This is distilled model from Bert Base uncased. It has 6 layers, 6 heads and 384 hidden Size. It has 29.8M parameter. Performance wise, it has the potential of 87% performance of bert base with has 12 layers and 12 heads with 110M parameters.
This is distilled model from Bert Base uncased. It has 6 layers, 6 heads and 384 hidden Size. It has 29.8M parameter. Performance wise, it has the potential of 87% performance of bert base with has 12 layers and 12 heads with 110M parameters.