Ollama supports quantization compression technologies such as 4-bit and 8-bit, reducing VRAM usage and computation by lowering the numerical precision of model weights.