Yuan 2.0
modelLLM series (2.1B, 51B, 102.6B) introducing Localized Filtering-based Attention (LFA) to incorporate local dependency priors. Strong capabilities in code generation, math, and Chinese reasoning. The 102B version was highly regarded for long-form Chinese text and specialized reasoning in legal and financial sectors.
Model Details
Architecture DENSE
Parameters 102B
Variants
| Name | Parameters | Notes |
|---|---|---|
| Yuan2-2B | 2.1B | — |
| Yuan2-51B | 51B | — |
| Yuan2-102B | 102.6B | — |
Paper
arXiv: 2311.15786