Report
Apertus 1.5 technical report is out, with evaluations claimed to be reproducible
A post lists 48 benchmarks for the 8B model and 47 for the 70B, inviting readers to rerun them.
TLDR
A post sharing the Apertus 1.5 technical report claims it is the first model release with every evaluation on Evaluation Cards and HF Community Evals, reproducible down to the generation settings. It lists 48 benchmarks for the 8B model and 47 for the 70B, and invites readers to rerun them.
Combined views
872
2 Sources, first seen ago
