XingChen-AGI/TeleOCR
Image-Text-to-Text • 1B • Updated • 40.7k • 1.28k
AI & ML interests
XingChen-AGI is the open-source organization, dedicated to building agent-era large language models. Our flagship Xing4.0-29B-A4B is a 29B MoE model with only 4B activated parameters, natively supporting 256K context (extensible to 512K). With 4-bit quantization, it runs on a single consumer GPU (~15GB VRAM) and scores 93.52 on SuperCLUE Agent benchmark.