Presentazione di Pixtral 12B

Pixtral 12B is a deprecated, natively multimodal model from Mistral AI that was trained on interleaved image and text data. It featured a 12B parameter multimodal decoder, a new 400M parameter vision encoder, and supported multiple images of varying sizes in a 128k token context window. The model demonstrated strong performance on multimodal reasoning benchmarks like MMMU, excelling at tasks such as chart and document understanding. Notably, it achieved these vision capabilities without sacrificing its state-of-the-art performance on text-only tasks, outperforming many comparable open-source models.