According to @JeffDean, the Gemma 3 27B model, previously only runnable on a single cloud H100 GPU, has been optimized with Quantization Aware Training (QAT) to fit on a home GPU like the RTX 3090. This breakthrough enables traders and developers to leverage the model's advanced capabilities directly from their desktops, potentially impacting algorithmic trading efficiency and data analysis. The optimization details and setup guidance are available via @ollama.