12 ms·
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.
by euclaise 3y ago
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.