Llama 3, Meta's open-weights AI model, offers two versions – pre-trained and instruction-tuned, in 8B and 70B parameter sizes accessible through Meta's website.
Meta trained Llama 3 on two 24,000-GPU clusters, with the 70B model utilizing around 15 trillion tokens of data before moving to other models.
[
add
]
[
|
|
...
]