Thursday (April 18th) Meta Introducing the most capable Large Language Model (LLM)The company also introduced an image generator that updates images in real-time even while users type prompts. Meta plans to integrate its latest model into its own virtual assistant, Meta AI.
Meta touts its latest model as the most sophisticated AI model, far ahead of competitors like Google, Mistral, and others in terms of performance and functionality. The updated Meta AI assistant will be integrated into Facebook, Instagram, WhatsApp, Messenger, and standalone websites much like OpenAI’s ChatGPT.
Here we take a look at what exactly Meta’s Llama 3 model is, how it’s different, and why Meta claims it’s its most capable model yet.
What is Rama 3?
Llama or Large-Scale Language Model Meta AI is a family of LLMs introduced by Meta AI in February 2023. The first version of the model was released in his four sizes: 7B, 13B, 33B and 65 billion parameters. Llama’s 13B model reportedly outperformed OpenAI’s GPT-3, which has 135 billion parameters. To simplify, the parameters here are a measure of the size and complexity of the AI model, and generally a higher number of parameters means a more complex and powerful AI model. Meta released his Llama 2, a significantly upgraded version of his first LLM, last July. Llama 2 was released with 7B, 13B, and 70B parameters and was trained on 40% more data compared to the previous version.
Now, Meta is back with Llama 3, the latest version of LLM, which is claimed to be the most sophisticated model with significant advances in terms of performance and AI features. Llama 3 is based on the Llama 2 architecture and is released in two sizes with parameters 8B and 70B. Both sizes come with a base model and instruction-tuned versions designed to improve performance for specific tasks. The tailored version of the command is reportedly intended to power AI chatbots aimed at conversing with users.
According to Meta, the company used Llama 3 to build the best open source model on par with the best proprietary model currently available. We also embrace the open source ethos of releasing early and enabling the development community. — A community of software engineers — To access models under development. For now, Meta has released a text-based model in the Llama 3 model collection. However, the company plans to make Llama 3 multilingual and multimodal and embrace longer contexts, while continuing to improve performance across his LLM abilities such as coding and reasoning.
All Llama 3 models support a context length of 8,000 tokens. This allows for more interactions and complex input processing compared to Llama 2 and 1. More tokens here means more content input or prompts from the user and more content in response from the model. When it comes to safety, Meta said it is committed to developing Rama 3 in a responsible manner. “We are providing a variety of resources for others to use responsibly, including new reliability and safety tools with Llama Guard 2, Code Shield, and CyberSec Eval 2. deployment,” the company said in a blog post.
How good is Rama 3?
Meta claims that the Llama 3 model with parameters 8B and 70B is a big leap forward from Llama 2. This was made possible through pre- and post-training improvements. “Our pre-trained and instruction-fine-tuned models are the best models that currently exist with parameter scales of 8B and 70B,” the company says on his website. According to the company, the post-training process improves Llama 3’s usability and significantly improves capabilities such as inference, code generation, and instructions.
Meta claims that Llama 3 8B outperformed other open source AIs such as Mistral 7B and Gemma 7B in benchmark evaluations. Llama 3 outperforms Google’s Gemma 7B and Mistral’s Mistral 7B, Anthropic’s Claude 3 Sonnet, with benchmarks such as MMLU 5 shots (Massive Multitask Language Understanding), GPQA 0 shots (graduate-level Google Proof Q&A benchmark), and HumanEval 0 shots. exceeded. (Benchmark for evaluating the multilingual ability of code generation models), GSM-8K 8-shot and Mathematics 4-shot, CoT (Mathematics and Word Problems).
Although Meta has not officially stated any use cases for Llama 3, given its similarities with existing AI chatbots, Llama 3 can be used to create various forms of text such as poetry, code, scripts, and musical compositions. Can be used for It can also be used to summarize factual topics and translate languages.
How can I try Llama 3?
Meta announced that it has integrated Llama 3 into Meta AI, which can be used on Facebook, Instagram, WhatsApp, Messenger, and the web. Meta has integrated his LLM into the Hugging Face ecosystem, so it’s readily available to developers. Also available from Perplexity Labs, Fireworks AI, and cloud provider platforms such as Azure ML and Vertex AI.
Llama 3 models will be available soon on AWS, Google Cloud, Hugging Face, Databricks, Kaggle, IBM WatsonX, Microsoft Azure, NVIDIA NIM, Snowflake, and more.
Meta AI is currently available in English on WhatsApp across the United States. Meta has also expanded to more countries including Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zimbabwe, and Zambia.
