Lamini

Lamini enables companies to develop, fine-tune, train, and deploy large language models (LLMs) tailored to their needs. The platform allows businesses to leverage their proprietary data to create customized GenAI capabilities on a secure basis on-premise or via public cloud. The company runs many of its models on AMD Instinct GPUs (MI300s and MI250s) while offering the NVIDIA H100 GPU.

Lamini differentiates itself from competitors via its “memory tuning” technique, which trains models on data to recall parts of that data exactly. The company claims that the technique delivers over 95% accuracy in information recall. 

Lamini generates revenue on a pay-as-you-go scheme (USD 0.5 for 1 million tokens used and USD 1 per tuning step). The company also offers custom pricing for enterprises requiring a dedicated GPU cluster or a virtual private cloud (VPC).

Key customers and partnerships

As of September 2024, notable enterprises using Lamini’s platform included AMD, AngelList, and NordicTrack. As of May 2024, the platform was also used by several undisclosed government agencies.

Key partnerships include Meta, which will improve LLaMa 3’s baseline performance by improving the quality of SQL queries (June 2024); Databricks, which will allow customers to train GenAI models using Databricks data (June 2023); and Mistral, which will allow Lamini customers to fine-tune the Mistral-7B model.