---
license: cc-by-nc-2.0
---

An experimentation regarding 'lasering' each expert to denoise and enhance model capabilities.

This follows the implementation of laserRMT @ https://github.com/cognitivecomputations/laserRMT

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.

We are looking forward to measure its evaluation over the most important benchmarks.