GLM-130B
Tsinghua University · 2022-08-04
GLM-130B is a bilingual English-Chinese bidirectional dense Transformer released by Tsinghua University and collaborators in August 2022. Its pretraining combined GLM autoregressive blank infilling for 95% of tokens with multi-task instruction pretraining for 5%; the team reports over 400 billion bilingual training tokens, so the exact cumulative exposure is undisclosed.
Benchmark scores
| Benchmark | Score |
|---|---|
| BoolQ | 78.4 |
| GSM8K | 6.1 |