Update README.md
#5
by 1rubbishyuan - opened
README.md
CHANGED
|
@@ -3,29 +3,36 @@ license: mit
|
|
| 3 |
configs:
|
| 4 |
- config_name: train
|
| 5 |
data_files:
|
| 6 |
-
|
| 7 |
-
|
| 8 |
- config_name: single_test
|
| 9 |
data_files:
|
| 10 |
-
|
| 11 |
-
|
| 12 |
- config_name: multiple_test
|
| 13 |
data_files:
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
rai:
|
| 18 |
-
|
|
|
|
|
|
|
| 19 |
|
| 20 |
-
|
|
|
|
| 21 |
|
| 22 |
-
|
|
|
|
| 23 |
|
| 24 |
-
|
|
|
|
| 25 |
|
| 26 |
-
|
|
|
|
|
|
|
| 27 |
|
| 28 |
-
|
| 29 |
---
|
| 30 |
|
| 31 |
## dataset structure
|
|
|
|
| 3 |
configs:
|
| 4 |
- config_name: train
|
| 5 |
data_files:
|
| 6 |
+
- split: train
|
| 7 |
+
path: train.jsonl
|
| 8 |
- config_name: single_test
|
| 9 |
data_files:
|
| 10 |
+
- split: train
|
| 11 |
+
path: single_test.jsonl
|
| 12 |
- config_name: multiple_test
|
| 13 |
data_files:
|
| 14 |
+
- split: train
|
| 15 |
+
path: multiple_test.jsonl
|
| 16 |
+
|
| 17 |
+
rai:dataLimitations: >
|
| 18 |
+
The current benchmark focuses on code generation scenarios with verifiable outcomes.
|
| 19 |
+
Extending it to broader domains, such as factual knowledge tasks, would help evaluate
|
| 20 |
+
the generality of our findings in more realistic settings.
|
| 21 |
|
| 22 |
+
rai:dataBiases: >
|
| 23 |
+
We do not identify significant systematic biases in the current benchmark construction.
|
| 24 |
|
| 25 |
+
rai:personalSensitiveInformation: >
|
| 26 |
+
The dataset does not intentionally include personal or sensitive information.
|
| 27 |
|
| 28 |
+
rai:dataUseCases: >
|
| 29 |
+
SE-Bench is intended for research on self-evolution and knowledge internalization.
|
| 30 |
|
| 31 |
+
rai:dataSocialImpact: >
|
| 32 |
+
The benchmark provides a controlled and clean environment for studying the self-evolution
|
| 33 |
+
capabilities of language models.
|
| 34 |
|
| 35 |
+
rai:hasSyntheticData: true
|
| 36 |
---
|
| 37 |
|
| 38 |
## dataset structure
|