Overview of Llama 3.2

A one-pager on the Llama models so far:

Llama 3.1 models

What’s new in Llama 3.1 and 3.2?

• Tokenizer → a new tokenizer with a vocabulary of 128K tokens.
• Context window → a larger context window of 128K tokens.
• Languages → native support of 8 languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
• Tool calling → Native support for tool calling.
• Llama stack → a set of APIs and CLI for the entire Llama lifecycle, and the API providers and distributions.

What’s new in Llama 3.2?

• Multimodal input in 11B and 90B models
  • Image (object, scenes, drawing) and OCR understanding
  • Captioning and QA
  • Visual reasoning (equations, charts, documents)
• Smaller sizes in 1B and 3B text-only models
  • New SLM models use cases:
    • On-device summarization
    • Writing, translation
    • QA in multiple languages

Vision