Opera Software made experimental changes to my opara browser One, adding the possibility of local launch of large language models (LLM) as part of the AI Feature Drop. So far, this function is available exclusively in the browser version for developers available only on desktop platforms.
The update adds to Opera One Developer 150 different LLM of 50 families, including popular Llama, Gemma and Mixtral.
The local work of large language models will probably be “much slower” than the use of online models on servers with powerful NVIDIA GPUs. However, a more serious problem than the low productivity of the home iron can be the required disk space: most models occupy from 2 to 10 GB of memory, and some models can even reach 70 GB of frightening. And if for some reason the user wants to install all 150 language models at once, even a 1 terabyte drive may not be enough.
The greatest irony is that it is the Internet browser, which involves the constant availability of a network for work, has become a single chat bots that do not require connection to the Internet. Although here, again, the emphasis is on the safety of data rather than on the use of LLM in Spartan conditions.
Despite the possible restrictions and temporary inaccessibility of the function for everyone, Opera One was the first browser in the world to offer a solution for the local launch of LLM on a PC and this definitely causes delight and a clear feeling that the future is already here.