Falcon returns after a year! 11 billion parameters and 5.5 trillion tokens, performance surpassing Llama 3

In the past few days, the world’s attention seems to have been attracted by the GPT-4o released by OpenAI. At the same time, OpenAI’s challengers are also making history simultaneously.

On May 14, the Technology Innovation Institute (TII) under the Abu Dhabi Advanced Technology Research Council (ATRC) released a new generation of Falcon 2 model.

Falcon 2 11B has opened access, Falcon 2 11B VLM will be open at 12 noon on May 14th

The new generation of "Falcon" (Falcon means falcon) returns to the arena. Once launched, it quickly topped the HN hot list.

Last year, Falcon shocked everyone with its debut technology, surpassing Llama with a crushing advantage.

According to the Hugging Face rankings, the performance of the Falcon 2 11B launched this time once again surpassed the Llama 3 8B and was equivalent to the leading Gemma 7B, with an average score difference of only 0.01.

Falcon 2 includes two versions: Falcon 2 11B and Falcon 2 11B VLM.

The former is a large language model trained on 5.5 trillion tokens, and the latter is a "visual to language" model developed on this basis.

As an open source large language model, Falcon 2 not only has multi-language capabilities, but also supports multi-modality and can seamlessly convert visual input into text output. This function is A first in today's large model market.

You only need to upload a picture in the user interface and enter a prompt to get a text description about the picture

Multi-language+Multi-modal

Falcon 2 11B is multi-language capable and can quickly respond to tasks in English, French, Spanish, German, Portuguese and various other languages.

Falcon 2 11B VLM As a visual language model, it can recognize and interpret various visual information from the environment, so it has a wide range of applications in medical, financial, e-commerce, education and legal fields. Application scenarios.

For example, it is used for tasks such as document management and contextual indexing, and its "visual to language" function is also expected to help the visually impaired in the future.

In addition, Falcon 2 requires only one GPU to run efficiently and can be easily deployed on a laptop.

In order to speed up the commercialization process of Falcon, in November last year, ATRC established a new company-AI71

Based on the Falcon model , AI71 has already entered the UAE's education, medical, legal and other fields, and plans to expand beyond the UAE soon. This upgrade of Falcon provides strong support for its commercialization process.

Adhere to open source Falcon

The artificial intelligence models released by TII have always been among the best in the world. The newly launched Falcon 2 11B model has once again stabilized TII's position .

More importantly, both model versions of this update have been open source, and developers around the world can use them without restrictions and develop and build on them. deploy. The excellent performance demonstrated by the model means that TII is expected to compete with Meta for the dominance of the open source universe.

Faisal Al Bannai, Secretary General of the Abu Dhabi Advanced Technology Research Council (ATRC), promised to continue to promote open source innovation and help developers around the world realize their AI journey.

TII principal researcher Dr. Hakim Hacid also said, “Developers are realizing the countless benefits of smaller, more efficient models... This is the next emerging big trend. In addition, Falcon 2’s visual The translanguage capability opens up new prospects for the ease of use of artificial intelligence, providing users with a revolutionary way to interact from images to text."

TII also revealed that it is in the process of brewing. Next-generation Falcon 2 models, due to be launched soon, will have their performance further enhanced through Mix of Experts (MoE) technology.

The battle between open source and closed source is undecided. But in any case, TII has become a strong force in the field of AGI. Behind this is the top-down joint efforts of the UAE. They are fully promoting the development of artificial intelligence and regard this technology as the next economic pillar after oil.

Falcon does not just exist as a "serious participant", but a "falcon" flying with great ambition.

The above is the detailed content of Falcon returns after a year! 11 billion parameters and 5.5 trillion tokens, performance surpassing Llama 3. For more information, please follow other related articles on the PHP Chinese website!