Remove unnecessary pyright ignores

scripts/user/tomtseng/mmlu_pro_tar_251203/compute_mmlu_accuracy.py

@@ -61,7 +61,7 @@ def build_ground_truth() -> list[dict[str, Any]]:
     """Load and preprocess MMLU-Pro test split, same as tamperbench."""
     ds = datasets.load_dataset(DATASET_NAME, split="test")
     # HuggingFace datasets library doesn't properly type Dataset iteration
-    rows = [dict(item) for item in ds]  # pyright: ignore[reportCallIssue, reportArgumentType]
+    rows = [dict(item) for item in ds]
     rows = preprocess(rows)
     return limit_examples_per_subject(rows, VAL_MAX_EXAMPLES_PER_SUBJECT)
 

scripts/user/tomtseng/mmlu_pro_tar_251203/eval_mmlu_pro_zeroshot.py

@@ -50,7 +50,7 @@ def build_ground_truth() -> list[dict[str, Any]]:
     """Load and preprocess MMLU-Pro test split, returning the same 140 questions."""
     ds = datasets.load_dataset(DATASET_NAME, split="test")
     # HuggingFace datasets library doesn't properly type Dataset iteration
-    rows = [dict(item) for item in ds]  # pyright: ignore[reportCallIssue, reportArgumentType]
+    rows = [dict(item) for item in ds]
     rows = preprocess(rows)
     return limit_examples_per_subject(rows, VAL_MAX_EXAMPLES_PER_SUBJECT)
 

src/tamperbench/whitebox/attacks/base.py

@@ -55,7 +55,7 @@ class TamperAttackConfig:
     random_seed: int
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> Self:  # pyright: ignore[reportExplicitAny]
+    def from_dict(cls, data: dict[str, Any]) -> Self:
         """All subclasses must implement a `from_dict` constructor.
 
         Args:
@@ -70,7 +70,7 @@ def from_dict(cls, data: dict[str, Any]) -> Self:  # pyright: ignore[reportExpli
         return cls(**data)
 
     @classmethod
-    def prepare_dict_for_init(cls, data: dict[str, Any]) -> dict[str, Any]:  # pyright: ignore[reportExplicitAny]
+    def prepare_dict_for_init(cls, data: dict[str, Any]) -> dict[str, Any]:
         """Transform dictionary data into proper types for dataclass initialization.
 
         Converts nested dictionaries into their corresponding dataclass instances
@@ -82,14 +82,14 @@ def prepare_dict_for_init(cls, data: dict[str, Any]) -> dict[str, Any]:  # pyrig
         Returns:
             dict[str, Any]: Dictionary with nested objects properly instantiated.
         """
-        model_config_dict = data.pop("model_config")  # pyright: ignore[reportAny]
-        model_config = ModelConfig.from_dict(model_config_dict)  # pyright: ignore[reportAny]
+        model_config_dict = data.pop("model_config")
+        model_config = ModelConfig.from_dict(model_config_dict)
 
         data.update({"model_config": model_config})
         return data
 
     @classmethod
-    def _validate_dict_keys(cls, data: dict[str, Any]) -> None:  # pyright: ignore[reportExplicitAny]
+    def _validate_dict_keys(cls, data: dict[str, Any]) -> None:
         """Check that the dictionary to be used to construct dataclass has correct fields.
 
         Args:

src/tamperbench/whitebox/attacks/embedding_attack/embedding_attack.py

@@ -31,7 +31,7 @@ class EmbeddingAttackConfig(TamperAttackConfig):
     soft_opt_config: SoftOptConfig
 
     @classmethod
-    def prepare_dict_for_init(cls, data: dict[str, Any]) -> dict[str, Any]:  # pyright: ignore[reportExplicitAny]
+    def prepare_dict_for_init(cls, data: dict[str, Any]) -> dict[str, Any]:
         """Transform dictionary data into proper types for dataclass initialization.
 
