Update README.md
Browse files
README.md
CHANGED
|
@@ -18,6 +18,22 @@ Success is a game of winners.
|
|
| 18 |
|
| 19 |
# The Human AI .
|
| 20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
# Deep thinking Model - Highly Trained om Multiple Datasets
|
| 23 |
|
|
|
|
| 18 |
|
| 19 |
# The Human AI .
|
| 20 |
|
| 21 |
+
# PERSONAL NOTE :
|
| 22 |
+
|
| 23 |
+
Problem is maths ! When maths goes up Musr Goes down even bbh goes down :
|
| 24 |
+
So i created the teacher model ( top Teacher ) problems is Merges dont always goe exactly to the calculations ! the tey acher model does have greater maths but it also went down !
|
| 25 |
+
SO! -- I have the batch training i done on that model , ie the scripts for the training session setup ( no Merging loras ! also not great ideas in the end !)
|
| 26 |
+
SO : I will retrain this model on Rank 64 + the new maths scripts ! And benchmark leader board again !
|
| 27 |
+
Currently this is the top model for mistral models ( 7B class )
|
| 28 |
+
------>>>> Lamma 8b is higher (only a few) and Qwens are higher !!! They are the current highest model for open source right now !
|
| 29 |
+
|
| 30 |
+
I only am competing with the 7b models !
|
| 31 |
+
|
| 32 |
+
a 1.5b model is top of the maths benchmark !
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
|
| 37 |
|
| 38 |
# Deep thinking Model - Highly Trained om Multiple Datasets
|
| 39 |
|