Publish crosscodeeval_v2 filtered by both models pass@16 results

Retain 863 of 6710 v1 tasks solved by Qwen3-30B Instruct or Thinking; archive 5847 newly unsolved tasks alongside 1053 prior exclusions. Preserve all task payloads, reuse pinned v1 images, and expose per-model success counts and selection examples.

crosscodeeval_v2/README.md

@@ -0,0 +1,84 @@
+# CrossCodeEval v2
+
+## Dataset stats
+
+- **6,710 v1 tasks → 863 retained; 5,847 newly archived.** All 863 retain their v1 repairs; v2 changes no task instructions, environments, verifiers or reference solutions.
+- **7,763 original tasks: 863 retained and 6,900 archived**, including 1,053 earlier v1 exclusions.
+- **Images:** the original 3 image definitions were not consolidated. Retained tasks reuse the 3 [v1 cached images](https://huggingface.co/datasets/FWeindel/validated-tasks/tree/main/crosscodeeval_environment-reliable/cached_images), pinned by immutable revision and checksum in the Parquet. No image copies are needed in v2.
+
+## Dataset characteristics
+
+Cross-file code completion in C#, Java, Python and TypeScript, with related-file context and executable verifiers.
+
+<details><summary>Domains</summary>
+
+Software Engineering; Code Completion; Cross-file Repository Understanding.
+
+</details>
+
+<details><summary>Capabilities</summary>
+
+Read related files, resolve identifiers, follow existing code patterns and complete missing statements.
+
+</details>
+
+**Likely transfer targets:** cross-file code completion and repository understanding.
+
+## Pass@16 filtering
+
+Keep a task if either model solved it at least once. Both models completed 16 attempts on every v1 task. Recorded terminal failures count as unsuccessful attempts; unsolved tasks are not necessarily defective.
+
+| Model | Tasks evaluated | Attempts | Tasks solved at least once |
+| --- | ---: | ---: | ---: |
+| Qwen3-30B Instruct | 6,710 | 107,360 | 641 |
+| Qwen3-30B Thinking | 6,710 | 107,360 | 627 |
+| Either model | 6,710 | 214,720 | **863** |
+
+Successful-attempt distribution **among the 863 retained tasks**. Each model column totals 863; rows with zero tasks for both models are omitted. A task with 0/16 for one model was solved by the other.
+
+| Successful attempts | Instruct tasks | Thinking tasks |
+| --- | ---: | ---: |
+| 0/16 | 222 | 236 |
+| 1/16 | 222 | 193 |
+| 2/16 | 70 | 94 |
+| 3/16 | 57 | 64 |
+| 4/16 | 40 | 51 |
+| 5/16 | 38 | 36 |
+| 6/16 | 33 | 22 |
+| 7/16 | 33 | 24 |
+| 8/16 | 20 | 24 |
+| 9/16 | 22 | 21 |
+| 10/16 | 17 | 16 |
+| 11/16 | 21 | 16 |
+| 12/16 | 18 | 19 |
+| 13/16 | 17 | 12 |
+| 14/16 | 13 | 13 |
+| 15/16 | 7 | 7 |
+| 16/16 | 13 | 15 |
+
+The retained tasks inherit v1's rule-based repairs, build checks, reference reward 1 and no-op reward 0. Teacher runs used the same task contents without the timeout/anti-cheating sentence; v2 preserves that v1 sentence. [V1 repair details](https://huggingface.co/datasets/FWeindel/validated-tasks/blob/main/crosscodeeval_environment-reliable/README.md) · [Teacher results](https://huggingface.co/datasets/FWeindel/validated-tasks/discussions/6).
+
+## Select tasks by successful-attempt count
+
+`tasks.parquet` includes `instruct_successes`, `thinking_successes` and `attempts_per_model` (16). Counts are also available in [selection.json](assets/selection.json). Earlier v1 archives have null teacher counts because they were not evaluated.
+
+```python
+import pyarrow.parquet as pq
+
+# Exactly 1 successful attempt out of 16 for Instruct:
+filters = [("instruct_successes", "=", 1)]
+
+# Alternatively, at least 3 successes for either model:
+# filters = [[("instruct_successes", ">=", 3)],
+#            [("thinking_successes", ">=", 3)]]
+
+subset = pq.read_table("tasks.parquet", filters=filters)
+pq.write_table(subset, "selected.parquet")
+```
+
+`patch.py` reproduces the release from the pinned v1 Parquets and recorded counts.
+
+## Runtime
+
+- **Versions:** Apptainer **1.5.4**, Python **3.12.14**, Harbor **0.8.1**.
+- **Implementation:** [Marin Harbor `7faf878c14b6`](https://github.com/marin-community/harbor/tree/7faf878c14b6d72737579ec936ebf5f74ba3194d) + [our runtime patches](https://github.com/MLI-lab/OT-next-data/tree/main/harbor_patches).

