Hacker News new | ask | show | jobs
by euclaise 988 days 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.