TransGPT is an open-source large language model tailored for the transportation industry, aiming to provide practical value in real-world applications. It serves professionals in road engineering, public transit, traffic safety, and urban planning by offering domain-specific knowledge and conversational AI capabilities.

How It Works

TransGPT leverages a transformer-based architecture, with models like TransGPT-7B and TransGPT-MM-6B. It is trained on a substantial corpus of transportation-specific text and dialogue data, augmented with general instruction-tuning datasets. This approach allows it to perform tasks such as traffic prediction, intelligent consultation, and safety education, with a multimodal version (TransGPT-MM) incorporating traffic signs and rules.

Quick Start & Requirements

• Installation: pip install sentencepiece transformers>=4.28.0
• Prerequisites: Python, CUDA-enabled GPU (8xA100 80G recommended for pre-training, 8xA40 45G for instruction tuning).
• Usage: The README provides Python code snippets for direct inference and integration with LangChain.
• Demo: An online demo is available via Gradio, with potential stability issues noted.

Highlighted Details

• Offers both unimodal (TransGPT-7B) and multimodal (TransGPT-MM-6B) versions.
• Includes dedicated datasets for domain pre-training and fine-tuning.
• Evaluated on transportation benchmarks with promising zero-shot performance compared to GPT-4.
• Codebase includes scripts for pre-training and instruction tuning, adapted from MedicalGPT.

Maintenance & Community

• Active development with recent releases of multimodal versions.
• Community discussion via WeChat.
• Roadmap and further details are available through provided links.

Licensing & Compatibility

• Licensed under Apache 2.0.
• Commercial use is permitted upon email application and official approval from duomo_tech@163.com.

Limitations & Caveats

The project disclaims responsibility for any misuse, harmful content generation, or unforeseen issues arising from model or data complexities. Users are advised to exercise caution, implement safety measures, and are responsible for any adverse impacts of public or commercial deployment.