After much waiting, the Llama 3 405B, originally scheduled for release on the 23rd, is finally here. In the product structure, the operating income of 100-300 billion yuan products is 4.01/12.88/0.06 billion yuan respectively, with a total sales volume of 18,000 kiloliters, up 28.10% year-on-year, showing significant growth.

As the top model in the Llama 3 series, the 405B version has 405 billion parameters and is one of the largest open-source models to date.

Last night, a leak of Llama 3.1-405B evaluation data occurred, and some netizens predict that a Llama 3.1-70B version could be released simultaneously, stating that "(early model leaks) is a tradition of META, and the Llama model did it last year."

Some analysts believe that Llama 3 405B is not just another upgrade in artificial intelligence capabilities, but also an opportunity for democratization and placing the most advanced AI directly into the hands of developers, potentially representing a ChatGPT moment for open-source AI.

Three predictions for the upcoming announcement of Llama 3 405B

Analysts have predicted the highlights of the upcoming Llama 3 405B announcement from three perspectives: data quality, model ecosystem, and API solutions.

Firstly, Llama 3 405B may completely change the data quality of specialized models.

For developers who specialize in building professional AI models, a long-term challenge they face is obtaining high-quality training data. Smaller expert models (1-10B parameters) usually use distillation techniques to enhance their training datasets with outputs from larger models. However, the use of such data from closed-source giants such as OpenAI is strictly limited, restricting commercial applications.

Llama 3 405B was created to address this. As an open-source giant with performance comparable to proprietary models, it provides a new foundation for developers to create rich, unrestricted datasets. This means that developers can freely use Llama 3 405B's distilled output to train niche models, greatly accelerating innovation and deployment cycles in professional fields. It is expected that high-performance, fine-tuned models will be developed, which are both powerful and in line with open-source ethical norms.

Secondly, Llama 3 405B will create a new model ecosystem, from basic models to expert combinations.

The release of Llama 3 405B may redefine the architecture of AI systems. The model's huge scale (405 billion parameters) may suggest a one-size-fits-all solution, but the real power lies in its integration with a hierarchical model system. This approach is particularly resonant for developers who use AI of different scales.

It is expected that there will be a shift towards a more dynamic model ecosystem, in which Llama 3 405B acts as the backbone and is supported by small and medium-sized models. These systems may use inference decoding and other techniques, where less complex models handle most of the processing, and calls to the 405B model are made for validation and error correction only when necessary. This can not only maximize efficiency, but also open up new avenues for optimizing computing resources and response times in real-time applications, especially when running on SambaNova RDU optimized for these tasks.

Finally, Llama 3 405B has the potential to compete with the most efficient API solutions.

The bigger the capacity, the greater the responsibility - for Llama 3 405B, deployment is a major challenge. Developers and organizations need to carefully address the complexity and operational requirements of the model. AI cloud providers will compete to provide the most efficient and cost-effective API solutions for deploying Llama 3 405B.

This situation provides a unique opportunity for developers to interact with different platforms and compare how various APIs handle such a large model. The winners in this field will be those who can provide APIs that not only manage computing loads effectively but also do not sacrifice the accuracy of the model or disproportionately increase its carbon footprint.

In conclusion, Llama 3 405B is not just another weapon in the AI arsenal; it represents a fundamental transformation towards open, scalable, and efficient AI development. It is believed that the arrival of Llama 3 405B will open up new horizons for users, whether in fine-tuning niche models, building complex AI systems, or optimizing deployment strategies.

How do netizens view this?

Netizens posted on the LocalLLaMA subreddit sharing information about the 405-billion-parameter Meta Llama 3.1. From the results of several key AI benchmark tests, its performance surpasses that of the current leader, OpenAI's GPT-4o, marking the first time that an open-source model may have beaten the most advanced closed-source LLM model.

As shown by benchmark tests, Meta Llama 3.1 outperforms GPT-4o in multiple tests such as GSM8K, Hellaswag, boolq, MMLU-humanities, MMLU-other, MMLU-stem, and winograd, but lags behind GPT-4o in HumanEval and MMLU-social sciences.

Ethan Mollick, Associate Professor at the Wharton School of Business at the University of Pennsylvania, writes:

If these statistics are true, it can be said that the top Al model will be open to everyone for free starting this week.

Governments, organizations and companies in every country in the world can use the same artificial intelligence functions as others. This will be very interesting.

Some highlights of the Llama 3.1 model were summarized by netizens:

The model was trained using publicly available 15T+ tokens, with a pre-training data cutoff date of December 2023;

Fine-tuning data includes publicly available instruction fine-tuning datasets (different from Llama 3) and 15 million synthetic samples;

The model supports multiple languages, including English, French, German, Hindi, Italian, Portuguese, Spanish and Thai.

Some netizens have expressed that this is the first open-source model to surpass closed-source models such as GPT4o and Claude Sonnet 3.5, achieving SOTA on multiple benchmarks.