October 24, 2024: Meta Launches Llama Models for Low-Power Devices - Meta Platforms Inc. has introduced quantized versions of its Llama 3.2 1B and 3B models to enhance AI accessibility on low-powered devices. These smaller models, revealed at the Connect 2024 event, utilize quantization techniques like QLoRA and SpinQuant to optimize performance and portability. Achieving a 56% reduction in model size, they deliver faster on-device inference with maintained accuracy. Collaborations with Qualcomm and MediaTek ensure compatibility with Arm-based hardware. Developers can now create private, AI-driven experiences on resource-constrained devices with these quantized models, available on Llama.com and Hugging Face.
