diff --git a/mojo_opset/backends/ttx/kernels/npu/__init__.py b/mojo_opset/backends/ttx/kernels/npu/__init__.py
index bd5c5108..bb487f87 100755
--- a/mojo_opset/backends/ttx/kernels/npu/__init__.py
+++ b/mojo_opset/backends/ttx/kernels/npu/__init__.py
@@ -49,6 +49,7 @@
 from .swa import swa_paged_prefill_impl
 from .swiglu import swiglu_bwd_impl
 from .swiglu import swiglu_fwd_impl
+from .int8_gemm import int8_gemm_dequant_impl, prepare_b_impl
 
 # triton-dist based comm kernels (requires triton_dist + shmem packages)
 allgather_gemm_impl = None
@@ -131,4 +132,6 @@
     "allgather_gemm_impl",
     "gemm_allreduce_impl",
     "gemm_reduce_scatter_impl",
+    "int8_gemm_dequant_impl", 
+    "prepare_b_impl",
 ]
diff --git a/mojo_opset/backends/ttx/kernels/npu/int8_gemm.py b/mojo_opset/backends/ttx/kernels/npu/int8_gemm.py
index 88ae0e1a..81e35439 100644
--- a/mojo_opset/backends/ttx/kernels/npu/int8_gemm.py
+++ b/mojo_opset/backends/ttx/kernels/npu/int8_gemm.py
@@ -110,7 +110,7 @@ def _pad_to(x, mult):
     return ((x + mult - 1) // mult) * mult
 
 
-def prepare_b(b: torch.Tensor) -> torch.Tensor:
+def prepare_b_impl(b: torch.Tensor) -> torch.Tensor:
     """Transpose B to (N, K) row-major and pad to block boundaries.
 
     For inference: weight B is fixed, call once and reuse.