Favorite Social Media Timesink
When you take a break from work, where are you going?
Video clips on TikTok/YouTube
X, Bluesky, Mastodon et al...
Web surfing
I do not get distracted by petty amusements
AI / Large Language Models

Monster API Delivers Advanced Machine Learning to the People

The platform serves up on-demand GPUs via distributed computing, APIs for AI models, and a no-code solution for fine-tuning models.
Jun 22nd, 2023 10:37am by
Featued image for: Monster API Delivers Advanced Machine Learning to the People

Monster API recently released its eponymous platform that’s designed to democratize the use of some of the most advanced machine learning models — and techniques — currently available.

The solution provides on-demand access to a pool of GPUs via a globally distributed computing network. That fact alone significantly lowers the barrier to entry (and costs) for training, refining, and deploying advanced statistical AI models, including Large Language Models (LLMs).

According to Monster API co-founder Gaurav Vij, the Monster API platform aims to “democratize AI models in general because the cost of GPU computing is prohibitively expensive. So far, it’s been restricted to large businesses that can afford it in the cloud. Our approach helps developers and machine learning engineers get access to GPU computing at a fraction of the cost.”

Underpinned by an array of GPUs at the disposal of the platform, Monster API provides APIs to some of the most useful advanced machine learning models for applications of natural language technologies and computer vision. The platform also includes specific pipelines for training models and capabilities for fine-tuning their hyper-parameters.

The optimization techniques involved, paired with inexpensive accessibility to a slew of GPUs, greatly decrease traditional inhibitors for developers utilizing these models. The result is “democratizing access to AI models and also the compute which is underneath those models,” commented Saurabh Vij, Monster API CEO.

GPUs On-Demand

The foundation of Monster API’s platform is its decentralized computing network of GPUs furnished by hundreds of data centers and individuals around the world. Consequently, customers can access as many GPUs as they require from regions in Europe, the United States, India, and others. The network involves more than 30,000 GPUs of numerous varieties, from conventional ones for gaming to those that are optimal for machine learning. “Just like AirBnB has access to underutilized and idle rooms and Uber does that for cars, we’re doing that for GPUs,” Saurabh revealed. Monster API automatically scales to handle workloads so developers can spin up GPU resources when needed.

Geographical concerns about where resources are spun up are abstracted from the users by an orchestrater, which handles this task. Commercial GPUs designed for gaming have been primed for AI models via a “package which adds in containers, GPU drivers, libraries, and more than 10 different frameworks so that they can then work for Machine Learning,” Saurabh explained. The platform itself is container-native and runs atop Kubernetes. Security measures involve five different encryption protocols, a variety of access control models, and container process isolation and data level isolation “so all the processes and data volumes inside the container are isolated from other server level processes to ensure there is additional security,” Gauruv maintained.

Impressive Optimization

The platform’s APIs let users access a plethora of cutting-edge AI models, including open source options like Whisper AI, Stable Diffusion, and more. In addition to optimizing consumer GPUs for machine learning, the platform involves optimization techniques for specific models to make training and deploying them markedly less expensive. For a specific use case involving Whisper AI for translation and speech-to-text transcription, “we increased the throughput and decreased the memory footprint [of the model],” Gauruv revealed.

According to Gaurav, using Whisper AI out of the box for a translation and speech-to-text transcription job on AWS would cost approximately $45,000. Using Monster API’s optimization methods, that same job would cost less than $3,000. “The cost reduction is multiplied because the model is also optimized, due to which the time required to process the request reduces,” Gaurav pointed out. “And, the cost of the GPU infrastructure is already low.”

Fine-Tuning Models

Monster API also addresses the notion of model fine-tuning, which is another aspect of data science that can become costly if left unchecked. The platform incorporates a no-code fine-tuning solution, which involves users accessing machine learning models and datasets used for tailoring models to an individual’s use case.

“You can take a pre-trained foundational model; you can take datasets from free datasets like Hugging Face and quickly start fine-tuning these foundational models for your custom dataset,” Saurabh explained.  “You can do it for under 30 to 40 dollars instead of hundreds of dollars, which you otherwise could spend on fine-tuning these models.”

Beyond Cost Reductions

The decrease in costs for implementing advanced machine learning with Monster API is remarkable. However, the platform does more than simply reduce the overhead for such work. The cost benefits of its distributed GPU network, optimization approaches, and model fine-tuning coincide with the overall accessibility it provides for quickly obtaining and curating these models for enterprise applications of choice.

As such, it has consequences for garage developers as well as C-level executives, both of whom can potentially benefit from it.

Group Created with Sketch.
THE NEW STACK UPDATE A newsletter digest of the week’s most important stories & analyses.