Update README.md
Browse files
README.md
CHANGED
@@ -36,7 +36,7 @@ question = 'When was Obama inaugurated?'
|
|
36 |
text = f'Text: {passage}.\nQuestion: {question}\nAnswer:{tokenizer.additional_special_tokens[0]}.'
|
37 |
encoded_input = tokenizer(text, return_tensors='pt')
|
38 |
output_ids = model.generate(input_ids=encoded_input.input_ids, attention_mask=encoded_input.attention_mask,
|
39 |
-
eos_token_id=tokenizer.additional_special_tokens_ids[1])
|
40 |
tokenizer.decode(output_ids[0])
|
41 |
```
|
42 |
The generated answer is then `"<pad><extra_id_0> 2009<extra_id_1>"`, while the one generated by the original [T5-v1.1-large](https://huggingface.co/google/t5-v1_1-large) is `"<pad><extra_id_0> On January 20, 2009<extra_id_1>"` - a correct yet non-extractive answer.
|
@@ -59,6 +59,22 @@ The gap between the two models diminishes as more training examples are introduc
|
|
59 |
|
60 |
### BibTeX entry and citation info
|
61 |
```bibtex
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
@misc{castel2021optimal,
|
63 |
title={How Optimal is Greedy Decoding for Extractive Question Answering?},
|
64 |
author={Or Castel and Ori Ram and Avia Efrat and Omer Levy},
|
@@ -66,9 +82,6 @@ The gap between the two models diminishes as more training examples are introduc
|
|
66 |
eprint={2108.05857},
|
67 |
archivePrefix={arXiv},
|
68 |
primaryClass={cs.CL}
|
69 |
-
}
|
70 |
-
<a href="https://huggingface.co/exbert/?model=distilbert-base-uncased">
|
71 |
-
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
|
72 |
-
</a>
|
73 |
-
|
74 |
|
|
36 |
text = f'Text: {passage}.\nQuestion: {question}\nAnswer:{tokenizer.additional_special_tokens[0]}.'
|
37 |
encoded_input = tokenizer(text, return_tensors='pt')
|
38 |
output_ids = model.generate(input_ids=encoded_input.input_ids, attention_mask=encoded_input.attention_mask,
|
39 |
+
eos_token_id=tokenizer.additional_special_tokens_ids[1], num_beams=1, max_length=512, min_length=3)
|
40 |
tokenizer.decode(output_ids[0])
|
41 |
```
|
42 |
The generated answer is then `"<pad><extra_id_0> 2009<extra_id_1>"`, while the one generated by the original [T5-v1.1-large](https://huggingface.co/google/t5-v1_1-large) is `"<pad><extra_id_0> On January 20, 2009<extra_id_1>"` - a correct yet non-extractive answer.
|
59 |
|
60 |
### BibTeX entry and citation info
|
61 |
```bibtex
|
62 |
+
@inproceedings{ram-etal-2021-shot,
|
63 |
+
title = "Few-Shot Question Answering by Pretraining Span Selection",
|
64 |
+
author = "Ram, Ori and
|
65 |
+
Kirstain, Yuval and
|
66 |
+
Berant, Jonathan and
|
67 |
+
Globerson, Amir and
|
68 |
+
Levy, Omer",
|
69 |
+
booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
|
70 |
+
month = aug,
|
71 |
+
year = "2021",
|
72 |
+
address = "Online",
|
73 |
+
publisher = "Association for Computational Linguistics",
|
74 |
+
url = "https://aclanthology.org/2021.acl-long.239",
|
75 |
+
doi = "10.18653/v1/2021.acl-long.239",
|
76 |
+
pages = "3066--3079",
|
77 |
+
},
|
78 |
@misc{castel2021optimal,
|
79 |
title={How Optimal is Greedy Decoding for Extractive Question Answering?},
|
80 |
author={Or Castel and Ori Ram and Avia Efrat and Omer Levy},
|
82 |
eprint={2108.05857},
|
83 |
archivePrefix={arXiv},
|
84 |
primaryClass={cs.CL}
|
85 |
+
}
|
|
|
|
|
|
|
|
|
86 |
|
87 |
+
```
|