BoDong's picture
add change
419289e
|
raw
history blame
621 Bytes
---
license: mit
---
# Sparse BERT mini model (uncased)
Finetuned model pruned to 1:4 structured sparsity.
The model is a pruned version of the [BERT mini model](https://huggingface.co/prajjwal1/bert-mini).
## Intended Use
The model can be used for inference with sparsity optimization.
For further details on the model and its usage will be soon available.
## Evaluation Results
We get the following results on the sst2 tasks development set:
| Task | SST-2 (Acc) |
|------|-------------|
| | 87.2 |
Better than dense [bert mini](https://huggingface.co/M-FAC/bert-mini-finetuned-sst2) which is 84.74%.