Model
Inference & Serving
CogVLM2
1Zhipu 2AI's 3visual 4expert 5language 6model
the six Ws · specification
W1
Who
Zhipu AI (THUDM) built CogVLM2 as a successor to the original CogVLM.
W2
What
A vision-language model using a visual expert module to fuse image and text features deeply.
W3
Where
Released on Hugging Face and GitHub under the THUDM organization.
W4
When
CogVLM2 was released in May 2024, with CogVLM originally launched in 2023.
W5
Why
It pioneered deep visual-language feature fusion, achieving strong results on OCR and chart-understanding benchmarks.
W6
With
Runs on PyTorch and transformers, built on a Llama3-based language backbone with a dedicated visual expert.
W7
Watch
Open-weight release by Zhipu AI/THUDM; free for research and, with registration, commercial use under its own license.
works with