MAINT: P311-19 hash pinned classifier datasets with hashlib.file_digest

pyrit/score/float_scale/local_violence_classifier_scorer.py

@@ -547,10 +547,11 @@ def compute_dataset_hashes() -> dict[str, str]:
             dict[str, str]: File name to current SHA-256, for re-pinning after a deliberate
             dataset update.
         """
-        return {
-            file_name: hashlib.sha256((_HARM_EVALS_PATH / file_name).read_bytes()).hexdigest()
-            for file_name in _TRAINING_DATASETS
-        }
+        hashes: dict[str, str] = {}
+        for file_name in _TRAINING_DATASETS:
+            with (_HARM_EVALS_PATH / file_name).open("rb") as dataset_file:
+                hashes[file_name] = hashlib.file_digest(dataset_file, "sha256").hexdigest()
+        return hashes
 
     def _build_identifier(self) -> ComponentIdentifier:
         """

pyrit/score/true_false/local_refusal_classifier_scorer.py

@@ -554,10 +554,11 @@ def compute_dataset_hashes() -> dict[str, str]:
             dict[str, str]: File name to current SHA-256, for re-pinning after a deliberate
             dataset update.
         """
-        return {
-            file_name: hashlib.sha256((_REFUSAL_EVALS_PATH / file_name).read_bytes()).hexdigest()
-            for file_name in _TRAINING_DATASETS
-        }
+        hashes: dict[str, str] = {}
+        for file_name in _TRAINING_DATASETS:
+            with (_REFUSAL_EVALS_PATH / file_name).open("rb") as dataset_file:
+                hashes[file_name] = hashlib.file_digest(dataset_file, "sha256").hexdigest()
+        return hashes
 
     def _build_identifier(self) -> ComponentIdentifier:
         """

tests/unit/score/test_local_refusal_classifier_scorer.py

@@ -57,6 +57,15 @@ def test_training_dataset_pins() -> None:
     assert LocalRefusalClassifierScorer.compute_dataset_hashes() == module._TRAINING_DATASETS
 
 
+def test_compute_dataset_hashes_reads_current_files(tmp_path: Path) -> None:
+    contents = {file_name: f"{file_name}\n".encode() * 50_000 for file_name in module._TRAINING_DATASETS}
+    for file_name, data in contents.items():
+        (tmp_path / file_name).write_bytes(data)
+    with patch.object(module, "_REFUSAL_EVALS_PATH", tmp_path):
+        hashes = LocalRefusalClassifierScorer.compute_dataset_hashes()
+    assert hashes == {file_name: hashlib.sha256(data).hexdigest() for file_name, data in contents.items()}
+
+
 def test_training_rejects_changed_data(tmp_path: Path) -> None:
     objective = "context " * 2000
     response = "response " * 1000

tests/unit/score/test_local_violence_classifier_scorer.py

@@ -268,6 +268,15 @@ def test_compute_dataset_hashes_matches_pins() -> None:
     assert LocalViolenceClassifierScorer.compute_dataset_hashes() == module._TRAINING_DATASETS
 
 
+def test_compute_dataset_hashes_reads_current_files(tmp_path: Path) -> None:
+    contents = {file_name: f"{file_name}\n".encode() * 50_000 for file_name in module._TRAINING_DATASETS}
+    for file_name, data in contents.items():
+        (tmp_path / file_name).write_bytes(data)
+    with patch.object(module, "_HARM_EVALS_PATH", tmp_path):
+        hashes = LocalViolenceClassifierScorer.compute_dataset_hashes()
+    assert hashes == {file_name: hashlib.sha256(data).hexdigest() for file_name, data in contents.items()}
+
+
 class _Tokenizer:
     """Tiny tokenizer with visible token IDs and the pinned encoder's framing budget."""