Mistral releases its first multimodal model, Pixtral 12B, available on GitHub and Hugging Face, and via API-serving platforms Le Chat and Le Platforme “soon”
French AI startup Mistral has released its first model that can process images as well as text.
Mistral released Pixtral 12B Vision Language Model 🔥 Some notes on the release: 1. Text backbone: Mistral Nemo 12B 2. Vision Adapter: 400M 3. Uses GeLU (for vision adapter) & 2D RoPE (for vision encoder) 4. Larger vocabulary - 131,072 5. Three new special tokens - ‘img’, [image]
Mistral seems to have dropped Pixtral-12b-240910, a multimodal LLM! Curious to see how this one will perform compared to the Lllama 3.1 variants as this one seems to be trained natively as a multimodal model. Still need to download this with my slow internet connection...
Mistral just dropped a new vision multimodal model called Pixtral 12b! Also downloaded params json - GeLU & 2D RoPE are used for the vision adapter. The vocab size also got larger - 131072 Also Mistral's latest tokenizer PR shows 3 extra new tokens (the image, the start & end). […