Skip to content

Llama-3.2-3B

Multilingual 1B and 3B models from Meta

Visit official site
Category AI models

What Llama-3.2-3B is

The model is described as a multilingual large language model collection from Meta with 1B and 3B pretrained and instruction-tuned variants for dialogue use cases.

What you can do with it

  • Build multilingual chatbots for customer service
  • Use it as a knowledge retrieval tool
  • Generate articles and summaries for content work
  • Run dialogue inference through the transformers pipeline

Model notes

  • Eight languages are listed, including Thai and Portuguese
  • An optimised Transformer is tuned with SFT and RLHF
  • Training used up to 9 trillion tokens and 916k GPU hours
  • Meta reports net-zero greenhouse gas emissions for training

Pros & cons

✓ What we like

  • Tuned with supervised fine-tuning and reinforcement learning from human feedback
  • Released with safety fine-tuning and system-level safety measures

! What to watch out for

  • Running inference needs transformers at version 4.43.0 or later
  • The training data stops at December 2023

FAQ

Which sizes are available?

The material lists 1B and 3B pretrained and instruction-tuned models.

How much compute was used?

The material states 916k GPU hours of training compute.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review