Book a call

News · April 18, 2024

Meta releases Llama 3, widening the choice of open-weight models

// ai.news :: 2024

Meta has released Llama 3 in two sizes with downloadable weights and broad cloud availability. For companies weighing hosted APIs against self-hosted models, the open option just became stronger.

What happened

  • Meta introduced Llama 3 on April 18, 2024 with pretrained and instruction-tuned models of 8B and 70B parameters, trained on more than 15 trillion tokens, according to its AI blog.
  • Meta said the models would be available on platforms including AWS, Databricks, Google Cloud, Hugging Face, Microsoft Azure, NVIDIA NIM and Snowflake, and that a model with more than 400B parameters was still in training.
  • According to Meta, the release came with safety tools including Llama Guard 2 and Code Shield, which filters insecure code at inference time; the context length at launch was 8,192 tokens.
  • Fortune reported that Meta's own assistant, Meta AI, now runs on Llama 3 inside WhatsApp, Instagram, Facebook and Messenger.

Why it matters

Open-weight models give teams options that hosted APIs do not: running inside their own cloud account, keeping sensitive data in place, fine-tuning for a narrow task and predicting cost per request. The trade-off is that the team now owns hosting, scaling, evaluation and safety filtering.

Our view: for focused tasks such as classification, extraction or support drafting, test the 8B model against your current API on your own data before deciding. Check the license terms for your use case, and keep a model-agnostic layer in the code so switching later is cheap.

Sources

  • Meta AI Blog: Introducing Meta Llama 3: The most capable openly available LLM to date
  • Fortune: As Meta debuts its Llama 3 model, today's generative AI landscape is a different animal

This post is CMM’s own summary and commentary on publicly reported news. Facts are attributed to the sources linked above; please read them for full details. CMM is not affiliated with these publishers or with the companies mentioned, and trademarks belong to their owners. Prepared with AI assistance. Corrections or requests: [email protected]. See our editorial policy.

Want help with this?

Tell us what you need, and we will suggest the right team model on a short call.

Talk to us