         Args:
@@ -42,11 +42,11 @@ def prepare_dict_for_init(cls, data: dict[str, Any]) -> dict[str, Any]:  # pyrig
         """
         data = super().prepare_dict_for_init(data)
 
-        soft_opt_config_dict = data.pop("soft_opt_config")  # pyright: ignore[reportAny]
+        soft_opt_config_dict = data.pop("soft_opt_config")
 
         # Pass random_seed from attack config to SoftOptConfig
         soft_opt_config_dict["seed"] = data["random_seed"]
-        soft_opt_config = SoftOptConfig(**soft_opt_config_dict)  # pyright: ignore[reportAny]
+        soft_opt_config = SoftOptConfig(**soft_opt_config_dict)
 
         data.update({"soft_opt_config": soft_opt_config})
         return data

src/tamperbench/whitebox/evals/ifeval/ifeval.py

@@ -90,7 +90,7 @@ def compute_inferences(self) -> DataFrame[InferenceSchema]:
         models like Qwen, Llama, etc.
         """
         dataset = datasets.load_dataset(DATASET_NAME, split=TEST_SPLIT)
-        test_docs = [dict(item) for item in dataset]  # pyright: ignore[reportArgumentType]
+        test_docs = [dict(item) for item in dataset]
 
         if self.eval_config.max_samples is not None:
             test_docs = test_docs[: self.eval_config.max_samples]
@@ -114,7 +114,7 @@ def compute_inferences(self) -> DataFrame[InferenceSchema]:
     def compute_scores(self, inferences: DataFrame[InferenceSchema]) -> DataFrame[ScoreSchema]:
         """Apply instruction-following checks (utils.process_results)."""
         dataset = datasets.load_dataset(DATASET_NAME, split=TEST_SPLIT)
-        test_docs = [dict(item) for item in dataset][: len(inferences)]  # pyright: ignore[reportArgumentType]
+        test_docs = [dict(item) for item in dataset][: len(inferences)]
 
         self._prompt_level_strict = []
         self._prompt_level_loose = []
@@ -132,8 +132,8 @@ def compute_scores(self, inferences: DataFrame[InferenceSchema]) -> DataFrame[Sc
             metrics = utils.process_results(doc, response)
             strict_prompt = float(metrics["prompt_level_strict_acc"])
             loose_prompt = float(metrics["prompt_level_loose_acc"])
-            inst_strict = list(metrics["inst_level_strict_acc"])  # pyright: ignore[reportArgumentType]
-            inst_loose = list(metrics["inst_level_loose_acc"])  # pyright: ignore[reportArgumentType]
+            inst_strict = list(metrics["inst_level_strict_acc"])
+            inst_loose = list(metrics["inst_level_loose_acc"])
 
             self._prompt_level_strict.append(strict_prompt)
             self._prompt_level_loose.append(loose_prompt)

src/tamperbench/whitebox/evals/mbpp/mbpp.py

@@ -225,7 +225,7 @@ class MBPPEvaluation(WhiteBoxEvaluation[MBPPEvaluationConfig]):
     def compute_inferences(self) -> DataFrame[InferenceSchema]:
         """Generate code completions for MBPP problems."""
         dataset = datasets.load_dataset(DATASET_PATH, split=TEST_SPLIT)
-        test_docs = [dict(item) for item in dataset]  # pyright: ignore[reportArgumentType]
+        test_docs = [dict(item) for item in dataset]
 
         # Validate dataset size
         assert len(test_docs) == 500, (
@@ -261,7 +261,7 @@ def _format_with_chat_template(self, prompt: str) -> str:
     def compute_scores(self, inferences: DataFrame[InferenceSchema]) -> DataFrame[ScoreSchema]:
         """Execute generated code and compute pass/fail scores."""
         dataset = datasets.load_dataset(DATASET_PATH, split=TEST_SPLIT)
-        test_docs = [dict(item) for item in dataset][: len(inferences)]  # pyright: ignore[reportArgumentType]
+        test_docs = [dict(item) for item in dataset][: len(inferences)]
 