crosscodeeval_v2/archive.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbfd98d499cb56d9c85c8ab2faf7ec97a30c8ed2699f487350ba4ae27c0b953a
+size 92382213

crosscodeeval_v2/patch.py

@@ -0,0 +1,136 @@
+"""Reproduce CrossCodeEval v2 from pinned v1 Parquets and the recorded pass@16 results.
+
+pip install pyarrow huggingface_hub
+python patch.py --out ./rebuilt
+Use --source-dir PATH to reuse the two original v1 Parquets locally.
+"""
+import argparse
+import copy
+import hashlib
+import json
+import io
+import tarfile
+from pathlib import Path
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+MODELS = {'instruct_successes': 'qwen3-30b-instruct-2507',
+          'thinking_successes': 'qwen3-30b-thinking'}
+
+def rows(path):
+    table = pq.ParquetFile(path)
+    for i in range(table.num_row_groups):
+        yield from table.read_row_group(i).to_pylist()
+
+def restrict_images(manifest, names):
+    result = copy.deepcopy(manifest)
+    result['tasks'] = {k: v for k, v in manifest['tasks'].items() if k in names}
+    bundles = {e['bundle'] for t in result['tasks'].values() for e in t['environments']}
+    result['bundles'] = {k: v for k, v in manifest['bundles'].items() if k in bundles}
+    result['missing'] = []
+    return result
+
+def build(source_dir, out, selection_path):
+    selection = json.loads(selection_path.read_text())
+    out.mkdir(parents=True, exist_ok=True)
+    if any((out / n).exists() for n in ('tasks.parquet', 'archive.parquet')):
+        raise ValueError('Output Parquets already exist; choose an empty output directory')
+    sources = {}
+    for name in ('tasks.parquet', 'archive.parquet'):
+        if source_dir is None:
+            from huggingface_hub import hf_hub_download
+            path = Path(hf_hub_download(selection['repo_id'], selection['source_folder'] + '/' + name,
+                        repo_type='dataset', revision=selection['source_revision'], cache_dir=out / '.source-cache'))
+        else:
+            path = source_dir / name
+        with path.open('rb') as handle:
+            digest = hashlib.file_digest(handle, 'sha256').hexdigest()
+        if digest != selection['source_sha256'][name]:
+            raise ValueError('Pinned source hash mismatch: ' + name)
+        sources[name] = path
+    tasks = selection['tasks']
+    kept = {name for name, task in tasks.items() if task['kept']}
+    assert len(tasks) == 6710 and len(kept) == 863
+    original_schema = pq.read_schema(sources['tasks.parquet'])
+    image_manifest = json.loads(original_schema.metadata[b'ot.images.v1'])
+    images = restrict_images(image_manifest, kept)
+    assert set(images['tasks']) == kept
+    release = {k: v for k, v in selection.items() if k != 'tasks'}
+    release['selection_file'] = 'assets/selection.json'
+    metadata = {b'ot.release.v2': json.dumps(release).encode(), b'ot.images.v1': json.dumps(images).encode()}
+    schema = original_schema.remove_metadata().append(pa.field('instruct_successes', pa.int64())).append(
+        pa.field('thinking_successes', pa.int64())).append(pa.field('attempts_per_model', pa.int64())).append(pa.field('archive_category', pa.string()))
+    seen, audit = set(), {}
+    with pq.ParquetWriter(out / 'tasks.parquet', schema.with_metadata(metadata), compression='zstd') as retain, \
+         pq.ParquetWriter(out / 'archive.parquet', schema.with_metadata({b'ot.release.v2': json.dumps(release).encode()}), compression='zstd') as archive:
+        counts = {'retained': 0, 'newly_archived': 0, 'prior_archived': 0}
+        for row in rows(sources['tasks.parquet']):
+            name = row['path']
+            if name in seen or name not in tasks:
+                raise ValueError('Duplicate or unexpected task: ' + name)
+            seen.add(name)
+            result = tasks[name]
+            assert row['content_sha256'] == result['content_sha256']
+            for column, model in MODELS.items():
+                rewards = result['models'][model]['rewards']
