fix model naming schemes for weight decay

[rhubarbwu]

There are several confusing model naming schemes. The most confusing is currently for weight decay.

• $\beta = 0$: TinyStories-01x0064_01n
• $\beta = 0.0005$: TinyStories-01x0064_01d
• $\beta = 0.1$: TinyStories-01x0064_01L

This is inconsistent across different $\beta$ values and also confusing with the $d$ and $L$ in referring to model scale.

We will likely have to write a script that auto-clones all of the models to update them. Then we'll have to correct our experimental scripts and artifacts themselves.