conda env create -f ODQA.yml
conda activate ODQA
Use Quesntion encoder and Passage encoder model of Dense Passage Retrieval which is trained on multi data. Then, use dot-product to perform similarity of embeddings.
python Retrieval/download_pretrained.py
python download_data.py
python Reader/download_pretrained.py
python Reader/train.py
We trained ELECTRA model on SQuADv2. You can download weights model here
| Epoch | Loss | Exact Match |
|---|---|---|
| 1 | 1.1731 | 0.83 |
| 2 | 0.7471 | 0.84 |
| 3 | 0.6016 | 0.85 |
python app.py
