Beyond just merging, we've boosted performance by continuing to pretrain with a massive dataset. Merging alone doesn't cut it for top-notch results.

→ View original post on X — @hunkims