Beijing Jiaotong University, together with China Computer Federation Smart Transportation Branch and Zuzhi Multi-Model Company, officially released TransGPT, the first comprehensive large-scale transportation model in China. The model has been open sourced on GitHub.
It is reported that TransGPT (TransGPT-7B) is a large text model based on Transformer architecture . This model combines multi-modal structure and real-time scene data call mode to form a comprehensive traffic model as the infrastructure and supplemented by traffic segment industry applications. architecture.
The model uses 346,000 pieces of text data for pre-training in the field, and 58,000 pieces of dialogue data for fine-tuning. The final model can realize traffic situation prediction, intelligent consultation assistant, public transportation service, traffic planning and design, traffic safety education, Functions such as assistance management, traffic accident reporting and analysis, and automated driving assistance systems .
Officials said that relevant departments can use this model to generate the optimal solution for the city's introduction of traffic plans, and create a management system for basic facilities such as traffic lights. The user only needs to propose the departure and arrival places, and the model can generate relevant traffic planning information and carry out a series of travel route planning behaviors.
At the same time, the model has received training on driving laws. Users can obtain detailed policy interpretation information by asking the model, and can also generate a series of safety education content to help relevant industry players conduct traffic safety education.
The official claims that TransGPT is not only completely open to academic research, but also free for commercial use only after applying by mail and obtaining an official commercial license .
TransGPT, as the first general common-sense transportation model in China, is currently a milestone. It can provide information for road engineering, bridge engineering, tunnel engineering, road transportation, waterway transportation, urban public transportation, transportation economy, transportation safety and other industries. Common sense. Based on this, it can be settled in specific traffic application scenarios.