Apple has made such an innovation that Android users will be very jealous

Apple’s research team said, “MM1They made great progress in the field of artificial intelligence with their new multi-modal big language model called “. MM1has remarkable capabilities in terms of both image recognition and natural language processing capabilities. The model is offered in three different sizes: 3 billion, 7 billion and 30 billion parameters.

Researchers conducted various experiments on these models and identified the key factors affecting performance. Interestingly, image resolution and number of image tags have more impact than visual-language connectors. Additionally, it appears that different pretraining datasets can significantly affect the effectiveness of the model.

Apple has made such an innovation that Android users will be very jealous

MM1“Blend of Experts” architecture and “Top-2 Door” method were used in the development of . This approach yielded excellent results both in pre-training and in current multimodal tasks. Even when fine-tuned for specific tasks, MM1 models continue to deliver competitive performance.

Tests performed, MM1-3B-Chat ve
MM1-7B-Chat reveals that their models are superior to most similarly sized competitors on the market. These models come to the fore in tasks such as answering questions about images and text, answering questions on text-based images, and answering scientific questions. However, MM1’s overall performance has yet to surpass that of Google’s Gemini or OpenAI.
GPT-4V It hasn’t outgrown its models. While MM1 is not an absolute leader, it represents a significant step forward in the field of artificial intelligence for Apple.