[R] Meta AI open sources new SOTA LLM called LLaMA. 65B version (trained on 1.4T tokens) is competitive with Chinchilla and Palm-540B. 13B version outperforms OPT and GPT-3 175B on most benchmarks. Submitted by MysteryInc152 t3_11awp4n on February 24, 2023 at 5:21 PM in MachineLearning https://twitter.com/GuillaumeLample/status/1629151231800115202?t=4cLD6Ko2Ld9Y3EIU72-M2g&s=19 Paper here - https://research.facebook.com/publications/llama-open-and-efficient-foundation-language-models/ No comments 2