The company Nvidia is playing hard in the AI world with its latest generation hardware, but now also adds software with NVLM. It will compete with GPT-4o and Llama through an open source LLM to add to multimodal models (image, text).
Nvidia dominates the sector AI hardware with its GPUs to centralize data, with the H100 and B200 models being the most representative. But with NVLM comes a new bet for the large language models (LLM) sector. Nvidia's proposal comes to fight in a sector dominated by OpenAI and its GPT-4º. However, there are other competitors too, such as Google's Gemini, Meta's Llama and Anhtropic's Claude 3.5.
What does Nvidia's NVLM propose to revolutionize the world of AI?
Nvidia NVLM 1.0 is the new competitor in the world of large language models for AI. The developers published a study detailing how this new proposal, which aims to compete in the AI software segment, works and what it will allow.
In short, NVLM is a whole family of multimodal LLMs that, according to Nvidia, provides remarkable results in vision and language. The study indicates an analysis and development capacity very similar to that of other popular models, such as GPT-4o.
In NVLM we find a model with 72.000 billion parameters, the most ambitious and capable number to date. According to Nvidia, its performance and type of responses are better than that of the Llama 3 405B, a much larger model according to performance tests.
Another advantage of Nvidia's proposal is that NVLM offers an open source AI model with open weights. Its developers promise to publish the code used to train the model, which is extremely useful. Developers will be able to use it in their own projects and forks.
What will NVLM offer to the AI world?
NVLM will allow you to analyze visual and text inputs. This translates into the AI's ability to interpret memes and analyze photographs. In the process, you will also be able to use this language model created by Nvidia to solve mathematical problems step by step.
How this system works combines OCR, localization, common sense, world knowledge and programming capabilities. All of this together allows NVLM to respond to different requests and situations with great versatility. Exploring the scope of this new AI language model, and its performance comparison with other players in the sector, allows us to understand where the technology is heading.
About pre-workout
The people behind NVLM training have used an improved architecture for training and reasoning their model. Its capabilities are very versatile and it is capable of processing 72.000 billion parameters and directly competing with GPT-4o, Llama 3-V-70B and Gemini 1.5 ProIts increased performance is currently proven in solving mathematical problems, image and text processing.
The data used for pre-training and training was carefully selected with fine-tuning and human supervision. The work served to verify the quality of the data sets as well as the diversity of compatible tasks and scale, even during the pre-learning stage.
El Powerful algorithm created by Nvidia It was released as open source, along with the model, instructions and training parameters. They can be used and modified free of charge, distributed through Megatron-Core, the firm's development library.
This is a real milestone in the industry, because Nvidia is making it easier for other small organizations or independent researchers to also contribute to the advancement of AI. free access to the tool and its characteristics similar to those of other big tech products will mark a before and after in the race for control of the sector.
The aim of this new LLM is to expand the user and customer base, favouring a business that is already lucrative. But now it will add even more enthusiasts and potential users around the world. An excellent step for Nvidia, which, in addition to dominating the AI sector through hardware, is now targeting the software sector directly.
Hybrid business strategy
The leading position that Nvidia has taken in the technology sector and in the development of Artificial Intelligence it responds to a hybrid business strategy. On the one hand, it contemplates the development and production of chips and the sale of advanced systems to promote the operation of very diverse algorithms. This accelerated the generation of income and obtained the approval of investors. On the other hand, work is being done on AI algorithms so that Nvidia can position itself in a segment where it is at a disadvantage compared to OpenAI, Google and Meta. But thanks to this strategy, NVLM is proving to be a great tool, capable of fighting head to head with AI giants that are already positioned.
With the new Nvidia processors, the speed and capacity for analysis and resolution of mathematical problems that users achieve is remarkable. The arrival of NVLM 1.0 will mark a before and after for the sector, generating the possibility of a new player with Open Source proposals to reach a greater number of users and small and medium-sized companies.
What will the AI market situation be like now that Nvidia has entered with NVLM?
Until now, Nvidia offered the Top processors and GPUs for all types of data centers and servers where AI models are run. But NVLM implies Nvidia's direct landing in the sector, and this may generate some friction with the rest of the competition. The open source initiative and the ambitious processing power make NVLM a very attractive language model. It remains to be seen how it adapts and the uses that the user community begins to give it to fully understand its scope. We will probably begin to hear important news from the AI sector now that Nvidia has entered the game.