Models

Llama 3.2 3B

Meta · September 2024

A small Llama for low-resource devices (~2.2 GB VRAM).

Llama 3.2 3B runs in your browser, on your own device. The model downloads once and stays there; what you type is never sent to Meta or to us , and no API key is involved.

  • Modalities

  • Context window

    131,072 tokens, ≈ 98,000 words · about 390 pages

    the model's own limit — how much your device can hold depends on its memory

  • Knowledge cutoff

    December 2023

  • Runs on: Your device, in the browser
  • Size: 3B

Files you can attach

  • ImagesNot supported
  • DocumentsNot supported
  • Text and codeTXT, MD, CSV, TSV, RPT, LOG, PY, TS, TSX, JS, JSX, CS, GO, JAVA, KT, SWIFT, CPP, C, H, PHP, RB, SQL, SH, PS1, HTML, CSS, SCSS, JSON, YAML, YML, XML, INI

Features

  • StreamingSupported
  • Function callingNot supported
  • Structured outputsSupported
  • Fine-tuningNot supported

Where it comes from

Model cardMeta's card for Llama 3.2 3B
WeightsThe build your browser downloads
RuntimeWebLLM

Ready to try it?

Log in

Runs on your device · No API key · Your chats stay in your browser