ai-forever commited on
Commit
db42fcc
1 Parent(s): bebd800

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +10 -2
README.md CHANGED
@@ -1,7 +1,7 @@
1
  ---
2
  license: apache-2.0
3
  language:
4
- - ru, en
5
  pipeline_tag: text-generation
6
  tags:
7
  - PyTorch
@@ -17,7 +17,7 @@ thumbnail: "https://github.com/sberbank-ai/mgpt"
17
 
18
  We introduce family of autoregressive GPT-like models with 1.3 billion parameters trained on 60 languages from 25 language families using Wikipedia and Colossal Clean Crawled Corpus.
19
 
20
- We reproduce the GPT-3 architecture using GPT-2 sources and the sparse attention mechanism, [Deepspeed](https://github.com/microsoft/DeepSpeed) and [Megatron]() frameworks allows us to effectively parallelize the training and inference steps. Resulting models show performance on par with the recently released [XGLM](https://arxiv.org/pdf/2112.10668.pdf) models at the same time covering more languages and enhance NLP possibilities for low resource languages.
21
 
22
  ## Code
23
  The source code for the mGPT XL model is available on [Github](https://github.com/sberbank-ai/mgpt)
@@ -33,8 +33,16 @@ bibtex
33
 
34
  ## Languages
35
 
 
 
 
36
  ## Training Data Statistics
37
 
 
 
 
 
 
38
 
39
  ## Details
40
  Model was trained with sequence length 1024 using transformers lib by [SberDevices](https://sberdevices.ru/) team on 80B tokens for 3 epochs. After that model was finetuned 1 epoch with sequence length 2048.
 
1
  ---
2
  license: apache-2.0
3
  language:
4
+ - en, az, sw, af, ar, ba, be, bxr, bg, bn, cv, hy, da, de, el, es, eu, fa, fi, fr, he, hi, hu, kk, id, it, ja, ka, ky, ko, lt, lv, mn, ml, os, mr, ms, my, nl, ro, pl, pt, sah, ru, tg, sv, ta, te, tk, th, tr, tl, tt, tyv, uk, en, ur, vi, uz, yo, zh, xal
5
  pipeline_tag: text-generation
6
  tags:
7
  - PyTorch
 
17
 
18
  We introduce family of autoregressive GPT-like models with 1.3 billion parameters trained on 60 languages from 25 language families using Wikipedia and Colossal Clean Crawled Corpus.
19
 
20
+ We reproduce the GPT-3 architecture using GPT-2 sources and the sparse attention mechanism, [Deepspeed](https://github.com/microsoft/DeepSpeed) and [Megatron](https://github.com/NVIDIA/Megatron-LM) frameworks allows us to effectively parallelize the training and inference steps. Resulting models show performance on par with the recently released [XGLM](https://arxiv.org/pdf/2112.10668.pdf) models at the same time covering more languages and enhance NLP possibilities for low resource languages.
21
 
22
  ## Code
23
  The source code for the mGPT XL model is available on [Github](https://github.com/sberbank-ai/mgpt)
 
33
 
34
  ## Languages
35
 
36
+ Model includes 60 languages: (iso codes)
37
+ ```az, sw, af, ar, ba, be, bxr, bg, bn, cv, hy, da, de, el, es, eu, fa, fi, fr, he, hi, hu, kk, id, it, ja, ka, ky, ko, lt, lv, mn, ml, os, mr, ms, my, nl, ro, pl, pt, sah, ru, tg, sv, ta, te, tk, th, tr, tl, tt, tyv, uk, en, ur, vi, uz, yo, zh, xal```
38
+
39
  ## Training Data Statistics
40
 
41
+ - Tokens: 559B
42
+
43
+ <img style="text-align:center; display:block;" src="https://huggingface.co/sberbank-ai/mGPT/blob/main/stats.png" width="200">
44
+ "General training corpus statistics"
45
+
46
 
47
  ## Details
48
  Model was trained with sequence length 1024 using transformers lib by [SberDevices](https://sberdevices.ru/) team on 80B tokens for 3 epochs. After that model was finetuned 1 epoch with sequence length 2048.