Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- athirdpath/Merge_Glue
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
### TeeZee/NEBULA-XB-v1.0_SFT_2_epoch ###
|
| 8 |
+
|
| 9 |
+
Experiment, can DUS be taken one or more steps further?
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
### Technical notes:
|
| 13 |
+
- pretrained model NEBULA-XB-v1.0 finetuned on 30k entries from Merge_Glue dataset
|
| 14 |
+
- 18 layers removed from both models of finetuned GALAXY-XB-v03
|
| 15 |
+
- model has 108 layers (((48-12)*2)-18)*2 = 108
|
| 16 |
+
- second step in scaling DUS procedure
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
### To evaluate
|
| 20 |
+
- model performance after merge, should be a little lover that GALAXY finetuned on 50k of slimorca
|