Expand description
SqueezeBERT (Iandola et al., 2020)
SqueezeBERT follows the same architecture as BERT, but replaces most matrix multiplications by grouped convolutions. This reduces the number of parameters and speeds up inference.
Structsยง
- Squeeze
Bert Config - SqueezeBert model configuration.
- Squeeze
Bert Encoder - SqueezeBERT encoder.