Apache 2.0 Archives - BonTech Labs

Line graph showing total tokens per second vs. batch size on a single h100. IBM Granite 4.0 rises sharply, outperforming OpenL-30B and OpenL-32B, both of which drop at higher batch sizes. RAM usage is indicated by shaded error bands.

IBM Granite 4.0 goes hybrid to cut RAM and cost

October 3, 2025 Gavin Bonshor 0

IBM has unveiled Granite 4.0, a new generation of open language models that blend Mamba-2 state-space layers with Transformer attention blocks to reduce memory requirements […]