Un-ignore ruff E741; add noqas

pyproject.toml

@@ -19,7 +19,6 @@ extend-select = [
 ignore = [
     "C901", # Function too complex
     "E501", # Line length (handled by ruff-format)
-    "E741", # Ambiguous variable name
     "UP007", # X | Y style Unions
 ]
 

src/peft/tuners/adalora/model.py

@@ -245,7 +245,7 @@ def forward(self, *args, **kwargs):
             for n, p in self.model.named_parameters():
                 if ("lora_A" in n or "lora_B" in n) and self.trainable_adapter_name in n:
                     para_cov = p @ p.T if "lora_A" in n else p.T @ p
-                    I = torch.eye(*para_cov.size(), out=torch.empty_like(para_cov))
+                    I = torch.eye(*para_cov.size(), out=torch.empty_like(para_cov))  # noqa: E741
                     I.requires_grad = False
                     num_param += 1
                     regu_loss += torch.norm(para_cov - I, p="fro")

src/peft/tuners/oft/layer.py

@@ -243,7 +243,7 @@ def _cayley_batch(self, data: torch.Tensor) -> torch.Tensor:
         b, r, c = data.shape
         # Ensure the input matrix is skew-symmetric
         skew = 0.5 * (data - data.transpose(1, 2))
-        I = torch.eye(r, device=data.device).unsqueeze(0).expand(b, r, c)
+        I = torch.eye(r, device=data.device).unsqueeze(0).expand(b, r, c)  # noqa: E741
 
         # Perform the Cayley parametrization
         Q = torch.bmm(I - skew, torch.inverse(I + skew))
@@ -267,7 +267,7 @@ def _block_diagonal(self, oft_r: torch.Tensor, rank: int) -> torch.Tensor:
     def _project_batch(self, oft_r, eps=1e-5):
         # scaling factor for each of the smaller block matrix
         eps = eps * 1 / torch.sqrt(torch.tensor(oft_r.shape[0]))
-        I = (
+        I = (  # noqa: E741
             torch.zeros((oft_r.size(1), oft_r.size(1)), device=oft_r.device, dtype=oft_r.dtype)
             .unsqueeze(0)
             .expand_as(oft_r)