diff --git a/onnxruntime/core/mlas/lib/q4_dq_cli.cpp b/onnxruntime/core/mlas/lib/q4_dq_cli.cpp
index b994f171c6..5cc66da357 100644
--- a/onnxruntime/core/mlas/lib/q4_dq_cli.cpp
+++ b/onnxruntime/core/mlas/lib/q4_dq_cli.cpp
@@ -254,13 +254,14 @@ dequantize(const Cli& cli)
         out.write((const char*)dstbuf.data(), std::streamsize(dstbuf.size()) * sizeof(float));
     } else {
         std::streambuf* buf;
+        std::ofstream file_output_stream;
         if (cli.output_file) {
-            std::ofstream out(cli.output_file, std::ios::out);
-            if (!out) {
+            file_output_stream.open(cli.output_file, std::ios::out);
+            if (file_output_stream.fail()) {
                 std::cerr << "Cannot open output file " << cli.output_file << std::endl;
                 return -1;
             }
-            buf = out.rdbuf();
+            buf = file_output_stream.rdbuf();
         } else {
             buf = std::cout.rdbuf();
         }
diff --git a/onnxruntime/python/tools/quantization/matmul_weight4_quantizer.py b/onnxruntime/python/tools/quantization/matmul_weight4_quantizer.py
index 44d870bb22..921e02fb69 100644
--- a/onnxruntime/python/tools/quantization/matmul_weight4_quantizer.py
+++ b/onnxruntime/python/tools/quantization/matmul_weight4_quantizer.py
@@ -189,7 +189,7 @@ class MatMulWeight4Quantizer:
                         # recursive call to take care of sub-graph
                         graph_stack.append(attr.g)
                         kv = {attr.name: self._process_subgraph(graph_stack)}
-                    elif attr.type == onnx.AttributeProto.GRAPH:
+                    elif attr.type == onnx.AttributeProto.GRAPHS:
                         value = []
                         for subgraph in attr.graphs:
                             # recursive call to take care of sub-graph