Range – open a 1 TB AI model in 3 seconds without downloading it (getrange.sh)

Range has launched a revolutionary tool that allows users to access AI models, such as Hugging Face repositories or S3 environments, instantly—without the need for downloading. This innovative approach enables users to open large models, even up to 1 TB, and utilize only the data their programs require, significantly reducing the time and bandwidth typically needed for deployment. For instance, accessing the 1.03 TB Kimi K2 model took just 3.4 seconds to read a portion, with subsequent access being nearly instantaneous as the system leverages indexed storage to fetch necessary bytes on demand. This breakthrough is particularly significant for the AI/ML community as it addresses common issues related to the processing and management of massive models, streamlining workflows and enhancing efficiency for developers and researchers. By transforming how environments are accessed—turning them into “disks” from which only specific data is transferred—Range minimizes latency and network load, making it easier and faster to experiment with advanced AI capabilities. This could potentially accelerate model training and deployment times, promoting innovation and collaboration across various AI applications.