Bump ruff version (#367)

* bump ruff version

* make style

Author: a-r-r-o-w

.github/workflows/pr_tests.yml

@@ -21,7 +21,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          pip install ruff==0.1.5
+          pip install ruff==0.9.10
       - name: Check quality
         run: make quality
       - name: Check if failure

finetrainers/args.py

@@ -842,9 +842,9 @@ def _map_to_args_type(args: Dict[str, Any]) -> BaseArgs:
 
 def _validate_model_args(args: BaseArgs):
     if args.training_type == "full-finetune":
-        assert (
-            "transformer" not in args.layerwise_upcasting_modules
-        ), "Layerwise upcasting is not supported for full-finetune training"
+        assert "transformer" not in args.layerwise_upcasting_modules, (
+            "Layerwise upcasting is not supported for full-finetune training"
+        )
 
 
 def _validate_dataset_args(args: BaseArgs):

finetrainers/data/dataset.py

@@ -436,9 +436,9 @@ def __init__(
     ) -> None:
         super().__init__()
 
-        assert weights == -1 or isinstance(
-            weights, dict
-        ), "`weights` must be a dictionary of probabilities for each caption column"
+        assert weights == -1 or isinstance(weights, dict), (
+            "`weights` must be a dictionary of probabilities for each caption column"
+        )
 
         self.dataset_name = dataset_name
         self.infinite = infinite
@@ -530,9 +530,9 @@ def __init__(
     ) -> None:
         super().__init__()
 
-        assert weights == -1 or isinstance(
-            weights, dict
-        ), "`weights` must be a dictionary of probabilities for each caption column"
+        assert weights == -1 or isinstance(weights, dict), (
+            "`weights` must be a dictionary of probabilities for each caption column"
+        )
 
         self.dataset_name = dataset_name
         self.infinite = infinite

finetrainers/utils/__init__.py

@@ -37,7 +37,7 @@ def get_parameter_names(obj: Any, method_name: Optional[str] = None) -> Set[str]
 
 
 def get_non_null_items(
-    x: Union[List[Any], Tuple[Any], Dict[str, Any]]
+    x: Union[List[Any], Tuple[Any], Dict[str, Any]],
 ) -> Union[List[Any], Tuple[Any], Dict[str, Any]]:
     if isinstance(x, dict):
         return {k: v for k, v in x.items() if v is not None}

finetrainers/utils/torch.py

@@ -266,10 +266,10 @@ def _get_total_norm(
     if len(tensors) == 0:
         return torch.tensor(0.0)
     first_device = tensors[0].device
-    grouped_tensors: dict[
-        tuple[torch.device, torch.dtype], tuple[list[list[torch.Tensor]], list[int]]
-    ] = _group_tensors_by_device_and_dtype(
-        [tensors]  # type: ignore[list-item]
+    grouped_tensors: dict[tuple[torch.device, torch.dtype], tuple[list[list[torch.Tensor]], list[int]]] = (
+        _group_tensors_by_device_and_dtype(
+            [tensors]  # type: ignore[list-item]
+        )
     )  # type: ignore[assignment]
 
     norms: List[torch.Tensor] = []
@@ -308,9 +308,9 @@ def _clip_grads_with_norm_(
     max_norm = float(max_norm)
     if len(grads) == 0:
         return
-    grouped_grads: dict[
-        Tuple[torch.device, torch.dtype], Tuple[List[List[torch.Tensor]], List[int]]
-    ] = _group_tensors_by_device_and_dtype([grads])  # type: ignore[assignment]
+    grouped_grads: dict[Tuple[torch.device, torch.dtype], Tuple[List[List[torch.Tensor]], List[int]]] = (
+        _group_tensors_by_device_and_dtype([grads])
+    )  # type: ignore[assignment]
 
     clip_coef = max_norm / (total_norm + 1e-6)
     # Note: multiplying by the clamped coef is redundant when the coef is clamped to 1, but doing so

requirements.txt

@@ -16,3 +16,4 @@ torchao>=0.7.0
 sentencepiece>=0.2.0
 imageio-ffmpeg>=0.5.1
 numpy>=1.26.4
+ruff==0.9.10

tests/models/ltx_video/_test_tp.py

@@ -151,7 +151,7 @@ def main(world_size: int, rank: int):
         torch.allclose(output, output_tp, atol=1e-5, rtol=1e-5),
         (output - output_tp).abs().max(),
     )
-    print(f"Max memory reserved ({rank=}): {torch.cuda.max_memory_reserved(rank) / 1024 ** 3:.2f} GB")
+    print(f"Max memory reserved ({rank=}): {torch.cuda.max_memory_reserved(rank) / 1024**3:.2f} GB")
 
     if rank == 0:
         print()