Tencent Mixed Element Large Model is a large-scale pre-trained language model developed by Tencent. According to the search results, Tencent officially released the hybrid model at the Global Digital Ecology Conference on September 7, 2023. The model has a parameter size of more than 10 billion, and the pre-trained corpus exceeds 2 trillion tokens. The hybrid model has been tested in internal businesses and products such as Tencent Cloud, Tencent Advertising, and WeChat Sou.
The hybrid model not only serves the internal business, but also can be applied to various industry scenarios, and customers can call the mixed element model based on APIs or make exclusive industry models based on it. Compared with ChatGPT, the Tencent hybrid model has strong performance in natural language understanding, and compared with ChatGPT, the hybrid model has stronger performance, but it may still have certain shortcomings compared with GPT4. It is worth mentioning that Tencent's hybrid model has also achieved certain achievements in the field of visual common sense reasoning, and has performed well in the international authoritative list in related fields.
Specifically, the mixed-element large model has made an important breakthrough in the task of visual common sense reasoning in the field of multimodal understanding, and has topped the international authoritative list VCR (Visual Commonsense Reasoning). In this task, the model not only needs to identify the perception level of the image, but also needs to carry out cognitive-level perception, such as judgment intention, logical reasoning, etc., with similar thinking ability to humans.
This breakthrough shows the potential of mixed-element large models in the multimodal field, and may be applied to a wider range of fields, such as intelligent interaction, intelligent recommendation and intelligent decision-making. For example, in intelligent interactions, mixed-element large models can better understand the text and images entered by the user and provide more accurate and personalized responses. In intelligent recommendation, the mixed-element large model can better recommend content suitable for the user's taste based on the user's visual and language information. In intelligent decision-making, the mixed-element large model can synthesize information from multiple modalities to make more comprehensive and accurate decisions
Compared with ChatGPT, Tencent's hybrid model performs better in the following aspects:
Larger parameter scale: The parameter size of the mixed-element large model exceeds 10 billion, compared to the smaller parameter scale of ChatGPT.
Pre-trained corpus is richer: The pre-trained corpus of mixed-element large models exceeds 2 trillion tokens, more than ChatGPT.
Application testing in internal business and products: The hybrid model has been tested in internal businesses and products such as Tencent Cloud, Tencent Advertising, and WeChat Sou, showing strong performance.
Support for multimodality: In contrast, ChatGPT is mainly biased towards natural language, while mixed-element large models support multimodality.
Excellent performance in visual common sense reasoning: Tencent Mixed Element Model has achieved certain achievements in the field of visual common sense reasoning, and has performed well in the international authoritative list in related fields.
Although the mixed-element large model is stronger than ChatGPT in terms of natural language understanding, it may still have certain shortcomings compared to GPT4#Tencent#
