Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
datasets:
|
4 |
+
- wikipedia
|
5 |
+
- bookcorpus
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
metrics:
|
9 |
+
- glue
|
10 |
+
library_name: transformers
|
11 |
+
---
|
12 |
+
|
13 |
+
This is our reproduction using the official HuggingFace `roberta` architecture with a medium size. On the architecture side, RoBERTa is exactly the same as BERT except for its larger vocabulary size.
|
14 |
+
|
15 |
+
According to Google's [BERT releases](https://huggingface.co/google/bert_uncased_L-8_H-512_A-8) and [BERT-Medium](https://huggingface.co/google/bert_uncased_L-8_H-512_A-8/blob/main/config.json), a medium sized model should have a config of Layer=8, Hidden=512, #AttnHeads=8, and IntermediateSize=2048. We follow this config to pre-train a RoBERTa-base model for reproduction.
|
16 |
+
|
17 |
+
We use the same datasets as BERT (English Wikipedia and Book Corpus) to pre-train for 30k steps with a batch size of 8,192. I also released the reproduction of this dataset [on HuggingFace](https://huggingface.co/datasets/JackBAI/bert_pretrain_datasets).
|
18 |
+
|
19 |
+
We utilized DeepSpeed ZeRO-2 for performance optimization.
|
20 |
+
|
21 |
+
Other training configuration:
|
22 |
+
|
23 |
+
| Parameter | Value |
|
24 |
+
|----------------------|-----------|
|
25 |
+
| WARMUP_STEPS | 1800 |
|
26 |
+
| LR_DECAY | linear |
|
27 |
+
| ADAM_EPS | 1e-6 |
|
28 |
+
| ADAM_BETA1 | 0.9 |
|
29 |
+
| ADAM_BETA2 | 0.98 |
|
30 |
+
| ADAM_WEIGHT_DECAY | 0.01 |
|
31 |
+
| PEAK_LR | 1e-3 |
|