Update README.md
Browse files
README.md
CHANGED
|
@@ -40,53 +40,53 @@ The model is pre-trained by [UER-py](https://github.com/dbiir/UER-py/) on [Tence
|
|
| 40 |
Stage1:
|
| 41 |
|
| 42 |
```
|
| 43 |
-
python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt
|
| 44 |
-
--vocab_path models/google_zh_vocab.txt
|
| 45 |
-
--dataset_path cluecorpussmall_lm_seq128_dataset.pt
|
| 46 |
--seq_length 128 --processes_num 32 --target lm
|
| 47 |
```
|
| 48 |
|
| 49 |
```
|
| 50 |
-
python3 pretrain.py --dataset_path cluecorpussmall_lm_seq128_dataset.pt
|
| 51 |
-
--vocab_path models/google_zh_vocab.txt
|
| 52 |
-
--config_path models/gpt2/distil_config.json
|
| 53 |
-
--output_model_path models/cluecorpussmall_gpt2_distil_seq128_model.bin
|
| 54 |
-
--world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7
|
| 55 |
-
--total_steps 1000000 --save_checkpoint_steps 100000 --report_steps 50000
|
| 56 |
-
--learning_rate 1e-4 --batch_size 64
|
| 57 |
-
--embedding word_pos --remove_embedding_layernorm
|
| 58 |
-
--encoder transformer --mask causal --layernorm_positioning pre
|
| 59 |
-
--target lm --
|
| 60 |
```
|
| 61 |
|
| 62 |
Stage2:
|
| 63 |
|
| 64 |
```
|
| 65 |
-
python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt
|
| 66 |
-
--vocab_path models/google_zh_vocab.txt
|
| 67 |
-
--dataset_path cluecorpussmall_lm_seq1024_dataset.pt
|
| 68 |
--seq_length 1024 --processes_num 32 --target lm
|
| 69 |
```
|
| 70 |
|
| 71 |
```
|
| 72 |
-
python3 pretrain.py --dataset_path cluecorpussmall_lm_seq1024_dataset.pt
|
| 73 |
-
--pretrained_model_path models/cluecorpussmall_gpt2_distil_seq128_model.bin-1000000
|
| 74 |
-
--vocab_path models/google_zh_vocab.txt
|
| 75 |
-
--config_path models/gpt2/distil_config.json
|
| 76 |
-
--output_model_path models/cluecorpussmall_gpt2_distil_seq1024_model.bin
|
| 77 |
-
--world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7
|
| 78 |
-
--total_steps 250000 --save_checkpoint_steps 50000 --report_steps 10000
|
| 79 |
-
--learning_rate 5e-5 --batch_size 16
|
| 80 |
-
--embedding word_pos --remove_embedding_layernorm
|
| 81 |
-
--encoder transformer --mask causal --layernorm_positioning pre
|
| 82 |
-
--target lm --
|
| 83 |
```
|
| 84 |
|
| 85 |
Finally, we convert the pre-trained model into Huggingface's format:
|
| 86 |
|
| 87 |
```
|
| 88 |
-
python3 scripts/convert_gpt2_from_uer_to_huggingface.py --input_model_path cluecorpussmall_gpt2_distil_seq1024_model.bin-250000
|
| 89 |
-
--output_model_path pytorch_model.bin
|
| 90 |
--layers_num 6
|
| 91 |
```
|
| 92 |
|
|
|
|
| 40 |
Stage1:
|
| 41 |
|
| 42 |
```
|
| 43 |
+
python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt \
|
| 44 |
+
--vocab_path models/google_zh_vocab.txt \
|
| 45 |
+
--dataset_path cluecorpussmall_lm_seq128_dataset.pt \
|
| 46 |
--seq_length 128 --processes_num 32 --target lm
|
| 47 |
```
|
| 48 |
|
| 49 |
```
|
| 50 |
+
python3 pretrain.py --dataset_path cluecorpussmall_lm_seq128_dataset.pt \
|
| 51 |
+
--vocab_path models/google_zh_vocab.txt \
|
| 52 |
+
--config_path models/gpt2/distil_config.json \
|
| 53 |
+
--output_model_path models/cluecorpussmall_gpt2_distil_seq128_model.bin \
|
| 54 |
+
--world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7 \
|
| 55 |
+
--total_steps 1000000 --save_checkpoint_steps 100000 --report_steps 50000 \
|
| 56 |
+
--learning_rate 1e-4 --batch_size 64 \
|
| 57 |
+
--embedding word_pos --remove_embedding_layernorm \
|
| 58 |
+
--encoder transformer --mask causal --layernorm_positioning pre \
|
| 59 |
+
--target lm --tie_weights
|
| 60 |
```
|
| 61 |
|
| 62 |
Stage2:
|
| 63 |
|
| 64 |
```
|
| 65 |
+
python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt \
|
| 66 |
+
--vocab_path models/google_zh_vocab.txt \
|
| 67 |
+
--dataset_path cluecorpussmall_lm_seq1024_dataset.pt \
|
| 68 |
--seq_length 1024 --processes_num 32 --target lm
|
| 69 |
```
|
| 70 |
|
| 71 |
```
|
| 72 |
+
python3 pretrain.py --dataset_path cluecorpussmall_lm_seq1024_dataset.pt \
|
| 73 |
+
--pretrained_model_path models/cluecorpussmall_gpt2_distil_seq128_model.bin-1000000 \
|
| 74 |
+
--vocab_path models/google_zh_vocab.txt \
|
| 75 |
+
--config_path models/gpt2/distil_config.json \
|
| 76 |
+
--output_model_path models/cluecorpussmall_gpt2_distil_seq1024_model.bin \
|
| 77 |
+
--world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7 \
|
| 78 |
+
--total_steps 250000 --save_checkpoint_steps 50000 --report_steps 10000 \
|
| 79 |
+
--learning_rate 5e-5 --batch_size 16 \
|
| 80 |
+
--embedding word_pos --remove_embedding_layernorm \
|
| 81 |
+
--encoder transformer --mask causal --layernorm_positioning pre \
|
| 82 |
+
--target lm --tie_weights
|
| 83 |
```
|
| 84 |
|
| 85 |
Finally, we convert the pre-trained model into Huggingface's format:
|
| 86 |
|
| 87 |
```
|
| 88 |
+
python3 scripts/convert_gpt2_from_uer_to_huggingface.py --input_model_path cluecorpussmall_gpt2_distil_seq1024_model.bin-250000 \
|
| 89 |
+
--output_model_path pytorch_model.bin \
|
| 90 |
--layers_num 6
|
| 91 |
```
|
| 92 |
|