Generative AI has opened up a whole world of possibilities for us, and its integration into various professional applications and tools has completely changed the way we create, collaborate, share and work. NVIDIA AI Workbench is one of the best examples in this sense because it allows us to do this simplify work processes and create our own RAG projectseven if we don’t have advanced knowledge.

If you missed it, don’t worry, RAG stands for “retrieval-augmented generation,” a technique that improves the accuracy and reliability of generative artificial intelligence models using materials and facts supplied from external sources. With it we can fill an important gap that has affected LLMs since their inception, the need to keep them updated and properly “fed” to preserve their ability to understand if we want to delve deeper into certain topics.

For example, imagine that you want to use generative artificial intelligence to create specific images of a product that you have in your head, but that artificial intelligence is unable to provide a satisfactory result because it has never seen anything like it and It does not have the necessary foundation to achieve relevant results. With RAG, we can overcome this obstacle and provide AI with everything it needs to produce satisfactory results.

All of this represented a workload that could have been quite complex, but was greatly simplified thanks to the NVIDIA AI Workbench. With it we can create our own RAG projects and it will help us develop, experiment, test and use various artificial intelligence applications at the prototype stage on a variety of GPUs, including everything from notebook models to more powerful data center solutions. Yes, it also works in the cloud.

NVIDIA AI Workbench: ready in minutes and it’s free

The installation process is very simple and once it is done we are ready to start working locally and remotely. We can use it start a new project or replicate one of the projects which are available on GitHub or GitLab, making it easy to not only start, but also carry out collaborative work through these platforms.

With NVIDIA AI Workbench, everything is much easier because Simplifies the process of setting up a GPU-accelerated development environmentsomething that is undoubtedly of great value to those new to this world, as it eliminates the workload and issues that can arise with GPU configuration, driver updates and the errors they can cause, as well as other issues related to data usage , inconsistencies and peculiarities of each model.

This also allows us to enjoy ourselves perfect cooperation, thanks to the integration of versioning and project management tools, and offers total consistency if at some point we want to scale from on-premises to the cloud, both up and down. This makes it very easy to move our projects and switch between on-premises and cloud depending on our needs.

To help users with less experience, NVIDIA offers The RAG Workbench Project, which are development samples designed to go live without the need for complex configurations. They run in a container that has everything needed to move an AI application and have a fully hybrid approach.

RAG Workbench projects are compatible with a wide range of LLMsand its hybrid nature gives us the ability to choose where we want the entire inference process to take place, whether in the cloud or on-premises. For example, we can run the integrated model on the host and the inference process on the “Hugging Face Text Generation Inference” server, in microservices such as NVIDIA NIM, or in other third-party services. These projects also include:

• Performance metrics which will allow us to evaluate the performance of each project with values ​​such as refresh time, time to first token and token rate.
• Transparent recovery, with a panel that accurately displays text fragments that have been recovered from the most contextually relevant content in the vector database that feeds the LLM and that improve the retrieval of relevant answers.
• Own answerswhich can be adjusted based on a wide range of parameters, including the maximum number of tokens that can be generated, their temperature and frequency penalty.

If you’re thinking about getting started with NVIDIA AI Workbench but have a lot of questions I recommend that you take a look at this guide, In it you will find all the information you need to start your new adventure off on the right foot.

AI generated cover image.