The new brainchild of Meta challenges the giants of the industry.
Meta has officially announced the release of its latest artificial intelligence model, Llama 3, which developers claim can compete with larger models from Google, Mistral and Anthropic.
Llama 3 is presented in versions with a different number of parameters: from 8 to more than 400 billion. Currently, options with 8 and 70 billion parameters are available. Meanwhile, the company continues to work on training larger and more complex language models.
One of the key improvements of Llama 3 is the use of a new tokenizer with a lexicon of 128 thousand tokens, which significantly increased the efficiency of natural language coding. Additional performance improvements are achieved through the use of high-quality data sets and additional adjustment steps after training.
Llama 3 training was conducted on the basis of more than 15 trillion tokens collected from publicly available sources. This is 7 times more than the previous Llama 2 model. Meta has also developed a special input filtering system to minimize training on low-quality information.
In addition, the company has implemented new security tools, including Llama Guard 2 and Cybersec Eval 2, designed to protect the model from abuse and attacks.
In the coming months, Meta plans to introduce additional models, including the aforementioned one for 400 billion parameters. These models will support advanced features and have much larger context windows that allow you to handle more complex queries.
The Llama3-8B and Llama3-70B models are already available for download on the Meta website. It will also be possible to deploy them in cloud platforms such as Amazon Web Services, Microsoft Azure, and Google Cloud.
Meta has officially announced the release of its latest artificial intelligence model, Llama 3, which developers claim can compete with larger models from Google, Mistral and Anthropic.
Llama 3 is presented in versions with a different number of parameters: from 8 to more than 400 billion. Currently, options with 8 and 70 billion parameters are available. Meanwhile, the company continues to work on training larger and more complex language models.
One of the key improvements of Llama 3 is the use of a new tokenizer with a lexicon of 128 thousand tokens, which significantly increased the efficiency of natural language coding. Additional performance improvements are achieved through the use of high-quality data sets and additional adjustment steps after training.
Llama 3 training was conducted on the basis of more than 15 trillion tokens collected from publicly available sources. This is 7 times more than the previous Llama 2 model. Meta has also developed a special input filtering system to minimize training on low-quality information.
In addition, the company has implemented new security tools, including Llama Guard 2 and Cybersec Eval 2, designed to protect the model from abuse and attacks.
In the coming months, Meta plans to introduce additional models, including the aforementioned one for 400 billion parameters. These models will support advanced features and have much larger context windows that allow you to handle more complex queries.
The Llama3-8B and Llama3-70B models are already available for download on the Meta website. It will also be possible to deploy them in cloud platforms such as Amazon Web Services, Microsoft Azure, and Google Cloud.
