Monster API uses generative AI to enable anyone to build generative AI

A startup called Monster API (officially known as Generative Cloud Inc.) features a GPT-based agent that simplifies and accelerates the fine-tuning and deployment of open-source generative artificial intelligence models.

According to the company, MonsterGPT helps reduce the time it takes to fine-tune and implement generative AI applications based on open source models such as Llama 3 and Mistral to just 10 minutes.

Open source generative AI models like Llama 3 have become very popular among companies looking for alternatives to expensive proprietary models created by OpenAI and Google LLC. However, the process of creating applications such as conversational assistants is still very difficult.

For example, if a company is looking to build a customer service bot that can handle common issues, the open source model used by the company can be based on the company's own data and knowledge base to learn about different products. Need to be fine-tuned. , the services and policies necessary to inform that response.

This is not easy to do and requires rigorous execution by developers and data scientists to adjust as many as 30 variables. This requires both knowledge of advanced AI optimization frameworks and an understanding of the underlying infrastructure such as graphics processing unit-based cloud steps, containerization, and Kubernetes.

According to Monster API, fine-tuning AI models is often a huge undertaking for companies, involving as many as 10 expert engineers and requiring dozens of hours of work. This assumes that the company actually has employees with the required skill set. There are many that are not.

Create generative AI using generative AI

It is this process that the Monster API aims to solve. We decided that the best approach to help fine-tune and customize generative AI models was to look at generative AI itself. Using that API, developers can simply issue a command like “tweak Llama 3” and the Monster API will start working.

“For the first time, we offer a solution based on an agent-driven approach for generative AI,” Saurabh Vij, CEO of Monster API, told SiliconANGLE. “The ease and speed of this process is like flying from New York to London in 90 minutes in a Mach 4 supersonic jet. At the end of this incredibly fast process, MonsterGPT provides developers with custom, fine-tuned Provides API endpoints for models.

Compared to traditional AI development, the Monster API enables a more “use case” oriented approach, allowing users to simply specify the task they want the model to accomplish, such as sentiment analysis or code generation, and create the optimal model. Masu. to perform that task.

Vij said there will be a lot of interest in the service, as the majority of small teams, startups and indie developers are not particularly familiar with the art of fine-tuning and deploying AI models. “Most developers aren't experts in the deeper nuances of how different models and architectures work, and they don't have experience working with complex cloud or GPU infrastructure,” he said. .

Vij said he envisions a future world where almost everyone can become a programmer. That's because you don't need any coding expertise; you can just use natural language to instruct the generative AI to write code. “All of our research and design is aimed at accelerating faster towards this future,” he added.

Open source AI vs. closed source AI

The Monster API is an API-based approach to generative AI development that has been linked to historical developments such as the first Macintosh computers, which introduced the concept of personal computing in the 1980s, and Mosaic, the first easy-to-use web browser that democratized We believe it reflects the advancement of technology. It's about making the Internet accessible to everyone.

The company wants to democratize access to generative AI development in a similar way, which requires a focus on open-source models rather than closed-source alternatives.

Bisi said the conflict between open source and closed source AI has historical precedent in the form of the battle for mobile supremacy between Apple Inc. and Google LLC's open source Android over the past decade. He said that there is. “Just as Android offers a flexible alternative to Apple's tightly controlled ecosystem, it also offers a competitive advantage to proprietary giants like OpenAI's GPT-4 for powering open source AI models. “There is a concerted effort,” he said.

Vij believes the Monster API will help advance open source AI by making it easier for even non-skilled people to build AI applications. These users will almost certainly choose open source models over proprietary versions, he said, as proprietary versions such as GPT-4 tend to be general rather than specialized.

He explained that most businesses require domain-specific AI models, so the basic model needs to be fine-tuned. However, he said, the fine-tuning process in a closed-source model is often limited to a few technologies provided by the vendor itself and can be very expensive.

“In the open source world, developers are free to experiment with multiple frameworks from Q-LORA to DPO,” Vij added. “You can try out different technologies and choose the one that best fits your use case and budget.”

Monster API's agents are said to be able to leverage powerful AI frameworks such as Q-LORA to enable fine-tuning and leverage vLLM to deploy customized models. It provides a simple, unified interface that covers the entire development lifecycle, from initial tweaks to deployment. “MonsterGPT saves you time and effort to start experimenting and fine-tuning your deployments,” he says Vij. “Users will now have more options in terms of optimization algorithms they can use to significantly improve throughput and latency.”

Another benefit of MonsterGPT is that users do not need to learn about the setup, GPU resources, and configurations of the many cloud infrastructures available. Because it automatically selects the best infrastructure based on your budget and goals.

Vij promised that this will be particularly beneficial as smaller, fine-tuned models are sufficient for many use cases and applications. For example, if a company is using his GPT-4 Turbo, OpenAI's most powerful model, for a simple customer service chatbot, its capabilities go far beyond that kind of use case. That's probably overkill. By using smaller models optimized for their applications, businesses can enjoy significant cost savings.

“The smaller models can fit into smaller, more affordable general-purpose GPUs,” Vij explained, adding that this is another benefit of open source. “A closed-source model leaves developers unable to do anything and forces them to use larger, more generalized models.”

Constellation Research Inc.'s Holger Mueller said he is impressed with the MonsterGPT concept, which represents a major advance in the recursive use of generative AI. “Just as we are already using robots to build more robots, we are now using AI software to create more AI software,” he said. Ta.

Analysts noted that use cases like this require some form of oversight, so it's important for humans to stay in the loop. However, he said the technology is very promising in its ability to lower barriers to entry to generative AI development.

“The goal of the Monster API is to enable anyone who can speak and validate AI models to build and deploy them without any special skills,” Mueller explained. “As with any new technology, especially AI, you should apply a healthy level of skepticism until you see practical use cases in production. But for now, the Monster API… We applaud their efforts to democratize AI development and change the future of work for developers and business users without prior AI experience.”

At the time of release, the Monster API supports over 30 popular open source LLMs, including Meta Platforms Inc.'s Llama 3, Mistral's Mistral-7B-Instruct-v0.2, Microsoft Corp's Microsoft/phi-2, and Stability AI Ltd. is supported. ■SDXL base.

Image: Microsoft Designer