Unsloth AI

Based in California, US, and founded in 2023 by Daniel Han (CEO) and Michael Han (CTO), Unsloth AI Inc. operates as a software company that provides open-source fine-tuning and training of LLMs. In September 2024, Unsloth AI Inc. The company engages in accelerating AI training, offering up to 30x faster performance with reduced memory usage, multi-GPU support, and scalable solutions, making LLM fine-tuning accessible for researchers, developers, and enterprises. The firm uses patented techniques such as manually developing computation-heavy mathematical steps using handwriting GPU kernels to provide training speeds 30 times faster than Flash Attention 2 (FA2) while using 90% less memory. Unsloth supports a wide range of hardware, including NVIDIA, AMD, and Intel GPUs, enabling users to achieve up to 32x faster training on multi-GPU systems. In addition, the company optimizes memory usage, allowing scalable deployment across diverse computing environments. The firm has also developed HyperLearn, a tool for streamlining machine learning workflows. The organization generates revenue by offering three tiers of service: Free, Pro, and Enterprise. The free plan supports basic single-GPU operations and open-source LLMs such as LLama models.5x faster training, and 20% reduced VRAM usage. Enterprise customers benefit from advanced features, including 30x faster training, multi-node support, and up to 30% improved accuracy for AI applications.

1. Seed

   Sep 30, 2024

   Amount

   Undisclosed

   Investors

   Locked