Puyu Lingbi 162

浦语·灵笔 is a vision-language large model developed based on the 书生·浦语 large language model, providing outstanding image-text understanding and generation capabilities, with several advantages:

Image-Text Interleaved Creation: 浦语·灵笔 can create exclusive articles with both engaging text and beautiful images for users. The generated articles feature brilliant prose complemented by well-matched illustrations, delivering an immersive reading experience. This capability is achieved through the following steps:

Understanding user instructions and creating long articles that meet the requirements.

Intelligently analyzing the article, automatically planning ideal positions for illustrations, and determining the required image content.

Multi-level intelligent filtering to select the most perfect images from the gallery.

Image-Text Understanding Based on Rich Multimodal Knowledge: 浦语·灵笔 has designed an efficient training strategy that injects massive multimodal concepts and knowledge data into the model, endowing it with powerful image-text understanding and dialogue capabilities.

Outstanding Performance: 浦语·灵笔 has achieved the best performance on multiple mainstream benchmarks for vision-language large models, including MME Benchmark (English evaluation), MMBench (English evaluation), Seed-Bench (English evaluation), CCBench (Chinese evaluation), and MMBench-CN (Chinese evaluation).

The open-sourced 浦语·灵笔 includes two versions:

InternLM-XComposer-VL-7B: A multimodal pre-training and multi-task training model based on the 书生·浦语 large language model, demonstrating outstanding performance on various benchmarks, such as MME Benchmark, MMBench, Seed-Bench, CCBench, and MMBench-CN.

InternLM-XComposer-7B: A fine-tuned model oriented towards image-text interleaved article creation and intelligent dialogue.

浦语·灵笔 is now open-sourced and available on GitHub, Hugging Face, and 魔搭社区 (ModelScope).