Zum Inhalt springen
KI-Atlas

Llama

Die Modellfamilie, die offene Gewichte zum Mainstream machte

Meta AI (FAIR) Modell Offene Gewichte
EingabeTextText

Der vollständige Artikel liegt auf Englisch vor; Titel und Zusammenfassung sind lokalisiert.

WAS ES IST

Llama is Meta’s large language model family, first released in February 2023 and distributed as open weights that can be downloaded and run in one’s own environment. From Llama 2 onward, Meta permitted commercial use and offered several sizes such as 7B, 13B and 70B; Llama 3.1 extended the largest version to 405B parameters. It is a pretrained base on which the community performs extensive instruction tuning and domain adaptation. Its open weights made it a default starting point for many open-source projects.

Warum es wichtig ist

By releasing large models as open weights under a commercially usable licence, it made “download a base model and fine-tune it yourself” common practice and opened the open-weight camp’s direct competition with the closed frontier.

Wichtige Eckdaten

Parameters
8B / 70B / 405B (Llama 3.1 family)
Context window
128K tokens (Llama 3.1)
Open weights
Yes
Released
2023-02

Fähigkeiten

Verwandte Konzepte

Vergleichbare Produkte