Update README.md

README.md

@@ -62,7 +62,7 @@ The rest of this is going to go over both my rationalization for the approach, i
 The model selection was done using an evolutionary approach where each generation added or removed a model and the outcomes were tested by (me) without the use of any automation or benchmarks.
 
 ## Why are these models candidates for selection? 
-The long story short, there's few choices for large trained models that are very different from eachother. Almost all RP models use similar datasets, and often exclusively train on a small distribution of specific data. This often means they differ numerically only very little from their root model, which makes merging less impactful. For this reason, many common roleplay models have been avoided as candidates, and I've only selected a single Euryale 2.2 from Sao. I did try two others, and the merge was significantly worse. This leaves the actual selection, which was the most varied models available in the limited selection, and let the evolutionary process decide the rest.
+The long story short, there's few choices for large trained models that are very different from eachother. Almost all RP models use similar datasets, and often exclusively train on a small distribution of specific data. This often means they differ numerically only very little from their root model, which makes merging less impactful. This leaves the actual selection, which was the most varied models available in the limited selection, and let the evolutionary process decide the rest.
 
 ## Base Models VS Instruct Tunes
 A very important point that is often missed is that instruct is often where the moralizing, dehumanization, and overcompliance comes in. This is not that surpising, instruct is designed to be compliant, and the nature of instruct tunes leads naturally to strong biases in the models vocabulary and personality. In this way, the goal was to use the base model merges like youko and the original llama 3 70B base as regularization to pull back from some of these learned habbits of the instruct models, which are not present (Or perhaps I should say, as prominent) in the base models.