diff --git a/flash_qla/ops/gated_delta_rule/chunk/__init__.py b/flash_qla/ops/gated_delta_rule/chunk/__init__.py
index 1db0116..a2d8ee0 100644
--- a/flash_qla/ops/gated_delta_rule/chunk/__init__.py
+++ b/flash_qla/ops/gated_delta_rule/chunk/__init__.py
@@ -7,10 +7,10 @@ import tilelang
 from flash_qla.utils import l2norm
 from flash_qla.ops.utils import chunk_local_cumsum, group_reduce_vector
 
-if tilelang.contrib.nvcc.get_target_compute_version() == "9.0":
+if tilelang.contrib.nvcc.get_target_compute_version() == "7.5":
     from .hopper import fused_gdr_fwd, fused_gdr_bwd, fused_gdr_h, kkt_solve
 else:
-    raise ValueError("FlashQLA now support sm90 only.")
+    raise ValueError("FlashQLA now support sm75 only.")
 from .cp_context import intra_card_cp_preprocess