InternLM-XComposer-2.5: A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output

We present InternLM-XComposer-2.5 (IXC-2.5), a versatile large-visionlanguage model that supports long-contextual input and output. IXC-2.5 excelsin various text-image comprehension and composition applications, achievingGPT-4V level capabilities with merely 7B LLM backend. Trained with 24Kinterleaved image-text contexts, it can seamlessly extend to 96K long contextsvia RoPE extrapolation. This long-context capability allows IXC-2.5 to excel intasks requiring extensive input and output contexts. Compared to its previous2.0 version, InternLM-XComposer-2.5 features three major upgrades invision-language comprehension: (1) Ultra-High Resolution Understanding, (2)Fine-Grained Video Understanding, and (3) Multi-Turn Multi-Image Dialogue. Inaddition to comprehension, IXC-2.5 extends to two compelling applications usingextra LoRA parameters for text-image composition: (1) Crafting Webpages and (2)Composing High-Quality Text-Image Articles. IXC-2.5 has been evaluated on 28benchmarks, outperforming existing open-source state-of-the-art models on 16benchmarks. It also surpasses or competes closely with GPT-4V and Gemini Pro on16 key tasks. The InternLM-XComposer-2.5 is publicly available athttps://github.com/InternLM/InternLM-XComposer.