Ollama v0.31.2-rc2 Enables iGPU MMProj Offload with Fit Padding

Core summary

One sentence to understand this update

Ollama v0.31.2-rc2 now allows integrated GPU (iGPU) offloading for multimodal projectors with "fit padding," improving efficiency by better handling text-model placement.

Impact & opportunity

What this could mean

This update could significantly boost performance for local LLMs on systems with iGPUs, opening up more accessible hardware options for running multimodal models.