Meta unveils ‘Llama 3.1’ after 3 months, trained with 16,000 Nvidia ‘H100’ for AI supremacy, “Facebook’s parent company Meta Platform has made a move to gain an edge in the AI arms race by introducing its new artificial intelligence (AI) model ‘Llama 3.1.”,
,
, “According to foreign media such as TechCrunch, Meta recently presented ‘Llama 3.1’ just 3 months after unveiling ‘Llama 3’ in April. ‘Llama’ stands for Meta’s Large Language Model (LLM).”,
,
, “With this release, Meta is expected to engage in fierce competition in the AI market with companies like OpenAI, which recently introduced ‘GPT-4o’, Google with ‘Gemini’, and Ansys with ‘Claude 3.5 Sonnet’ leading the way.”,
,
, “Meta believes that ‘Llama 3.1’, having been trained with more data than its previous version, will be helpful for companies looking to utilize powerful AI models.”,
,
, ‘Meta emphasized, “Llama 3.1 surpassed OpenAI’s latest model ‘GPT-4o’ and Ansys’s ‘Claude 3.5 Sonnet’ in several evaluation criteria of the large-scale Multi-Modal Language Understanding (MMLU) AI model evaluation test.”‘,
,
,
Mark Zuckerberg, Meta CEO (Photo=Meta Newsroom),
, “This model is released in three versions. The largest version, ‘Llama 3.1 405B’, has a staggering 405 billion parameters related to AI data processing capabilities, more than twice the scale of GPT-3’s 175 billion parameters.”,
,
, “Meta also introduced the small model ‘Llama 3.1 8B’ with 8 billion parameters and the medium model ‘Llama 3.1 70B’ with 70 billion parameters. These small models are known to be suitable for operating chatbots and software coding AI.”,
,
, “Similar to its previous models, Meta has made ‘Llama 3.1’ open-source for anyone to use for free, in contrast to OpenAI and Google keeping their AI models private.”,
,
, “Some view Meta’s move as a strategy to strengthen its dominance by securing a larger user base and obtaining an industry standard position. However, there is disappointment that ‘Llama 3.1’ is not a multimodal model and cannot understand or process images.”,
,
, ‘Meta stated, “Llama 3.1 costs only about half of OpenAI’s GPT-4o operational costs,” and emphasized, “We are collaborating with over 20 companies including Microsoft, Amazon, Google, Nvidia, and more to make this model accessible to many developers.”‘,
,
,
Llama 3.1 is released in three versions. (Photo=Meta Newsroom),
, “Furthermore, ‘Llama 3.1’ is reported to have been trained on 16,000 of Nvidia’s latest commercial AI chips, the ‘H100’ graphics processing units (GPUs). Although Meta did not disclose the development cost of ‘Llama 3.1,’ given that the ‘H100’ chips cost between $30,000 to $40,000 per unit, the chip purchase alone would have amounted to hundreds of millions of dollars.”,
,
, ‘Mark Zuckerberg, Meta CEO, stated, “\’Llama 3.1\’ is a product that can compete with the most advanced models,” and expressed his expectation that upcoming Llama models to be released next year will be the most advanced models in the industry.’,
,