Update README.md
Browse files
README.md
CHANGED
@@ -278,4 +278,58 @@ This is a dataset built from [CommitPackFT](https://huggingface.co/datasets/bigc
|
|
278 |
- Scala
|
279 |
- Bash
|
280 |
|
281 |
-
The goal of this dataset is to evaluate the ability of models to retrieve a diff given it's instruction.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
278 |
- Scala
|
279 |
- Bash
|
280 |
|
281 |
+
The goal of this dataset is to evaluate the ability of models to retrieve a diff given it's instruction.
|
282 |
+
|
283 |
+
|
284 |
+
### Code To Produce Dataset
|
285 |
+
Below is the code to reproduce this dataset:
|
286 |
+
```py
|
287 |
+
import datasets
|
288 |
+
from tqdm import tqdm
|
289 |
+
import difflib
|
290 |
+
|
291 |
+
|
292 |
+
outrepo = "cassanof/CodeEditSearch"
|
293 |
+
LANGS = ["python", "javascript", "go", "ruby", "java", "php", "c", "c++", "rust", "swift",
|
294 |
+
"typescript", "scala", "kotlin", "r", "perl", "haskell", "lua", "shell", "dart", "julia"]
|
295 |
+
|
296 |
+
processed = []
|
297 |
+
|
298 |
+
|
299 |
+
def get_udiff(a, b):
|
300 |
+
a = a.splitlines()
|
301 |
+
b = b.splitlines()
|
302 |
+
diff = difflib.unified_diff(a, b, lineterm="")
|
303 |
+
return "\n".join(diff)
|
304 |
+
|
305 |
+
|
306 |
+
for lang in tqdm(LANGS):
|
307 |
+
print(f"Processing {lang}")
|
308 |
+
ds = datasets.load_dataset("bigcode/commitpackft", lang, split="train")
|
309 |
+
ds = ds.shuffle(seed=42)
|
310 |
+
print(f"{lang}: {len(ds)}")
|
311 |
+
ds = ds.filter(lambda x: len(
|
312 |
+
x["new_contents"] + x["old_contents"]) < 2500, num_proc=8)
|
313 |
+
ds = ds.filter(lambda x: len(x["new_contents"].strip()) > 0 and len(
|
314 |
+
x["old_contents"].strip()) > 0, num_proc=8)
|
315 |
+
if len(ds) < 2000:
|
316 |
+
print(f"Skipping {lang} due to insufficient data")
|
317 |
+
continue
|
318 |
+
print(f"{lang} after: {len(ds)}")
|
319 |
+
ds = ds.select(range(2000))
|
320 |
+
diffs = [get_udiff(a, b)
|
321 |
+
for a, b in zip(ds["old_contents"], ds["new_contents"])]
|
322 |
+
ds = {
|
323 |
+
"after": ds["new_contents"],
|
324 |
+
"before": ds["old_contents"],
|
325 |
+
"diff": diffs,
|
326 |
+
"instruction": ds["message"],
|
327 |
+
}
|
328 |
+
ds = datasets.Dataset.from_dict(ds)
|
329 |
+
ds = ds.filter(lambda x: len(x["diff"].splitlines()) > 10, num_proc=8)
|
330 |
+
print(f" ******* Final {lang}: {len(ds)} *******")
|
331 |
+
ds.push_to_hub(outrepo, lang)
|
332 |
+
processed.append(lang)
|
333 |
+
|
334 |
+
print(processed)
|
335 |
+
```
|