Search papers, labs, and topics across Lattice.
1
0
3
4
By reusing existing data mixture ratios and only recomputing for affected domains, Olmix slashes compute costs by 74% without sacrificing downstream task performance during iterative LM development.
