Predibase has just announced a new collection of fine-tuned LLMs in a package called LoRA country. LoRA Land contains just over 25 LLMs that are optimized for specific purposes and perform as well or better than GPT-4.
Examples of available models include code generation, customer support automation, SQL generation, and more.
According to the company, many organizations are starting to realize the benefits of having smaller models that work really well for one specific purpose. Our own research shows that 65% of organizations plan to implement two or more fine-tuned LLMs in the next 12 months.
Predibase believes that the release of LoRA Land will help companies implement these models without paying to build them from scratch, which is often not feasible in many companies’ budgets.
There are also savings in actually running fine-tuned LLMs as they would typically require a dedicated GPU for each model, while multiple LoRA Land models can be run from a single GPU. This is because the models are developed and trained using the company’s open source code LoRAX a project that allows running up to 100 fine-tuned models from a single GPU.
“We built LoRA Land to provide a real-world example of how smaller, fine-tuned, task-specific models can cost-effectively outperform leading commercial alternatives. Predibase makes it much faster and more resource-efficient for organizations to fine-tune and serve their own LLMs, and we’re happy to provide the tools and infrastructure for teams looking to start implementing specialized LLMs to power their business,” the company wrote in a statement. blog post.