v4.5 dev update. (#3153)

2026-04-08 00:16:05 +08:00
parent 418d38a5de
commit a221da7ccf
265 changed files with 4913 additions and 1478 deletions
--- a/examples/python/CuTeDSL/hopper/grouped_gemm.py
+++ b/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`.
 """