About Model Performance

[zgzjdx]

Hi!
Why i just got about 53% acc on dev set? What's the result you get on the dev set?

[nishanthta]

Try using this dataset (~10k examples)
https://github.com/brmson/dataset-sts/tree/master/data/sts/sick2014

I was able to bump it to about 91% on this dev set using the same model. Just make sure to bump up the embedding dimension and the number of LSTM units accordingly