Skip to content

Latest commit

 

History

History
74 lines (49 loc) · 2.29 KB

README.md

File metadata and controls

74 lines (49 loc) · 2.29 KB

Datasets

Catalogue:

1. WIT Benchmark:

Unzip the file (wit.zip) under the current directory.

**** The resulting directory looks like

.
├── ./wit/                    
    └── wit_test_set.json # The test of WIT dataset

2. XSum Benchmark:

We provide the in-context learning examples as well as the test set of XSum benchmark that are used in our experiments.

**** The directory looks like

.
├── ./xsum/                    
    ├── /one-shot/ # Contains the one-shot in-context examples.
        ├── xsum_train_one-shot-1.json # The first random one-shot in-context learning example.
        ├── xsum_train_one-shot-2.json # The second random one-shot in-context learning example.
        └── xsum_train_one-shot-3.json # The third random one-shot in-context learning example.
    ├── /two-shot/ # Contains the two-shot in-context examples
        ├── xsum_train_two-shot-1.json # The first random two-shot in-context learning example.
        ├── xsum_train_two-shot-2.json # The second random two-shot in-context learning example.
        └── xsum_train_two-shot-3.json # The third random two-shot in-context learning example.
    └── xsum_test.json # The test set of XSum benchmark.

3. IWSLT14 De-En Benchmark:

We provide the IWSLT14 De-En dataset that is used in our experiments.

**** After unzipping, the directory looks like

.
├── ./translation/iwslt14/de-to-en/                    
    ├── train.json # The training set of IWSLT14 De-En dataset.
    ├── validation.json # The validation set of IWSLT14 De-En dataset.
    └── test.json # The test set of IWSLT14 De-En dataset.

4. WebText Benchmark:

We provide the held-out set of WebText.

**** The directory looks like

.
├── ./webtext/                    
    └── webtext.test.jsonl # The held-out set of the WebText benchmark.