LLaMA
Meta's original 2023 open-weight release that kicked off the open LLM boom
LLaMA is Meta’s original large language model, released in February 2023 in four sizes ranging from 7B to 65B parameters, with a modest 2,048-token context window by today’s standards. Unlike later Llama generations, access was initially restricted: Meta granted weights to researchers on a case-by-case basis under a non-commercial license rather than publishing them freely, though the weights leaked publicly within weeks of release. Meta reported that its 13B model beat GPT-3’s 175B parameter version on several benchmarks, and that the 65B model was competitive with PaLM and Chinchilla, a claim that got the open-source LLM community paying close attention to efficient training. LLaMA is the model that set off the wave of open fine-tunes and derivatives, like Alpaca and Vicuna, that defined open-source LLM research through 2023, even though it has since been fully superseded by Llama 2, 3, and 4.