---
base_model:
- schonsense/70B_llama312_RP_ft
---

- imx_plus quants are experimental full bf16 precision on the entirety of layer 0 as well as the output.
- https://arxiv.org/html/2408.15301v1
- I made a mistake on my 'plus' quanting, and overweighted every tenth layer instead of just layer 0. I will change this note once I have rectified the issue.
- imx quants are normally formulated with a model specific custom imatrix dataset.