From 1726cb64a9b0744b7f516f6c974f81d9ba68a26e Mon Sep 17 00:00:00 2001 From: biondizzle Date: Tue, 2 Jun 2026 08:29:03 +0000 Subject: [PATCH] fix: interleave_l1_weights granularity_bf16 (not granularity) in SE --- dsv4/layers/shared_expert.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dsv4/layers/shared_expert.py b/dsv4/layers/shared_expert.py index a1800ecc..2f92808e 100644 --- a/dsv4/layers/shared_expert.py +++ b/dsv4/layers/shared_expert.py @@ -119,7 +119,7 @@ class Nvfp4SharedExpert: # The fused kernel's SwiGLU epilogue expects granularity-8 interleaved gate/up. # The unfused path (if _fused_swiglu=False) deinterleaves the GEMM output before splitting. if self._fused_swiglu: - l1_stacked = interleave_l1_weights(l1_stacked, granularity=8) + l1_stacked = interleave_l1_weights(l1_stacked, granularity_bf16=8) # Stack weights and convert to K-major self._l1_mat_b = make_b_k_major(l1_stacked) # (1, K_packed, N_packed) self._l2_mat_b = make_b_k_major(l2_stacked)