diff --git a/onediff_comfy_nodes/utils/diffusers_quant_utils.py b/onediff_comfy_nodes/utils/diffusers_quant_utils.py
index d94015105..ca29170b1 100644
--- a/onediff_comfy_nodes/utils/diffusers_quant_utils.py
+++ b/onediff_comfy_nodes/utils/diffusers_quant_utils.py
@@ -35,7 +35,7 @@ def _use_graph():
     os.environ["ONEFLOW_CONV_ALLOW_HALF_PRECISION_ACCUMULATION"] = "1"
     os.environ["ONEFLOW_MATMUL_ALLOW_HALF_PRECISION_ACCUMULATION"] = "1"
     os.environ["ONEFLOW_LINEAR_EMBEDDING_SKIP_INIT"] = "1"
-    os.environ["ONEFLOW_KERNEL_GLU_ENABLE_DUAL_GEMM_IMPL"] = "0"
+    # os.environ["ONEFLOW_KERNEL_GLU_ENABLE_DUAL_GEMM_IMPL"] = "0"
     os.environ["ONEFLOW_MLIR_GROUP_MATMUL_QUANT"] = "1"
     os.environ["ONEFLOW_FUSE_QUANT_TO_MATMUL"] = "0"
     # os.environ["ONEFLOW_MLIR_FUSE_KERNEL_LAUNCH"] = "1"
diff --git a/src/infer_compiler_registry/register_diffusers_quant/quant_diffusion_pipeline.py b/src/infer_compiler_registry/register_diffusers_quant/quant_diffusion_pipeline.py
index 56766813d..282c6eec7 100644
--- a/src/infer_compiler_registry/register_diffusers_quant/quant_diffusion_pipeline.py
+++ b/src/infer_compiler_registry/register_diffusers_quant/quant_diffusion_pipeline.py
@@ -29,7 +29,7 @@ def _use_graph():
     os.environ["ONEFLOW_CONV_ALLOW_HALF_PRECISION_ACCUMULATION"] = "1"
     os.environ["ONEFLOW_MATMUL_ALLOW_HALF_PRECISION_ACCUMULATION"] = "1"
     os.environ["ONEFLOW_LINEAR_EMBEDDING_SKIP_INIT"] = "1"
-    os.environ["ONEFLOW_KERNEL_GLU_ENABLE_DUAL_GEMM_IMPL"] = "0"
+    # os.environ["ONEFLOW_KERNEL_GLU_ENABLE_DUAL_GEMM_IMPL"] = "0"
     os.environ["ONEFLOW_MLIR_GROUP_MATMUL_QUANT"] = "1"
     os.environ["ONEFLOW_FUSE_QUANT_TO_MATMUL"] = "0"
     # os.environ["ONEFLOW_MLIR_FUSE_KERNEL_LAUNCH"] = "1"
diff --git a/src/onediff/infer_compiler/with_oneflow_compile.py b/src/onediff/infer_compiler/with_oneflow_compile.py
index 04b23b69c..21c9ba02a 100644
--- a/src/onediff/infer_compiler/with_oneflow_compile.py
+++ b/src/onediff/infer_compiler/with_oneflow_compile.py
@@ -379,7 +379,7 @@ def __init__(self, model):
         os.environ["ONEFLOW_CONV_ALLOW_HALF_PRECISION_ACCUMULATION"] = "1"
         os.environ["ONEFLOW_MATMUL_ALLOW_HALF_PRECISION_ACCUMULATION"] = "1"
         os.environ["ONEFLOW_LINEAR_EMBEDDING_SKIP_INIT"] = "1"
-        os.environ["ONEFLOW_KERNEL_GLU_ENABLE_DUAL_GEMM_IMPL"] = "0"
+        # os.environ["ONEFLOW_KERNEL_GLU_ENABLE_DUAL_GEMM_IMPL"] = "0"
         os.environ["ONEFLOW_MLIR_GROUP_MATMUL_QUANT"] = "1"
         # TODO: enable this will cause the failure of multi resolution warmup
         # os.environ["ONEFLOW_MLIR_FUSE_KERNEL_LAUNCH"] = "1"