diff --git a/deep_gemm/mega/__init__.py b/deep_gemm/mega/__init__.py
index ca60573..213054a 100644
--- a/deep_gemm/mega/__init__.py
+++ b/deep_gemm/mega/__init__.py
@@ -317,11 +317,12 @@ def fp8_nvfp4_mega_moe(y: torch.Tensor,
     Activation format: E2M1 packed uint8 + UE4M3 scales (computed by staging kernel)
     Recipe: (1, 1, 16) — kGranK=16 for NVFP4 group_size=16.
     """
-    for name, t in [("l1_w", l1_weights), ("l1_w_sf", l1_weights_sf),
-                     ("l2_w", l2_weights), ("l2_w_sf", l2_weights_sf)]:
+    l1_w, l1_w_sf = l1_weights
+    l2_w, l2_w_sf = l2_weights
+    for name, t in [("l1_w", l1_w), ("l1_w_sf", l1_w_sf),
+                     ("l2_w", l2_w), ("l2_w_sf", l2_w_sf)]:
         print(f"[debug] {name}: dtype={t.dtype} shape={tuple(t.shape)} contig={t.is_contiguous()}", flush=True)
 
-    # Also check symm buffer views
     for name, t in [("sym_x", sym_buffer.x), ("sym_x_sf", sym_buffer.x_sf),
                      ("sym_l1_acts", sym_buffer.l1_acts), ("sym_l1_acts_sf", sym_buffer.l1_acts_sf),
                      ("sym_l2_acts", sym_buffer.l2_acts), ("sym_l2_acts_sf", sym_buffer.l2_acts_sf)]: