Uploaded July 2024 | Updated September 2026, 3 days ago
With Llama 3 we switched from using SentencePiece to Tiktoken. Aston Zhang, author of D2L and researcher on the Llama team at Meta shared a little bit more on some of the efficiencies this is leading to and how it's enabling stronger performance and potentially driving down inference costs.
--
Subscribe: youtube.com/aiatmeta?sub_confirmation=1
Learn more about our work: ai.meta.com
Follow us on Twitter: twitter.com/aiatmeta
Follow us on Facebook: facebook.com/aiatmeta
Connect with us on LinkedIn: linkedin.com/showcase/aiatmeta
Meta focuses on bringing the world together by advancing AI, powering meaningful and safe experiences, and conducting open research.
With Llama 3 we switched from using SentencePiece to Tiktoken. Aston Zhang, author of D2L and researcher on the Llama team at Meta shared a little bit more on some of the efficiencies this is leading to and how it's enabling stronger performance and potentially driving down inference costs.
--
Subscribe: youtube.com/aiatmeta?sub_confirmation=1
Learn more about our work: ai.meta.com
Follow us on Twitter: twitter.com/aiatmeta
Follow us on Facebook: facebook.com/aiatmeta
Connect with us on LinkedIn: linkedin.com/showcase/aiatmeta
Meta focuses on bringing the world together by advancing AI, powering meaningful and safe experiences, and conducting open research.




