tfivezhang
commited on
Commit
•
424e225
1
Parent(s):
4fc838d
decontamination
Browse files- README.md +16 -1
- tinystories-3.5.parquet +3 -0
- tinystories-4.parquet +3 -0
README.md
CHANGED
@@ -2,4 +2,19 @@
|
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
|
5 |
-
[TinyStories](https://huggingface.co/datasets/roneneldan/TinyStories) cleaned and deduplicated, including 2,205,910 stories from GPT-3.5, and 2,733,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
|
5 |
+
[TinyStories](https://huggingface.co/datasets/roneneldan/TinyStories) cleaned and deduplicated, including 2,205,910 stories from GPT-3.5, and 2,733,747 from GPT-4.
|
6 |
+
|
7 |
+
This dataset has also been decontaminated with respect to the following benchmarks based on n-gram overlap:
|
8 |
+
|
9 |
+
- GLUE (dev set of SST-2, CoLA, QQP, WNLI, RTE, QNLI, MNLI; test set of MPRC)
|
10 |
+
- SIQA, PIQA, QASC, CSQA (all dev set)
|
11 |
+
- BoolQ (dev set)
|
12 |
+
- WinoGrande (dev set)
|
13 |
+
- ANLI (test set)
|
14 |
+
- ARC easy and challenge (test set)
|
15 |
+
- RACE middle and high (test set)
|
16 |
+
- MMLU (dev, val, and test sets)
|
17 |
+
- MATH, GSM8K (test set)
|
18 |
+
- HumanEval (test set)
|
19 |
+
|
20 |
+
2 documents are removed in this step.
|
tinystories-3.5.parquet
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:7cc244d51003fa28d563ef789b923d64253cc63a94076d00717f62eba57abdd4
|
3 |
+
size 794023027
|
tinystories-4.parquet
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f87a6bd5a17114161da3cadcf60bb90cbb66a44b4e6e768babd2004820fb7096
|
3 |
+
size 1127563685
|