On April 18th local time, Meta released its latest open-source model Llama 3, providing pre trained and instruction fine-tuning versions of 8B and 70B. This is currently the most advanced open source model, which will create strong competition for large model companies, including OpenAI, and also benefit many artificial intelligence companies that rely on open source models.
Amidst a slight decline in the overall market, Meta's stock price rose by 1.54%.
According to Meta's official introduction, Llama 3 was trained on two customized 24K GPU clusters based on data exceeding 15T - which is 7 times larger and 4 times more code than the dataset used by Llama 2, and Llama 3 supports 8K context length, which is twice the capacity of Llama 2.
Meta Official
In addition, Meta has also released parameter comparisons between two versions of Llama 3 and competitors such as Google Gemma, Google Gemini, Mistral, and Antioptic's Claude 3.
For users, they can also experience Llama 3 through Meta's own Meta AI. Meta CEO Zuckerberg introduced that Meta AI will be integrated into the search boxes at the top of Meta's major products WhatsApp, Instagram, Facebook, and Messenger, and has created a website called meta.ai to make it easier to use.
According to Yann LeCun, Chief Scientist of Meta, Llama will continue to release more versions in the coming months and will soon release research papers on Llama 3. Jim Fan, a senior scientist at Nvidia, believes that future versions of Llama 3-400B and above may become a "watershed" for the open source community to use GPT-4 level models.
It is worth noting that the launch of Llama 3, like its previous version, caused a sensation in the open source community. But earlier this week, Baidu founder Robin Lee publicly said that the open source model will be more and more backward.
Robin Lee said that the smaller size model cut by developers through dimension reduction of Baidu's Wenxin 4.0 is significantly better than the model directly brought out from the open source model under the same size; Under the same effect, the cost is significantly lower. "People used to think that open source was cheap when using it, but in large model scenarios, open source is actually the most expensive. Therefore, open source models will become increasingly outdated."