diff --git a/ggml-qnn.cpp b/ggml-qnn.cpp
index 4eeb048f2ef9b2..918e2fb4a59277 100644
--- a/ggml-qnn.cpp
+++ b/ggml-qnn.cpp
@@ -1,8 +1,4 @@
 /*
- * MIT license
- * Copyright (C) 2024 GGML Authors
- * SPDX-License-Identifier: MIT
- *
  * this is implementation of ggml QNN(Qualcomm Neural Network, aka AI Engine Direct) backend
  *
  * status:
@@ -959,7 +955,7 @@ static int free_qnn_tensor(Qnn_Tensor_t & tensor) {
         QNN_LOG_INFO("it should not happen, pls check");
     } else {
         //TODO:why crash in here? why pointer changed with mul_mat?
-        //memory leak after comment above line
+        //memory leak after comment below line
         //free(QNN_TENSOR_GET_DIMENSIONS(tensor));
     }
 
@@ -1464,7 +1460,7 @@ static void ggml_qnn_log_internal(ggml_log_level level, const char * file, const
 #if (defined __ANDROID__) || (defined ANDROID)
             __android_log_print(level, "llamacpp", "%s", s_ggml_qnn_log_internal_buf);
 #else
-            printf("%s", buffer); //Qualcomm's QNN could running on Window over ARM
+            printf("%s", buffer); //Qualcomm's QNN could running on Windows over ARM(aka WoA)
 #endif
         }
         va_end(args);
@@ -2108,11 +2104,11 @@ int qnn_instance::load_system() {
 
     _system_lib_handle = dlopen(system_lib_path.c_str(), RTLD_NOW | RTLD_LOCAL);
     if (nullptr == _system_lib_handle) {
-        QNN_LOG_WARN("can not pen QNN library %s, error: %s\n", system_lib_path.c_str(), dlerror());
+        QNN_LOG_WARN("can not open QNN library %s, error: %s\n", system_lib_path.c_str(), dlerror());
         return 1;
     }
 
-    auto *get_providers = reinterpret_cast<_pfn_QnnSystemInterface_getProviders *>(dlsym(
+    auto * get_providers = reinterpret_cast<_pfn_QnnSystemInterface_getProviders *>(dlsym(
             _system_lib_handle, "QnnSystemInterface_getProviders"));
     if (nullptr == get_providers) {
         QNN_LOG_WARN("can not load QNN symbol QnnSystemInterface_getProviders: %s\n", dlerror());
@@ -3605,7 +3601,6 @@ bool ggml_qnn_compute_forward(struct ggml_compute_params * params, struct ggml_t
     }
 
 
-    //ok, real show time in Qualcomm's QNN internal
     if (nullptr != func)
         func(tensor->src[0], tensor->src[1], tensor);
     if (nullptr != func_common)
@@ -3845,7 +3840,7 @@ static size_t ggml_backend_qnn_buffer_type_get_alignment(ggml_backend_buffer_typ
 static size_t ggml_backend_qnn_buffer_type_get_max_size(ggml_backend_buffer_type_t buft) {
     GGML_UNUSED(buft);
 
-    return (38 * 1024 * 1024);
+    return (96 * 1024 * 1024);
 }
 
 
diff --git a/ggml-qnn.h b/ggml-qnn.h
index dae4b476134299..75602fafd30159 100644
--- a/ggml-qnn.h
+++ b/ggml-qnn.h
@@ -1,8 +1,4 @@
 /*
- * MIT license
- * Copyright (C) 2024 GGML Authors
- * SPDX-License-Identifier: MIT
- *
  * this is implementation of ggml QNN(Qualcomm Nerual Network, aka AI Engine Direct) backend
  */
 #pragma once