Quantifying the Capabilities of LLMs across Scale and Precision