Recently, we discussed the latest AI models, especially XAI GROK-1 And google Gemini And Gem. Now it's time to highlight Meta Ai. Last week, the company proudly presented its new creation among the LLM: LAMA 3. Built on previous iterations, this version marks significant progression in AI technology.
Llama 3 is available in two sizes to meet various needs:
- LLAMA 3 8B: Suitable for effective deployment and development on general public GPUs.
- LLAMA 3 70B: Designed for large -scale AI applications.
The two versions include basic models (pre-formed) and refined versions, with a length of context of 8K tokens and adaptable compatibility with diversified consumption equipment.
What distinguishes Llama 3 is his understanding of unrivaled language, her understanding of the nuanced context and her competence in complex tasks such as the translation and generation of dialogue. It transcends the traditional treatment of language, excelling it in reasoning, generation of code and the following instruction.
Trained on a set of detailed data of more than 15 billions of tokens – seven times larger than its predecessor, Llama 2 – Llama 3 pushes the limits of capacity, managing transparent tasks in several stages. Whether you develop conversational agents, support systems or content generators, Llama 3's scalability ensures versatility.
New features in Llama 3:
- Expanded vocabulary: Llama 3 has a new tokenzer with a vocabulary size of 128,256, improving the effectiveness of text coding and multilingual capacities.
- Larger input and output matrices: the increase in the size of the vocabulary improves performance and extends input and integration output matrices.
- LLAMA GURD 2: This adjusted safety model classifies LLM inputs and responses, ensuring the safety of risk taxonomy.
Equipped with advanced automatic learning algorithms, Llama 3 continues continuously and adapts according to the interactions and user comments. This self-learning capacity guarantees that Llama 3 becomes more competent over time, offering increasingly precise and personalized results.
Meta has already started to integrate Llama 3 into its products and services. In particular, Llama 3 feeds the AI ​​assistant from Meta, which helps users by providing useful suggestions, answering questions and facilitating interactions.
Meta's Llama 3 is a model of large open source language. The company announced that it would soon be available on AWS, Databricks, Google Cloud, Hugging Face, Kaggle, IBM Watsonx, Microsoft Azure, Nvidia Nim and Snowflake.
