| --- |
| pretty_name: Code Functional Identity Tasks |
| configs: |
| - config_name: cluster_similar |
| data_files: |
| - split: train |
| path: data/cluster_similar*.parquet |
| - config_name: pick_similar |
| data_files: |
| - split: train |
| path: data/pick_similar*.parquet |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/*.parquet |
| --- |
| |
| ## Code Functional Identity |
|
|
| Tasks related to the functional identity of code: i.e do these two codes achieve the same outcome? |
|
|
| Data sourced from Aletheia artifacts (codes and accompanying pass rates. Codes with PR=1.0 for the same prompt are functionally identical) |
|
|
| Two categories |
| - pick_similar: Given a reference and 5 candidates, choose the ones that are similar to the reference (multiple correct) |
| - cluster_similar: Given 5 candidates, cluster them into groups that are functionally identical (e.g. (1, 2, 4), (3)) |
