Meta Llama 3.1 LLM

From GM-RKB
Jump to navigation Jump to search

A Meta Llama 3.1 LLM is a Llama 3 LLM that introduces some enhancements.



References

2024

2024

  • (AI@ Meta Llama Team, 2024) ⇒ AI@Meta Llama Team. (2024). “The Llama 3 Herd of Models.” In: Meta AI Research.
    • NOTE: The paper introduces Llama 3, a set of foundation models supporting multilinguality, coding, reasoning, and tool usage. The largest model has 405B parameters and performs competitively with GPT-4. It includes multimodal capabilities but is still under development.
    • NOTE: Llama 3.1 models support long context windows up to 128K tokens, enhancing their ability to handle extensive input sequences effectively.
    • NOTE: The Llama 3 series includes models with varying parameters, such as 8B, 70B, and 405B, catering to different levels of computational needs and applications.
    • NOTE: The Llama 3 models are pre-trained on a large-scale, high-quality dataset of 15T multilingual tokens, significantly improving over previous versions.
    • NOTE: Llama 3 models utilize a dense Transformer architecture with enhancements like Grouped Query Attention (GQA) for improved inference speed and reduced memory usage.
    • NOTE: The development of Llama 3 involved extensive empirical evaluations, demonstrating competitive performance with state-of-the-art models on tasks like coding, reasoning, and multilingual processing.
    • NOTE: The Llama 3 models are part of Meta's initiative to release open-access, high-performance language models, aiming to foster innovation and responsible AI development in the research community.
Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.