Llama 3.3 70B

Llama 3.3 70B is Meta’s large, text-only autoregressive language model for chat, generation, and other language tasks. It should not be confused with the Llama 3.2 11B Vision model: this 70B entry does not accept images. Its memorable scale is about 70 billion parameters, giving it a much larger footprint than the 1B, 3B, and 8B Llama entries in the same matrix. TT-Metal shows two Wormhole deployments for the same model: TP=8 on QuietBox and TP=32 on Galaxy, both at batch 32. Those duplicate hardware rows collapse into one model item here; they show that one checkpoint can be partitioned across different chip counts.