HOW TO IMPROVE 24 POINTS IN MMLU WITH OLMO

Summary from the TLDR AI newsletter

The latest completely open 7B language model had an upgrade with a much-improved reasoning score on MMLU. Its development team found that changing the data mixture had a substantial effect on performance. They have provided the exact statistics of new data sources and percentages that led to this improvement.