Meta AI’s Llama 3.1 405b is a hit that has managed to impress many users. The new model is no small fry – it’s a big fish in an ever-growing pool of language models. Let’s take a look at the AI that has everyone talking and writing.
Meta AI’s Llama 3.1 405b is, as the name suggests, a big language model with 405 billion parameters. It is part of Meta’s Llama 3 series, launched in April 2024. Early benchmarks suggest that this model could outperform the current leaders in several key AI tests.
A horse of a different color: Meta AI’s Llama 3.1 405b specs
Meta AI, formerly known as Facebook AI Research, is the artificial intelligence research division of Meta Platforms. They introduced Llama (Large Language Model Meta AI) in 2023 as an open-source alternative to proprietary language models. Llama quickly gained popularity in the AI community. Building on this success, Meta released Llama 2 in 2023, which showed significant improvements.
Now, with Llama 3, Meta has pushed the boundaries even further, culminating in the powerful Llama 3.1 405b model we’re discussing today. This rapid progression showcases Meta’s commitment to advancing open-source AI technology.
In this era where artificial intelligence is all around us, companies do not stop. In an age where we look at old game graphics and wonder how much more can be improved, we’ve come to forgive new graphics, and the same is true for artificial intelligence.
Meta AI is no slouch when it comes to performance. This model has 405 billion parameters, making it a heavyweight player in the AI arena. So, what does this 405b parameter mean?
Neigh-ver say never: Meta AI’s Llama 3.1 405b vs competitors
Meta AI’s Llama 3.1 405b is showing impressive results in early benchmarks. It outperforms GPT-4 in several tests, including GSM8K, Hellaswag, Boolq, and various MMLU categories. However, it lags behind in areas like HumanEval and MMLU social sciences.
The model’s performance is particularly strong in math and coding tasks. For example, in the GSM8K test, Meta AI’s Llama 3.1 405b scored 96.8, while its 70B counterpart achieved 94.8. In HumanEval, the 405B model reached 85.3, compared to 79.3 for the 70B version.
These figures are based on the base model. Instruction tuning could potentially improve these results further. So we can say that these numbers represent processing power, the bigger the number the better (but sometimes), but sometimes the models are crushed under their load.
Hoof It to the future: Meta AI’s Llama 3.1 405b and open-source AI
The fact that Meta AI’s Llama 3.1 405b model is open-source AI and that for the first time, an open-source model can beat the best-closed source LLM available in various benchmarks, may be a sign of things to come, even if not fully understood at the moment. For now, we can make our GPTs in ChatGPT. This dependency may be broken in the future.
Meta AI’s Llama 3.1 405b is a strong new player in the AI space. Its strong performance on various benchmarks and open-source nature make it a model to watch. As AI continues to evolve, Meta AI’s Llama 3.1 405b can play an important role in shaping the future of language models and AI technology.
Featured image credit: Meta AI Blog
