Meta launched its next-generation artificial intelligence (AI) models Llama 3 8B and 70B on Thursday. Llama 3, short for Large Language Model Meta-AI, offers improvements over its predecessor. The company also adopted new training methods to optimize the efficiency of the model. Interestingly, the largest model in Llama 2 was the 70B, but this time the company says its large model will contain more than 400 billion parameters. Notably, a report last week suggested that Meta would launch a smaller AI model in April and a larger model later in the summer.

Those interested in trying new AI models are in luck, as Meta is taking a community-first approach with Llama 3. The new base model will be open source like previous models.Yuan Zai Qi blog post”, “Llama 3 models will soon be available on AWS, Databricks, Google Cloud, Hugging Face, Kaggle, IBM WatsonX, Microsoft Azure, NVIDIA NIM and Snowflake, supported by hardware platforms from AMD, AWS, Dell, Intel , Nvidia and Qualcomm. “

The list includes all major cloud, hosting, and hardware platforms, which should make it easier for hobbyists to master AI models. In addition, Meta has integrated Llama 3 with its own Meta AI, which can be accessed via Facebook Messenger, Instagram and WhatsApp in supported countries.

During the show, the social media giant shared benchmark scores for Llama 3 pre-trained and guided models. For reference, the pre-trained is a general conversational AI, while the guidance model is designed to accomplish a specific task. Llama 3 70B’s pre-trained model surpassed Google Gemini 1.0 Pro in MMLU (79.5 vs 71.8), BIG-Bench Hard (81.3 vs 75.0) and DROP (79.7 vs 74.1) benchmarks, with the 70B Instruct model outscoring Gemini 1.5 Pro model in MMLU, HumanEval and GSM-8K benchmarks, based on data shared by the company.

Meta chose a decoder-only Transformer architecture for the new AI model, but with some improvements over its predecessor. Llama 3 now uses a tokenizer with a vocabulary of 128K tokens, and the company has adopted Grouped Query Attention (GQA) to improve inference efficiency. GQA helps improve the AI’s attention so that it doesn’t break away from the specified context when answering queries. The social media giant has pre-trained the model using more than 15T tokens, which it claims are derived from public data.