Commit
•
e6c5ece
1
Parent(s):
df30192
Update README.md
Browse files
README.md
CHANGED
|
@@ -8,8 +8,17 @@ An experimentation regarding 'lasering' each expert to denoise and enhance model
|
|
| 8 |
|
| 9 |
This model has half size in comparison to the Mixtral 8x7b Instruct. And it basically has the same level of performance (we are working to get a better MMLU score).
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
It follows the implementation of laserRMT @ https://github.com/cognitivecomputations/laserRMT
|
| 12 |
|
| 13 |
Here, we are controlling layers checking which ones have lower signal to noise ratios (which are more subject to noise), to apply Laser interventions, still using Machenko Pastur to calculate this ratio.
|
| 14 |
|
| 15 |
We intend to be the first of a family of experimentations being carried out @ Cognitive Computations.
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
This model has half size in comparison to the Mixtral 8x7b Instruct. And it basically has the same level of performance (we are working to get a better MMLU score).
|
| 10 |
|
| 11 |
+
Used models (all lasered using laserRMT):
|
| 12 |
+
cognitivecomputations/dolphin-2.6-mistral-7b-dpo
|
| 13 |
+
mlabonne/Marcoro14-7B-slerp
|
| 14 |
+
beowolx/CodeNinja-1.0-OpenChat-7B
|
| 15 |
+
Q-bert/MetaMath-Cybertron-Starling
|
| 16 |
+
WizardLM/WizardMath-7B-V1.1
|
| 17 |
+
|
| 18 |
It follows the implementation of laserRMT @ https://github.com/cognitivecomputations/laserRMT
|
| 19 |
|
| 20 |
Here, we are controlling layers checking which ones have lower signal to noise ratios (which are more subject to noise), to apply Laser interventions, still using Machenko Pastur to calculate this ratio.
|
| 21 |
|
| 22 |
We intend to be the first of a family of experimentations being carried out @ Cognitive Computations.
|
| 23 |
+
|
| 24 |
+
In this experiment we have observe very high truthfulness and high reasoning capabilities.
|