6,294
edits
Meta Platforms:AI Capabilities: Difference between revisions
→Development Process
| Line 34: | Line 34: | ||
=== Development Process === | === Development Process === | ||
Here is a summary of Llama 3 development process<ref>https://www.youtube.com/watch?v=r3DC_gjFCSA</ref>: | |||
* The generative AI team at Meta was formed in February 2023. | * The generative AI team at Meta was formed in February 2023. | ||
* Meta Llama 3 exists in two forms 8B and 70B parameters. | * Meta Llama 3 exists in two forms 8B and 70B parameters. | ||
* The development process of Llama 3 is more powerful than that of Llama 2. | * The development process of Llama 3 is more powerful than that of Llama 2. | ||
* According to Meta, Llama 3 still lacks a lot of things they planned to put into it. [[File:Llama 3 vs Llama 2.png|left|thumb|Source: Meta]] [[File:Evolution of Llama.png|left|thumb|330x330px|Source: Meta]] <br /> | * According to Meta, Llama 3 still lacks a lot of things they planned to put into it. | ||
* Meta said most of the time is now spend on post training. | |||
* Meta already partners with a number of enterprises to open source its Llama 3 model including Google, Microsoft Azure, AWS, IBM, etc. | |||
* Llama 3 performed well when tested for cyber attack and prompt injection rate. <br />[[File:Llama 3 vs Llama 2.png|left|thumb|Source: Meta]] [[File:Evolution of Llama.png|left|thumb|330x330px|Source: Meta]] <br /> | |||
| Line 61: | Line 64: | ||
[[File:Development approach.png|left|thumb|Source: Meta]] | |||