We lifted 10 functions out of this paper's own repositories and ran 8 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| see--/natural-question-answering | pwc_unofficial | 8 of 10 |
| Function | Status | Where it lives |
|---|---|---|
| convert_to_functional_api | Ran | see--/natural-question-answering/export_tfhub.py code served (permissive licence) · get_code("3c6f00ab6a9f35b3") |
| enumerate_tags | Ran | see--/natural-question-answering/nq_to_squad.py code served (permissive licence) · get_code("28da9238ac641da4") |
| find_closing_tag | Ran | see--/natural-question-answering/utils_nq.py code served (permissive licence) · get_code("3c9035c96be8a79b") |
| get_add_tokens | Ran | see--/natural-question-answering/utils_nq.py code served (permissive licence) · get_code("62b21856204f94f1") |
| get_f1 | Ran | see--/natural-question-answering/eval_server.py code served (permissive licence) · get_code("47a5135d3347930d") |
| in_shorts | Ran | see--/natural-question-answering/eval_server.py code served (permissive licence) · get_code("c1853b4dcdc05921") |
| read_candidates | Ran | see--/natural-question-answering/utils_nq.py code served (permissive licence) · get_code("b29ee0435be8282b") |
| reshape_df | Ran | see--/natural-question-answering/eval_server.py code served (permissive licence) · get_code("3c42ee2b19a937f1") |
| convert_nq_to_squad | Not yet run | see--/natural-question-answering/nq_to_squad.py code served (permissive licence) · get_code("f9ad382c926b25b3") |
| create_optimizer | Not yet run | see--/natural-question-answering/optimization.py code served (permissive licence) · get_code("21a37595f49175a0") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
This technical note describes a new baseline for the Natural Questions. Our model is based on BERT and reduces the gap between the model F1 scores reported in the original dataset paper and the human upper bound by 30% and 50% relative for the long and short answer tasks respectively. This baseline has been submitted to the official NQ leaderboard at ai.google.com/research/NaturalQuestions. Code, preprocessed data and pretrained model are available at https://github.com/google-research/language/tree/master/language/question_answering/bert_joint.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("1901.08634")
get_code_for_paper("1901.08634")
have("1901.08634")
Connect an agent — have() is free.