By: Adam Sharp
The HIVE Newsletter
Since the public release of ChatGPT in November of 2022, the world has been captivated by large language models (LLMs). OpenAI owns ChatGPT, yet it is a “closed” model, meaning the code is private and proprietary.
Inevitably, we’ve started to see similar models released under open-source software licenses. A model that is “open- source” can be freely used and customized by anyone, even for commercial purposes.
In the past few months, we’ve begun to see a flood of powerful open-source LLMs, including Meta’s Llama-2, Falcon, and Mistral 7B.
Although these open-source models are new, they are quickly approaching ChatGPT in terms of capability. To demonstrate their potential, I asked Llama-2 70B to explain why open-source LLMs are a disruptive force.
As shown above , Llama-2 already provides remarkably high-quality answers. Now that this open-source code is available for anyone to use and customize, the AI industry is about to experience a period of extreme innovation.
How Open-Source LLMs Benefit Private Businesses
Let’s look at a real example of how enterprises might use open-source LLMs. A good friend of mine is an accountant. His firm took an open-source LLM and “fine-tuned” it using hundreds of thousands of emails between accountants and clients. After significant training and tinkering, this open-source model could correctly and professionally answer client questions on almost any topic.
After integrating this open-source model into the firm’s email, now, whenever an accountant gets an email from a client, the model scans it and automatically provides a draft response. The accountant reviews the response, makes any necessary changes, and then sends it to the client. The accounting firm successfully created their own specialized large language model.
This example gives us a glimpse into the massive potential of open-source LLMs in the workplace. The technology is nascent but already incredibly useful. At HIVE, we believe that this technology has a role to play within almost every company and industry.
Announcing HIVE’s Open-Source GPU Grant Program
The HIVE team is busy converting our fleet of 38,000 industrial GPUs into on-demand cloud infrastructure (see our recent press release for details).
We are particularly interested in cutting-edge generative AI technologies such as LLMs. To support and better understand this market, HIVE has launched a GPU grant program for open-source AI developers.
In this initial program, we’re giving away 1,000 hours on a powerful Supermicro server with 10x data center GPUs with 48 GB of VRAM each (an estimated value of $10,000). This hardware offers robust capabilities for running and modifying LLMs, similar to Llama-2 and Mistral 7B.
HIVE invites all open-source AI developers to apply for our GPU grant program by October 31st, 2023, for the chance to win. All winner(s) will be announced by November 3rd, 2023 and granted access to these servers by November 6th, 2023. For more information on how to apply, visit the official contest page.
HIVE is excited to offer this contest to support an important field of research. Our team is excited to learn what modern AI developers want and need from their GPU cloud providers, to better service our clients and the industry.
Cheers,
Adam Sharp
The HIVE Newsletter