Llama-3.2-3B

Llama-3.2-3B is a foundational large language model released by Meta on September 25, 2024, as part of the broader Llama 3.2 family of models and trained weights distributed through Meta's official channels. The model is published under the Llama 3.2 Community License Agreement, which governs use, reproduction, distribution, and modification of the Llama Materials and accompanying Documentation hosted at the Meta documentation portal. Access to the official weights on the meta-llama Hugging Face organization is gated, requiring users to agree to share contact information in accordance with Meta's Privacy Policy before downloading the model.

Beyond the base 3B model, an instruction-tuned variant called Llama-3.2-3B-Instruct is also distributed within the same family, reflecting the common Llama practice of pairing a pretrained foundational model with a fine-tuned conversational counterpart. A third-party Kaggle mirror distributes this Instruct variant as a PyTorch model that is fine-tunable, suggesting practical adaptability for downstream tasks. Users seeking to evaluate or build on Llama-3.2-3B should obtain the weights through Meta's official gated distribution to ensure compliance with the community license terms.