Mixtral 8x7B architecture: Sparse Mixture-of-Experts (8 experts × 7B params, 2 experts routed per token).
We use analytics cookies to learn which pages help. Privacy Details