goldfish-models commited on
Commit
70529c9
1 Parent(s): bdcc43e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +67 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ ---
3
+ license: apache-2.0
4
+ language:
5
+ - chv
6
+ datasets:
7
+ - allenai/MADLAD-400
8
+ - cis-lmu/Glot500
9
+ - legacy-datasets/wikipedia
10
+ - oscar-corpus/OSCAR-2109
11
+ library_name: transformers
12
+ pipeline_tag: text-generation
13
+ tags:
14
+ - goldfish
15
+
16
+ ---
17
+
18
+ # chv_cyrl_100mb
19
+
20
+ Goldfish is a suite of monolingual language models trained for 350 languages.
21
+ This model is the <b>Chuvash</b> (Cyrillic script) model trained on 100MB of data, after accounting for an estimated byte premium of 1.80; content-matched text in Chuvash takes on average 1.80x as many UTF-8 bytes to encode as English.
22
+ The Goldfish models are trained primarily for comparability across languages and for low-resource languages; Goldfish performance for high-resource languages is not designed to be comparable with modern large language models (LLMs).
23
+
24
+ Note: chv_cyrl is an [individual language](https://iso639-3.sil.org/code_tables/639/data) code. It is not contained in any macrolanguage codes contained in Goldfish (for script cyrl).
25
+
26
+ All training and hyperparameter details are in our paper, [Goldfish: Monolingual Language Models for 350 Languages (Chang et al., 2024)](https://github.com/tylerachang/goldfish/blob/main/goldfish_paper_20240815.pdf).
27
+
28
+ Training code and sample usage: https://github.com/tylerachang/goldfish
29
+
30
+ Sample usage also in this Google Colab: [link](https://colab.research.google.com/drive/1rHFpnQsyXJ32ONwCosWZ7frjOYjbGCXG?usp=sharing)
31
+
32
+ ## Model details:
33
+
34
+ To access all Goldfish model details programmatically, see https://github.com/tylerachang/goldfish/model_details.json.
35
+ All models are trained with a [CLS] (same as [BOS]) token prepended, and a [SEP] (same as [EOS]) token separating sequences.
36
+ Details for this model specifically:
37
+
38
+ * Architecture: gpt2
39
+ * Parameters: 124770816
40
+ * Maximum sequence length: 512 tokens
41
+ * Training text data (raw): 179.70MB
42
+ * Training text data (byte premium scaled): 100.005MB
43
+ * Training tokens: 32877056 (x10 epochs)
44
+ * Vocabulary size: 50000
45
+ * Compute cost: 1.67833620578304e+17 FLOPs or ~15.9 NVIDIA A6000 GPU hours
46
+
47
+ Training datasets (percentages prior to deduplication):
48
+ * 53.68880%: [Languages of Russia](http://web-corpora.net/wsgi3/minorlangs/download)
49
+ * 24.72870%: [MADLAD-400 (CommonCrawl)](https://huggingface.co/datasets/allenai/MADLAD-400)
50
+ * 14.16329%: [Glot500](https://huggingface.co/datasets/cis-lmu/Glot500), including [Earthlings](https://publicdata.canterbury.ac.nz/Research/Geocorpus/CCGLU_v5.0/), [Wortschatz Leipzig Data](https://wortschatz.uni-leipzig.de/en/download), [OSCAR](https://oscar-project.org/), [Tatoeba](https://tatoeba.org/en/), [TIL](https://github.com/turkic-interlingua/til-mt), [W2C](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0022-6133-9), [Wikipedia Hugging Face](https://huggingface.co/datasets/legacy-datasets/wikipedia)
51
+ * 4.06333%: [Wikipedia 2023/08](https://dumps.wikimedia.org/)
52
+ * 3.35083%: [OSCAR 2021/09](https://huggingface.co/datasets/oscar-corpus/OSCAR-2109)
53
+ * 0.00505%: [Tatoeba](https://tatoeba.org/en/)
54
+
55
+
56
+ ## Citation
57
+
58
+ If you use this model, please cite:
59
+
60
+ ```
61
+ @article{chang-etal-2024-goldfish,
62
+ title={Goldfish: Monolingual Language Models for 350 Languages},
63
+ author={Chang, Tyler A. and Arnett, Catherine and Tu, Zhuowen and Bergen, Benjamin K.},
64
+ journal={Preprint},
65
+ year={2024},
66
+ }
67
+ ```