February 24, 2023 High Open Source Models · 1 min read

LLaMA: Meta opens foundation models to research

In one sentence Meta releases LLaMA in four sizes (7B, 13B, 33B, 65B), available to researchers on request. One week later, the weights leak publicly.

Verified Official source

ShareLinkedIn X

Reading level

Meta introduces LLaMA, a family of language models in "reasonable" sizes: from a small 7-billion-parameter model up to a 65-billion one.

The selling point: performance comparable to GPT-3 (175B) with a tenth of the parameters, because the model is trained on far more data. Meta distributes it on request to researchers, but within days the weights show up online via torrent.

At that point anyone with a decent GPU can run a near-state-of-the-art model. The open LLM ecosystem really begins.

Companies