Commit
•
30971b3
1
Parent(s):
9f37d2b
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,22 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: llama2
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama2
|
| 3 |
+
datasets:
|
| 4 |
+
- pkupie/mc2_corpus
|
| 5 |
+
- togethercomputer/RedPajama-Data-1T
|
| 6 |
+
language:
|
| 7 |
+
- en
|
| 8 |
+
- mn
|
| 9 |
+
base_model:
|
| 10 |
+
- meta-llama/Llama-2-7b-hf
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
A continually pre-trained model based on Llama-2-7b-hf.
|
| 14 |
+
|
| 15 |
+
We use the **Traditional Mongolian texts** in MC^2 and **English texts** in RedPajama with a proportion of **4:1** for training.
|
| 16 |
+
|
| 17 |
+
#### Hyper-parameters:
|
| 18 |
+
* lr: 3e-5
|
| 19 |
+
* batch size: 1M (2K*512)
|
| 20 |
+
* lr scheduler: cosine
|
| 21 |
+
* min lr: 1e-6
|
| 22 |
+
* lr decay iters: 10240
|