Update dataset/model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
tags: [code, assertion-completion, repository-level, ood]
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# code2lora-data-ood — post-cutoff OOD evaluation split
|
| 7 |
+
|
| 8 |
+
Out-of-distribution evaluation slice: commits from repositories whose first
|
| 9 |
+
release is **strictly after** the Qwen2.5-Coder pretraining cutoff. Used to
|
| 10 |
+
isolate cross-repository transfer from training-set memorisation.
|
| 11 |
+
|
| 12 |
+
## Layout
|
| 13 |
+
|
| 14 |
+
```
|
| 15 |
+
commits.parquet per-commit metadata + diff/repo-state embeddings
|
| 16 |
+
qna.parquet per-assertion rows (prefix, target, lineno, ...)
|
| 17 |
+
```
|
| 18 |
+
|
| 19 |
+
Both files share the same `(repo_id, commit_index, commit_sha)` key, so a
|
| 20 |
+
left join on those three columns reconstructs the per-commit QnA bundle.
|
| 21 |
+
|
| 22 |
+
## Schema
|
| 23 |
+
|
| 24 |
+
`commits.parquet` matches `code2lora/code2lora-data-commits/commits/*.parquet`.
|
| 25 |
+
`qna.parquet` matches `code2lora/code2lora-data-snapshots/qna/*.parquet`.
|
| 26 |
+
|
| 27 |
+
## Companion datasets
|
| 28 |
+
|
| 29 |
+
* `code2lora/code2lora-data-commits` in-distribution per-commit dataset
|
| 30 |
+
* `code2lora/code2lora-data-snapshots` static-snapshot variant
|