diff --git a/moonep/grad_reduce.py b/moonep/grad_reduce.py index 7a5c37f..937597e 100644 --- a/moonep/grad_reduce.py +++ b/moonep/grad_reduce.py @@ -292,7 +292,7 @@ def kernel( cd_state = pipeline.make_pipeline_state(pipeline.PipelineUserType.Consumer, STAGES) acc_tid = tidx - 32 H_PER = cutlass.const_expr(TILE_ELEMS // ACC_THREADS) - acc_reg = cute.make_fragment((H_PER,), Float32) + acc_reg = cute.make_rmem_tensor((H_PER,), Float32) # ============ phase 1: warp-specialized accumulate, no clears ======== total_work = acnt[0] * TILES_PER_EXPERT diff --git a/setup.py b/setup.py index cd50e28..f45e591 100644 --- a/setup.py +++ b/setup.py @@ -70,7 +70,7 @@ version="0.0.1", packages=find_packages(include=["moonep", "moonep.*"]), install_requires=[ - "nvidia-cutlass-dsl==4.4.2", + "nvidia-cutlass-dsl==4.6.0", ], ext_modules=ext_modules, cmdclass={"build_ext": BuildExtension},