Question 1

What is GLM-130B?

Accepted Answer

GLM-130B is a 130-billion-parameter dense, pre-trained General Language Model (GLM) released as an open bilingual (Chinese–English) model for research and certain commercial uses.

Question 2

Who developed GLM-130B?

Accepted Answer

It was developed by THUDM (THUKEG), the Knowledge Engineering Group at Tsinghua University, which maintains the broader GLM family.

Question 3

What are the main capabilities of GLM-130B?

Accepted Answer

It supports natural language understanding and generation tasks such as question answering, dialogue, and text completion using the GLM framework.

Question 4

Is GLM-130B open source and how is it licensed?

Accepted Answer

The repository is open, with code and model released under licenses included in the repo for academic research and some limited commercial scenarios; review the repo for exact terms.

Question 5

What languages does GLM-130B support?

Accepted Answer

It is bilingual, supporting both Chinese and English.

Question 6

How can I use GLM-130B in my own projects?

Accepted Answer

Clone the GLM-130B repository from THUDM, follow the setup instructions, and load the provided checkpoints to run inference with the supplied scripts.

Question 7

Where can I find related models or tools from the same team?

Accepted Answer

See the THUDM/THUKEG GitHub organization and their Hugging Face page for GLM, ChatGLM, GLM-4, CogVLM, CodeGeeX, CogView, CogVideoX, and more.

Question 8

How can I contact the GLM-130B/THUDM team?

Accepted Answer

Use the contact email listed on the THUKEG/THUDM GitHub organization page: keg.cs.tsinghua@gmail.com.

Question 9

What training scale was used for GLM-130B?

Accepted Answer

According to the repository, GLM-130B was trained on 400 billion text tokens using the GLM training framework.

Question 10

Does the repository include checkpoints and inference code?

Accepted Answer

Yes. The repo provides downloadable model checkpoints and inference/deployment scripts, including multi-GPU and mixed-precision configurations.

Question 11