[euclaise]

There's a new 7B version that was trained on more tokens, with longer context, and there's now a 14B version that competes with Llama 34B in some benchmarks.