         # Get prompts and references
         prompts = [get_prompt(doc) for doc in test_docs]

src/tamperbench/whitebox/evals/minerva_math/minerva_math.py

@@ -83,7 +83,7 @@ def compute_inferences(self) -> DataFrame[InferenceSchema]:
         for subset in DATASET_SUBSETS:
             dataset = datasets.load_dataset(DATASET_NAME, subset, split=TEST_SPLIT)
             processed = utils.process_docs(dataset)
-            all_test_docs.extend([dict(item) for item in processed])  # pyright: ignore[reportArgumentType]
+            all_test_docs.extend([dict(item) for item in processed])
 
         # Get few-shot examples
         fewshot_examples = utils.list_fewshot_samples()
@@ -111,7 +111,7 @@ def compute_scores(self, inferences: DataFrame[InferenceSchema]) -> DataFrame[Sc
         for subset in DATASET_SUBSETS:
             dataset = datasets.load_dataset(DATASET_NAME, subset, split=TEST_SPLIT)
             processed = utils.process_docs(dataset)
-            all_test_docs.extend([dict(item) for item in processed])  # pyright: ignore[reportArgumentType]
+            all_test_docs.extend([dict(item) for item in processed])
 
         # Extract and normalize answers from responses
         responses_list = list(inferences[InferenceSchema.response])

src/tamperbench/whitebox/evals/mmlu_pro/mmlu_pro.py

@@ -111,7 +111,7 @@ def compute_scores(self, inferences: DataFrame[InferenceSchema]) -> DataFrame[Sc
 
     @override
     def compute_results(self, scores: DataFrame[ScoreSchema]) -> DataFrame[EvaluationSchema]:
-        acc = float(pl.Series(scores[ScoreSchema.score]).mean())  # pyright: ignore[reportArgumentType]
+        acc = float(pl.Series(scores[ScoreSchema.score]).mean())
         df = pl.from_dict(
             {
                 EvaluationSchema.metric_name: [str(MetricName.MMLU_PRO_ACCURACY)],
@@ -126,7 +126,7 @@ def _load_split(self, split_name: str) -> list[dict]:
             if SPLIT_SAMPLE_LIMIT is None or SPLIT_SAMPLE_LIMIT <= 0
             else f"{split_name}[:{SPLIT_SAMPLE_LIMIT}]"
         )
-        ds = datasets.load_dataset(DATASET_NAME, split=split)  # pyright: ignore[reportArgumentType]
+        ds = datasets.load_dataset(DATASET_NAME, split=split)
         rows = [dict(item) for item in ds]
         return mmlu_api.preprocess(rows)
 

src/tamperbench/whitebox/utils/analysis/builders.py

@@ -43,9 +43,9 @@ class HeatmapData:
     utility_delta: NDArray[np.float32]
     models: list[str]
     attacks: list[AttackName]
-    metadata: dict[str, Any] = field(default_factory=dict)  # pyright: ignore[reportExplicitAny]
+    metadata: dict[str, Any] = field(default_factory=dict)
 
-    def to_dict(self) -> dict[str, Any]:  # pyright: ignore[reportExplicitAny]
+    def to_dict(self) -> dict[str, Any]:
         """Convert to dictionary for JSON serialization.
 
         Converts numpy arrays to lists and attack names to strings for JSON compatibility.
@@ -68,7 +68,7 @@ def to_dict(self) -> dict[str, Any]:  # pyright: ignore[reportExplicitAny]
             "metadata": self.metadata,
         }
 
-    def _to_list(self, array: NDArray[np.float32]) -> list[Any] | Any:  # pyright: ignore[reportExplicitAny]
+    def _to_list(self, array: NDArray[np.float32]) -> list[Any] | Any:
         """Convert numpy array to list, or return as-is if not an array."""
         return array.tolist() if hasattr(array, "tolist") else array