diff --git a/vibe_kernels/loss/benchmark.py b/vibe_kernels/loss/benchmark.py
index 5b483c1..947c7fc 100644
--- a/vibe_kernels/loss/benchmark.py
+++ b/vibe_kernels/loss/benchmark.py
@@ -92,11 +92,11 @@ def cutedsl_impl():
         baseline()
         try:
             triton_impl()
-        except:
+        except Exception:
             pass
         try:
             cutedsl_impl()
-        except:
+        except Exception:
             pass
 
     # --- Quack ---
@@ -177,7 +177,7 @@ def quack_impl():
                 .view(-1)
             )
             tri_diff = torch.max(torch.abs(ref - ours_tri)).item()
-        except:
+        except Exception:
             tri_diff = -1.0
 
         # CuTeDSL
diff --git a/vibe_kernels/sampling/benchmark_all_topk.py b/vibe_kernels/sampling/benchmark_all_topk.py
index 9440f3d..b40b65d 100755
--- a/vibe_kernels/sampling/benchmark_all_topk.py
+++ b/vibe_kernels/sampling/benchmark_all_topk.py
@@ -110,7 +110,7 @@ def benchmark_quack(
             else:
                 # Cutlass doesn't have cuda module in this version
                 return None
-        except:
+        except Exception:
             return None
 
         from quack.topk import topk
diff --git a/vibe_kernels/vibe_attention/interface.py b/vibe_kernels/vibe_attention/interface.py
index b5ac717..930b0c3 100644
--- a/vibe_kernels/vibe_attention/interface.py
+++ b/vibe_kernels/vibe_attention/interface.py
@@ -1276,7 +1276,7 @@ def backward(ctx, dout, *args):
         q, k, v, out, lse, cu_seqlens_q, cu_seqlens_k, seqused_q, seqused_k = (
             ctx.saved_tensors
         )
-        assert seqused_q == seqused_k == None
+        assert seqused_q is None and seqused_k is None
         assert ctx.softcap == 0.0
         dq, dk, dv = _flash_attn_bwd(
             q,