Welcome to the Runpod Roundup for the week ending August 19, 2023. In this issue, we'll be discussing the newest advancements in AI models over the past week, with a focus on new offerings that you can run in a RunPod instance right this second. In this issue, we'll be looking at large language model evaluation, vector search, and 3D scene reconstruction.
Arthur Releases Open-Source LLM Evaluator
Large language models are a prime use case of RunPod, but there's one problem: there's just so many of them. It takes time to evaluate whether a model is right for you – and not only that, even individual quantizations can vary widely as to their performance. Arthur has released a new open source model evaluator called Bench that helps you evaluate specific models for whatever your use case might be. You can also use it to test whether a specific open-source LLM might do as well or even better compared to closed-source offerings based on your use case. To get started using Bench, you can go ahead and grab it straight from their Github.
(In the market to try a new LLM, by the way? Check out the LLM Explorer.)
NVidia Releases Neuralangelo To Reconstruct 3D Scenes from Photographs
NVidia has released Neuralangelo, which utilizes 3D hash grids and neural surface rendering from the information collected from still photographs. There are a number of potential use cases for these reconstructed scenes, such as porting them to game or VR engines. It appears that the package can even work with video footage as well, which would provide an even more detailed result from the additional depth information presented from the camera movement. You can grab the repo from the official NVLabs Github.
Marqo Offers GPU-Enabled Open Source Vector Search
Vector search is an emerging machine learning technique that leverages ML to analyze and categorize unstructured data. The beauty of vector search is that you do not have to annotate your data to use it effectively – although it certainly helps, vector search also can search for similarities that it finds in text, images, and video that go beyond the need for any human annotation. Marqo (polo?) is a tensor search that you can run right in your RunPod instance in concert with any other solutions you might be using.
Feel free to reach out to RunPod directly if you have any questions about these latest developments, and we'll see what we can do for you!