uer commited on
Commit
b9b9ffc
1 Parent(s): 004bc3d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +21 -15
README.md CHANGED
@@ -33,39 +33,45 @@ Training data contains 3,000,000 ancient Chinese which are collected by [daizhig
33
 
34
  ## Training procedure
35
 
36
- The model is pre-trained by [UER-py](https://github.com/dbiir/UER-py/) on [Tencent Cloud TI-ONE](https://cloud.tencent.com/product/tione/). We pre-train 500,000 steps with a sequence length of 320. We use extended vocabulary to handle out-of-vocabulary words. The Chinese character that occurs greater than or equal to 100 in ancient Chinese corpus is added to the vocabulary.
37
 
38
  ```
39
- python3 preprocess.py --corpus_path corpora/ancient_chinese.txt \
40
- --vocab_path models/google_zh_vocab.txt \
41
- --dataset_path ancient_chinese_dataset.pt --processes_num 16 \
42
  --seq_length 320 --target lm
43
  ```
44
 
45
  ```
46
- python3 pretrain.py --dataset_path ancient_chinese_dataset.pt \
47
- --vocab_path models/google_zh_vocab.txt \
48
- --config_path models/bert_base_config.json \
49
- --output_model_path models/ancient_chinese_base_model.bin \
50
- --world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7 \
51
- --total_steps 500000 --save_checkpoint_steps 100000 --report_steps 10000 \
52
- --learning_rate 5e-4 --batch_size 32 \
53
- --embedding word_pos --remove_embedding_layernorm \
54
- --encoder transformer --mask causal --layernorm_positioning pre \
55
  --target lm --tie_weight
56
  ```
57
 
58
  Finally, we convert the pre-trained model into Huggingface's format:
59
 
60
  ```
61
- python3 scripts/convert_gpt2_from_uer_to_huggingface.py --input_model_path ancient_chinese_base_model.bin-500000 \
62
- --output_model_path pytorch_model.bin \
63
  --layers_num 12
64
  ```
65
 
66
  ### BibTeX entry and citation info
67
 
68
  ```
 
 
 
 
 
 
69
  @article{zhao2019uer,
70
  title={UER: An Open-Source Toolkit for Pre-training Models},
71
  author={Zhao, Zhe and Chen, Hui and Zhang, Jinbin and Zhao, Xin and Liu, Tao and Lu, Wei and Chen, Xi and Deng, Haotang and Ju, Qi and Du, Xiaoyong},
 
33
 
34
  ## Training procedure
35
 
36
+ The model is pre-trained by [UER-py](https://github.com/dbiir/UER-py/) on [Tencent Cloud](https://cloud.tencent.com/). We pre-train 500,000 steps with a sequence length of 320. We use extended vocabulary to handle out-of-vocabulary words. The Chinese character that occurs greater than or equal to 100 in ancient Chinese corpus is added to the vocabulary.
37
 
38
  ```
39
+ python3 preprocess.py --corpus_path corpora/ancient_chinese.txt \\
40
+ --vocab_path models/google_zh_vocab.txt \\
41
+ --dataset_path ancient_chinese_dataset.pt --processes_num 16 \\
42
  --seq_length 320 --target lm
43
  ```
44
 
45
  ```
46
+ python3 pretrain.py --dataset_path ancient_chinese_dataset.pt \\
47
+ --vocab_path models/google_zh_vocab.txt \\
48
+ --config_path models/bert_base_config.json \\
49
+ --output_model_path models/ancient_chinese_base_model.bin \\
50
+ --world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7 \\
51
+ --total_steps 500000 --save_checkpoint_steps 100000 --report_steps 10000 \\
52
+ --learning_rate 5e-4 --batch_size 32 \\
53
+ --embedding word_pos --remove_embedding_layernorm \\
54
+ --encoder transformer --mask causal --layernorm_positioning pre \\
55
  --target lm --tie_weight
56
  ```
57
 
58
  Finally, we convert the pre-trained model into Huggingface's format:
59
 
60
  ```
61
+ python3 scripts/convert_gpt2_from_uer_to_huggingface.py --input_model_path ancient_chinese_base_model.bin-500000 \\
62
+ --output_model_path pytorch_model.bin \\
63
  --layers_num 12
64
  ```
65
 
66
  ### BibTeX entry and citation info
67
 
68
  ```
69
+ @article{radford2019language,
70
+ title={Language Models are Unsupervised Multitask Learners},
71
+ author={Radford, Alec and Wu, Jeff and Child, Rewon and Luan, David and Amodei, Dario and Sutskever, Ilya},
72
+ year={2019}
73
+ }
74
+
75
  @article{zhao2019uer,
76
  title={UER: An Open-Source Toolkit for Pre-training Models},
77
  author={Zhao, Zhe and Chen, Hui and Zhang, Jinbin and Zhao, Xin and Liu, Tao and Lu, Wei and Chen, Xi and Deng, Haotang and Ju, Qi and Du, Xiaoyong},