Real-time technology can achieve all these features with minimum resources: mobile experience, conversation-and-visual comprehension, and multimodal interaction. Thus, BlueLM-3B by Vivo has become quite an interesting subject for technology enthusiasts.
What is BlueLM-3B talking about?
BlueLM-V-3B is a multimodal language-and-image-understanding model, developed for mobile or edge devices - designed by Vivo AI and MMLab of the Chinese University of Hong Kong.
The language portion of that model consists of approximately 2.7 billion parameters, whereas the image part has 400 million parameters.
The model represents open-source average scores from the OpenCompass benchmark of 66.1 among ≤4B, outperforming most of the larger models.
With 24.4 tokens/s to his credit in production speed, this makes a very worthy AI performer for mobile processors.
Performance and Resource Proficiency of Vivo BlueLM-3B
In balanced mode of power usage comparisons, Blue center 7B truly showed productivity improvements of ~ 46% under mode conditions that were bettered by Blue LM-3B, compared to its predecessor.
At ~1.4GB memory footprint, this model is a huge threshold with overheads for mobile.
Then, it has a great capacity to understand the Graphics User Interface: the model analyzes the screen UI elements: buttons, menus, and icons.
The model has two states: thought mode and no-thought mode, and there is a budgeting mechanism called thinking budget control, which is a concatenation of imagination and depth of work.
Cloud Number One on the Sub-10 Billion Count - Rankings and Awards
In fact, this model is said to be the most standing on its own among models in the range of 7B-9B shelf.
Forums have claimed it as the best in the sub-10B category based on SuperCLUE and Equal Eval tests.
Challenges and Future Opportunities
However, this model is quite competent but needs to be localized around many different cultures and languages around the world. Bengali and its environment remain a huge place of concern in this case.
Challenge in the implementation of the model is due to less RAM, battery being consumed by mobiles, among other hardware restrictions and constraints.
If Vivo plans to make this model available in global variants in the future, along with the enhancements of the Bengali language model, the experiences of users in the Bengali language will surely receive a further upgrade.
Thus, Vivo's BlueLM-3B stands as one of those "sub-10B mobile AI models" in high-performance, resource-efficient, and multi-modal capabilities. Only time will tell its relevance.
Follow our WhatsApp channel for the latest news and updates