+                assert len(rewards) == 16 and result['models'][model]['complete'] and result['models'][model]['missing'] == 0
+                row[column] = sum(reward >= 1 for reward in rewards)
+                assert row[column] == result['models'][model]['solved_attempts']
+            row['attempts_per_model'] = 16
+            selected = row['instruct_successes'] > 0 or row['thinking_successes'] > 0
+            assert selected == result['kept']
+            audit[name] = {'binary_sha256': hashlib.sha256(row['task_binary']).hexdigest(),
+                           'content_sha256': row['content_sha256'], 'kept': selected}
+            if selected:
+                row['stages_passed'] = ','.join(dict.fromkeys(row['stages_passed'].split(',') + ['6']))
+                retain.write_table(pa.Table.from_pylist([row], schema=schema.with_metadata(metadata)))
+                counts['retained'] += 1
+            else:
+                row['archive_stage'] = 6
+                row['archive_reason'] = 'pass16-unsolved: no successful attempt in 16 Instruct and 16 Thinking trials; this is not a confirmed task defect.'
+                row['archive_labels'] = list(row.get('archive_labels') or []) + ['pass16-unsolved']
+                archive.write_table(pa.Table.from_pylist([row], schema=schema))
+                counts['newly_archived'] += 1
+        assert seen == set(tasks)
+        for row in rows(sources['archive.parquet']):
+            assert row['path'] not in seen
+            seen.add(row['path'])
+            # V1's original-source archive predates the validation columns.
+            # Preserve its category and reason while computing its content hash.
+            with tarfile.open(fileobj=io.BytesIO(row['task_binary']), mode='r:*') as archive_tar:
+                files = [(str(Path(m.name)), archive_tar.extractfile(m).read())
+                         for m in archive_tar.getmembers() if m.isfile()]
+            digest = hashlib.sha256()
+            for name, content in sorted(files):
+                encoded = name.encode()
+                digest.update(len(encoded).to_bytes(8, 'big')); digest.update(encoded)
+                digest.update(len(content).to_bytes(8, 'big')); digest.update(content)
+            row['content_sha256'] = digest.hexdigest()
+            row['archive_stage'] = 0
+            row['stages_passed'] = ''
+            for column in MODELS:
+                row[column] = None
+            row['attempts_per_model'] = None
+            audit[row['path']] = {'binary_sha256': hashlib.sha256(row['task_binary']).hexdigest(),
+                                  'content_sha256': row['content_sha256'], 'kept': False}
+            archive.write_table(pa.Table.from_pylist([row], schema=schema))
+            counts['prior_archived'] += 1
+    assert counts == {'retained': 863, 'newly_archived': 5847, 'prior_archived': 1053}
+    checked = set()
+    for filename, expected_kept in [('tasks.parquet', True), ('archive.parquet', False)]:
+        for row in rows(out / filename):
+            prior = audit[row['path']]
+            assert row['path'] not in checked and prior['kept'] == expected_kept
+            assert hashlib.sha256(row['task_binary']).hexdigest() == prior['binary_sha256']
+            assert row['content_sha256'] == prior['content_sha256']
+            checked.add(row['path'])
+    assert checked == seen
+    print(json.dumps(dict(counts, preserved_task_payloads=len(checked), cached_bundles_reused=len(images['bundles']))))
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--source-dir', type=Path)
+    parser.add_argument('--out', type=Path, required=True)
+    parser.add_argument('--selection', type=Path, default=Path(__file__).parent / 'assets' / 'selection.json')
+    args = parser.parse_args()
+    build(args.source_dir, args.out, args.selection)

crosscodeeval_v2/tasks.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21a25eac1b374c251f67f76e7abbb865435eeba329c7598ba210ba274d31f648
+size 11270178