Ollama, the framework for running large language models locally, recently released version 0.30.5. This update primarily addresses a "gemma4:12b floating point exception crash" that affected the newly supported Gemma 4:12B model. The fix in v0.30.5 resolves an issue identified in the preceding v0.30.4 release, which itself followed the addition of Gemma 4:12B support in v0.30.3. Alongside this crash resolution, v0.30.4 also included an update to the underlying inference on various hardware, including local CPUs and GPUs.">llama.cpp version, which can bring performance and compatibility improvements for various models. These updates ensure more stable and efficient local inference for users leveraging Gemma models on their hardware.