Update README.md
Browse files
README.md
CHANGED
|
@@ -30,4 +30,37 @@ Original model: https://huggingface.co/cognitivecomputations/laserxtral
|
|
| 30 |
|
| 31 |
Credit to Bartowski for help and model card formatting
|
| 32 |
|
| 33 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
|
| 31 |
Credit to Bartowski for help and model card formatting
|
| 32 |
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
## Original Model Card Below
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
by David, Fernando and Eric
|
| 40 |
+
|
| 41 |
+
Sponsored by: [VAGO Solutions](https://vago-solutions.de)
|
| 42 |
+
|
| 43 |
+
Join our Discord! https://discord.gg/vT3sktQ3zb
|
| 44 |
+
|
| 45 |
+
An experimentation regarding 'lasering' each expert to denoise and enhance model capabilities.
|
| 46 |
+
|
| 47 |
+
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).
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
# Laserxtral - 4x7b (all, except for base, lasered using laserRMT)
|
| 51 |
+
|
| 52 |
+
This model is a Mixture of Experts (MoE) made with [mergekit](https://github.com/cg123/mergekit) (mixtral branch). It uses the following base models:
|
| 53 |
+
* [cognitivecomputations/dolphin-2.6-mistral-7b-dpo](https://huggingface.co/cognitivecomputations/dolphin-2.6-mistral-7b-dpo)
|
| 54 |
+
* [mlabonne/Marcoro14-7B-slerp (base)](https://huggingface.co/mlabonne/Marcoro14-7B-slerp)
|
| 55 |
+
* [beowolx/CodeNinja-1.0-OpenChat-7B](https://huggingface.co/beowolx/CodeNinja-1.0-OpenChat-7B)
|
| 56 |
+
* [Q-bert/MetaMath-Cybertron-Starling](https://huggingface.co/Q-bert/MetaMath-Cybertron-Starling)
|
| 57 |
+
* [WizardLM/WizardMath-7B-V1.1](https://huggingface.co/WizardLM/WizardMath-7B-V1.1)
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
It follows the implementation of laserRMT @ https://github.com/cognitivecomputations/laserRMT
|
| 61 |
+
|
| 62 |
+
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.
|
| 63 |
+
|
| 64 |
+
We intend to be the first of a family of experimentations being carried out @ Cognitive Computations.
|
| 65 |
+
|
| 66 |
+
In this experiment we have observed very high truthfulness and high reasoning capabilities.
|