Llama 3.2 1B Instruct

Amazon Web Services published a launch post on September 25, 2024, announcing availability of Meta's Llama 3.2 models in Amazon SageMaker JumpStart. The post falls within AWS's Machine Learning blog and covers the new family, including the small text models and the multimodal Vision variants intended for visual understanding workloads. The page is the official AWS announcement confirming day-one support for the 1B Instruct variant on a managed developer platform. By distributing the models through JumpStart, AWS provides developers a managed path to deploy and experiment with Llama 3.2 without standing up their own inference infrastructure, complementing Meta's direct release on the same date. The blog post is the primary AWS-served record of integration with SageMaker for the Llama 3.2 line, including the 1B Instruct target model.

Meta's Llama 3.2 1B Instruct is available on NVIDIA NGC as part of the Llama 3.2 multilingual LLM family released on September 25, 2024. It is a pre-trained and instruction-tuned generative model in the 1B and 3B text-in/text-out size range, optimized for multilingual dialogue use cases including agentic retrieval and summarization tasks. The model uses an optimized transformer architecture and is instruction-tuned with supervised fine-tuning (SFT) plus reinforcement learning from human feedback (RLHF) to align with human preferences for helpfulness and safety. It was developed from the Llama-3.2-1B base model and is accelerated by TensorRT-LLM for inference on NVIDIA GPUs, with commercial use permitted under the Llama 3.2 Community License.

Fireworks AI published a partner blog on September 25, 2024, announcing support for Meta's Llama 3.2 release across serverless on-demand, on-demand, and enterprise reserved deployment modes. The post explicitly names the 1B and 3B text-only Instruct variants, along with the 11B and 90B Vision models, and highlights fine-tuning availability. It positions the 1B as ideal for retrieval, summarization, multilingual knowledge retrieval, and rewriting tasks. The blog frames Llama 3.2 as a building block for compound AI systems that combine text and multimodal models with external tools, and notes that Meta's Llama Guard 3 safety models are also available on-demand. Fireworks supports both instruct and non-instruct variants on private GPU instances billed per GPU-second, giving developers flexible serving options for the 1B target model on launch day.

Databricks announced a partnership with Meta to launch the Llama 3.2 series on its Data + AI Platform, making the 1B and 3B Instruct text models available alongside the new 11B and 90B Vision-Instruct multimodal variants. The post, attributed to Ahmad Al-Dahle, Head of GenAI at Meta, positions the smaller text models for low-latency, low-cost enterprise tasks such as entity extraction, multilingual translation, summarization, and retrieval-augmented generation. Deployment is offered through Databricks' Agent Bricks AI Playground, with tuning via Model Training and orchestration through Agent Framework. The release marks the first time Llama models gain visual understanding via the 11B and 90B Vision-Instruct variants, paired with a new Llama-Guard-3-11B-Vision safety model for responsible multimodal deployment. Databricks cites 10x year-over-year growth for Llama adoption on its platform as motivation for the integration. The announcement frames the 1B and 3B models as purpose-built, fast, and cheap alternatives for narrow enterprise tasks after on-data tuning.

Meta released Llama 3.2 1B Instruct on September 25, 2024, with the model card published on Hugging Face under the official meta-llama organization. The page identifies the release version as "Llama 3.2" and presents the Llama 3.2 Community License Agreement, which grants a non-exclusive, worldwide, non-transferable, royalty-free limited license to use, reproduce, distribute, and create derivative works of the Llama Materials. Distribution is governed by Meta Platforms Ireland Limited or Meta Platforms, Inc. depending on licensee location. The Hugging Face model card is the canonical source for the variant's attribution, version stamp, and licensing terms, anchoring the model within the Llama 3.2 family released that day. The card also links to Meta's documentation at llama.meta.com/doc/overview and the download portal at llama.com/llama-downloads for the underlying weights and materials. Licensees must accept Meta's privacy and acceptable-use terms to access the gated artifacts.