diff --git a/backends/gaudi/server/text_generation_server/models/custom_modeling/flash_qwen3_moe_modeling.py b/backends/gaudi/server/text_generation_server/models/custom_modeling/flash_qwen3_moe_modeling.py
index f1e73f46..583c4ad0 100644
--- a/backends/gaudi/server/text_generation_server/models/custom_modeling/flash_qwen3_moe_modeling.py
+++ b/backends/gaudi/server/text_generation_server/models/custom_modeling/flash_qwen3_moe_modeling.py
@@ -185,7 +185,6 @@ class Qwen3MoeAttention(nn.Module):
                 seqlen=seqlen,
                 softmax_scale=self.scaling,
                 window_size_left=self.max_past,
-                num_key_value_groups=self.num_key_value_groups,
             )
         # Decode
         else: