v4.5 dev update. (#3153)

examples/python/CuTeDSL/hopper/fmha.py

@@ -527,7 +527,6 @@ class HopperFusedMultiHeadAttentionForward:
                 grid=grid,
                 block=[self.threads_per_cta, 1, 1],
                 cluster=self.cluster_shape_mnk,
-                smem=self.shared_storage.size_in_bytes(),
                 stream=stream,
                 min_blocks_per_mp=1,
             )

examples/python/CuTeDSL/hopper/grouped_gemm.py

@@ -107,7 +107,7 @@ Constraints (same as dense_gemm_persistent.py plus):
 * Cluster shape M/N: power of 2, total <= 4
 * Contiguous dim must be 16-byte aligned
 
-Debug environment knobs:
+Debug environment options:
 * `GROUPED_GEMM_FORCE_CUTE_COPY=1`
     Disable the non-mcast NVVM TMA load path and always use `cute.copy`.
 """