Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,40 @@
|
|
1 |
---
|
2 |
license: other
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: other
|
3 |
---
|
4 |
+
|
5 |
+
```python
|
6 |
+
#!/bin/bash
|
7 |
+
# Apache v2 license
|
8 |
+
# Copyright (C) 2021 Intel Corporation
|
9 |
+
# SPDX-License-Identifier: Apache-2.0
|
10 |
+
|
11 |
+
# Teacher Preparation
|
12 |
+
|
13 |
+
# Notes:
|
14 |
+
# Auto mixed precision can be used by adding --fp16
|
15 |
+
# Distributed training can be used with the torch.distributed.lauch app
|
16 |
+
|
17 |
+
TEACHER_PATH=./bert-base-uncased-teacher-preparation-pretrain
|
18 |
+
OUTPUT_DIR=$TEACHER_PATH
|
19 |
+
DATA_CACHE_DIR=/root/kaokao/Model-Compression-Research-Package/examples/transformers/language-modeling/wikipedia_processed_for_pretrain
|
20 |
+
|
21 |
+
python -m torch.distributed.launch \
|
22 |
+
--nproc_per_node=8 \
|
23 |
+
../../examples/transformers/language-modeling/run_mlm.py \
|
24 |
+
--model_name_or_path bert-base-uncased \
|
25 |
+
--datasets_name_config wikipedia:20200501.en \
|
26 |
+
--data_process_type segment_pair_nsp \
|
27 |
+
--dataset_cache_dir $DATA_CACHE_DIR \
|
28 |
+
--do_train \
|
29 |
+
--learning_rate 5e-5 \
|
30 |
+
--max_steps 100000 \
|
31 |
+
--warmup_ratio 0.01 \
|
32 |
+
--weight_decay 0.01 \
|
33 |
+
--per_device_train_batch_size 8 \
|
34 |
+
--gradient_accumulation_steps 4 \
|
35 |
+
--logging_steps 10 \
|
36 |
+
--save_steps 5000 \
|
37 |
+
--save_total_limit 2 \
|
38 |
+
--output_dir $OUTPUT_DIR \
|
39 |
+
--run_name pofa-teacher-prepare-pretrain
|
40 |
+
```
|