMPRESSED

![image/png](https://cdn-uploads.huggingface.co/production/uploads/62a4a59791cfdc7b365ff5da/VwuifFrxpATEAPGOvYOHe.png)


# somewheresystems/dataclysm-wikipedia-titles

This dataset comprises of 3,360,984 English language arXiv papers from the Cornell/arXiv dataset, with two new columns added: title-embeddings and abstract-embeddings. These additional columns were generated using the bge-small-en-v1.5 embeddings model. The dataset was sourced from the Cornell/arXiv GCP bucket's json manifest for arXiv metadata, as of January 14th, 2024 [gs://arxiv-dataset/metadata-v5/arxiv-metadata-oai.json](gs://arxiv-dataset/metadata-v5/arxiv-metadata-oai.json) 

# Embeddings Model

We used https://huggingface.co/BAAI/bge-small-en-v1.5 to embed the `title` and `abstract` fields. 
We'll be adding a column for URL so that retrieval of the source document is possible without having to perform a URL rebuilding step using the ID.

## Contact

Please contact hi@dataclysm.xyz for inquiries.