Commit
•
2e299be
1
Parent(s):
eb6a982
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
DeBERTa trained from scratch
|
| 2 |
+
|
| 3 |
+
continued training from https://huggingface.co/mikesong724/deberta-wiki-2006
|
| 4 |
+
|
| 5 |
+
Source data: https://dumps.wikimedia.org/archive/2010/
|
| 6 |
+
|
| 7 |
+
Tools used: https://github.com/mikesong724/Point-in-Time-Language-Model
|
| 8 |
+
|
| 9 |
+
2010 wiki archive 6.1 GB trained 18 epochs = 108GB + 2006 (65GB)
|
| 10 |
+
|
| 11 |
+
GLUE benchmark
|
| 12 |
+
|
| 13 |
+
cola (3e): matthews corr: 0.3640
|
| 14 |
+
|
| 15 |
+
sst2 (3e): acc: 0.9106
|
| 16 |
+
|
| 17 |
+
mrpc (5e): F1: 0.8505, acc: 0.7794
|
| 18 |
+
|
| 19 |
+
stsb (3e): pearson: 0.8339, spearman: 0.8312
|
| 20 |
+
|
| 21 |
+
qqp (3e): acc: 0.8965, F1: 0.8604
|
| 22 |
+
|
| 23 |
+
mnli (3e): acc_mm: 0.8023
|
| 24 |
+
|
| 25 |
+
qnli (3e): acc: 0.8889
|
| 26 |
+
|
| 27 |
+
rte (3e): acc: 0.5271
|
| 28 |
+
|
| 29 |
+
wnli (5e): acc: 0.3380
|