Ontocord.AI commited on
Commit
a8c1847
·
1 Parent(s): 3368fc0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +30 -0
README.md CHANGED
@@ -1,3 +1,33 @@
1
  ---
2
  license: apache-2.0
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
  ---
4
+ ## Anh multilingual chat dataset
5
+
6
+ This is about 24M multilingual synthetic instructions intended to train a chatbot.
7
+
8
+ - cross_lingual.jsonl (~800000)
9
+ This dataset contains both the multi-lingual and cross-lingual version of the Anh data in the form of `Human: instruction\nAssistant: response` described here:
10
+ https://github.com/LAION-AI/Anh/tree/main/data .
11
+ The data is translated from a portion of the OIG dataset, which includes synthic_qa, prosocial and anthropic data. Read more about the data in LAION's OIG hf repo.
12
+ Covers these langs: zh, vi, ru, ms, pt, ja, id, hi, fr, es, de.
13
+
14
+ - xp3_sample.jsonl (~650000)
15
+ This dataset also contains a portion of the xp3 dataset converted into the standard Human/Assistant format.
16
+ See https://huggingface.co/datasets/bigscience/xP3 for the 43 languages covered by xp3.
17
+
18
+ - sungai_ul2_instructions.jsonl (~23000000)
19
+ This dataset also contains a UL2 like instruction set based on 140 languages from a subset of cc100, OSCAR and mc4.
20
+ You can find the individual datasets from which this UL2 version was created here: https://github.com/ontocord/sungai
21
+
22
+ ## Disclaimer
23
+ - Translations may be inaccurate. The web text found in the UL2 file may contain inappropriate content as it is based on web scrapped data.
24
+ - Translations were generated by M2M 12B and the output generations were limited at 512 tokens due to VRAM limit (40G).
25
+
26
+ ## License
27
+ The Anh dataset that is authored by LAION volunteers is released under an Apache 2.0 license. However, the data also includes content licensed under other permissive licenses, or web-crawled data which is used under fair use principles.
28
+
29
+ ## Acknowledgement
30
+ - Thanks to LAION's Anh multilingual chat team: @yp_yurilee, @cahya, @kevin ko, @lasse, @mattdf, @theblackcat102, @yongzx, @acul3, @logus2, @paulovn, and many others.
31
+ - Thanks to @rallio67 for the original English version of the cross_lingual dataset.
32
+ - Thanks to @theblackcat102 for his translations at https://huggingface.co/datasets/theblackcat102/instruction_translations, from which the cross-lingual data is based.
33
+ - Thanks to the authors of all the underlying datasets from which Anh is based, including the xp3 authors, OSCAR, cc100 and mc4 authors.