Datasets:

Modalities:
Text
Languages:
English
ArXiv:
DOI:
License:
Files changed (1) hide show
  1. README.md +25 -3
README.md CHANGED
@@ -1,3 +1,25 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: odc-by
3
+ task_categories:
4
+ - text-generation
5
+ language:
6
+ - en
7
+ pretty_name: Zyda
8
+ size_categories:
9
+ - n>1T
10
+ ---
11
+
12
+ Zyda is a unified dataset, released under a permissive license, comprising most of the largest and highest quality existing open-source datasets available.
13
+ Upon these, we have performed extensive additional filtering, beyond what was originally applied, and thorough intra- and inter-dataset deduplication.
14
+ Our aim with this work is to create a growing and extendable meta-dataset which can be easily used by practitioners to train trillion-token scale language models
15
+ while also consolidating and unifying the efforts made by disparate open-source groups who have released their datasets. Ultimately, we hope that our work can
16
+ provide an 'off-the-shelf' accessible trillion-scale high quality pretraining dataset which can be used by groups interested in pretraining their own LLM models.
17
+
18
+ While not performing new filtering from common-crawl, we believe that our work is an important and valuable step towards the creation of large scale
19
+ high quality open datasets given that there exist a number of high quality existing datasets but few to none of them individually reach the scale
20
+ necessary for training frontier models. Collating, filtering, and deduplicating the existing datasets needed to create a trillion token dataset is
21
+ nontrivial work and extremely important to raise the quality of the dataset and prevent significant amounts of inter-dataset duplicates.
22
+ This latter operation of deduplication between datasets is extremely important given the degree of duplicated documents we discovered in
23
+ the datasets that we collected in this work.
24
+
25
+ Here we are releasing a version of the dataset that was deduplicated using LSH minhash technique with 40% Jaccard similarity threshold.