Predibase, the leading development platform for LLM (large language model) fine-tuning, has launched LoRa Land, a collection of 25 open-source fine-tuning models that the company claims can challenge or even surpass the very popular OpenAI's GPT-40。
LoRa Land is powered by Predibase's serverless fine-tuning endpoints and the open-source LoRax framework. The new platform offers a wide range of use cases, from sentiment analysis to summarization.
GPT-4 is one of the most widely used LLMs in the world, and it was a daunting challenge for LoRa Land to surpass it. However, Predibase seems to be confident in the capabilities of its latest product. The company claims that LoRa Land offers a more cost-effective way for organizations to train highly accurate and professional GenAI (generative artificial intelligence) applications.
Due to the prohibitively high cost of building GPT models from scratch and fine-tuning LLMS, using specialized LLMS is becoming a popular alternative, and this is exactly where Predibase may position LoRa Land in the competitive landscape.
Using smaller and more specialized LLMS, developers leverage techniques such as parameter-efficient fine-tuning and low-level adaptation to create high-performance AI applications to reduce the cost of fine-tuning LLMS. Predibase says it has integrated these technologies into its own platform, giving users the option to choose the LLM that best suits their application and fine-tune it accordingly.
Traditionally, one of the reasons why fine-tuned LLMs have been so expensive to put into production is that they require a dedicated GPU for each model. For users who need to deploy LLMs to address a variety of use cases, GPU fees add up to be a major barrier to growth and innovation. While the initial experiment of accessing LLMs via APIs is relatively inexpensive, the cost can quickly rise when the deployment is fully implemented.
From a resource perspective, fine-tuning open source LLMs is not only costly, but also has a major problem of a lack of AI skills, which is one of the main barriers to AI adoption.
Predibase overcomes cost challenges by designing LoRa Land to serve multiple fine-tuned LLMs on a single GPU. According to Predibase, 25 LLMs in LoRa Land are capable of fine-tuning with an average GPU cost of less than $8. Not only is this cheaper, but users also don't have to wait for the GPU warm-up to boot up before fine-tuning each model. Other benefits offered by LoRa Land include a highly scalable infrastructure model and instant deployment and prompting.
Dev Rishi, co-founder and CEO of Predibase, said, "Organizations are increasingly recognizing the benefits of having many smaller, more granular models for different use cases and customers. "According to internal data, 65% of organizations surveyed plan to deploy two or more fine-tuned LLMs in the next 12 months, and 18% plan to deploy six or more. ”
The efficiency and affordability offered by LoRa Land level the playing field for smaller companies in the AI race. This is good news for companies that want to deploy a wide range of specialized LLMs to drive their business forward.
The introduction of new platforms not only provides technological innovation, but also has the potential to change the landscape of AI development. Predibase aspires to set a new standard for the industry with its high performance, cost-effectiveness, and accessibility.