Between chatbot more promising, now available at no cost, there is also that of NVIDIA. ChatRTX is a tool that combines the use of Large Language Model (LLM) provided by the company led by Jensen Huang with the acceleration of NVIDIA RTX cards in order to bring the capabilities of generative AI directly to Windows PCs with GeForce GPUs.
ChatRTX allows users to add their own local files as a dataset to extend the potential of open source language models. In this way, it is easy to activate RAG functionality (Retrieval-Augmented Generation) to carry out data research personal and corporate, obtaining relevant and personalized responses. Not only. Since all the processing they happen in locale, no personal or confidential data ever leaves your PC: users can thus enjoy maximum guarantees in terms of security and privacy. In another article we saw how to run the chatbot locally on a QNAP NAS: the goal is to combine the abilities of ChatRTX with the volume of data typically stored in a device mostly storage what is a server NAS.
To use NVIDIA’s intelligent chatbot, you must be using a Windows 10 or 11 machine, have at least 100 GB of free space, a GeForce RTX 3000 or better GPU with at least 8 GB of dedicated VRAM, and the latest drivers. ChatRTX can be downloaded from this page at no cost.
The new ChatRTX recognizes images and opens up to other LLMs
NVIDIA spokespeople have announced the availability of a completely revamped version of ChatRTX. Differently from what was possible up to now, from now on users can choose the LLM to use with the chatbot: you can therefore also draw on the open source models developed and made available by parties other than NVIDIA.
Therefore, alongside the Mistral and Llama 2 models, initially offered by ChatRTX, NVIDIA now also offers alternatives such as Gemma, ChatGLM3 and CLIP.
CLIP it is a model for the image recognition governed by artificial intelligence that maximizes ChatRTX’s abilities. You can activate it and use it, for example, to enable local image search without having carried out a previous “tagging” activity. A demonstration of the potential of ChatRTX with CLIP is available in this YouTube video.
It all comes down to choosing CLIP from the drop-down menu Select AI Modelin the local ChatRTX web interface and the next time you insert one description in natural language of the photos you wish to locate in your archive.
Voice recognition and voice prompt support
The other important new feature integrated into ChatRTX consists in the addition of Vocal recognition. The operation of this feature is model-driven Whisper Of OpenAI.
The system of automatic speech recognition uses AI to process requests made using voice. This way, users can submit vocal query to the application and ChatRTX will provide textual responses.
The expansion of ChatRTX’s abilities looks at the goal towards which many other companies are aiming, evidently starting with OpenAI. NVIDIA users can benefit from a multimodal system that is, capable of combining texts, images and audio. CLIP is a neural network that, through training and refinement, learns visual concepts from natural language supervision: it recognizes what it “sees” in image collections.
By combining the use of this model with Whisper, an advanced tool is available thanks to which it becomes possible to interact not only with “generalist” LLMs but with the “knowledge base” availability within your business or enterprise.