diff --git a/CMakeLists.txt b/CMakeLists.txt
index 786b1322b346631d1570a6ebd9c572302531db4e..288be06b78ac6d6439ea48b39a2e86ee0c364a97 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -24,6 +24,11 @@ endif(WITH_PADDLE_MOBILE)
 
 # set(CMAKE_BUILD_TYPE DEBUG)
 
+
+SET(CMAKE_BUILD_TYPE "Release")
+SET(CMAKE_CXX_FLAGS_RELEASE "$ENV{CXXFLAGS} -O3 -Wall")
+
+
 set(PADDLE_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR})
 set(PADDLE_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR})
 set(CMAKE_CXX_STANDARD 11)
diff --git a/lite/backends/fpga/KD/llapi/filter.cpp b/lite/backends/fpga/KD/llapi/filter.cpp
index b6932bc27f0019af58cea00e4b5422396d838208..54c10f50af1d2f3cd4244d11f0d8dbcba95b35e2 100755
--- a/lite/backends/fpga/KD/llapi/filter.cpp
+++ b/lite/backends/fpga/KD/llapi/filter.cpp
@@ -234,8 +234,8 @@ int8_t* format_filter(float* data_in,
     float* filter_start = data_in + n * chw;
     float f_max = find_max(filter_start, chw);
     int8_t* quantized_start = quantized_data + n * chw;
-    quantize(filter_start, quantized_start, chw, max);
-    filter_max.push_back(max);
+    quantize(filter_start, quantized_start, chw, f_max);
+    filter_max.push_back(f_max);
   }
 
   int8_t* hwc_data =
diff --git a/lite/backends/fpga/KD/llapi/zynqmp_api.cpp b/lite/backends/fpga/KD/llapi/zynqmp_api.cpp
index 68d0b6c68b722f9c5cf31139ed7308516889bd8c..b8b7dfb8876ea5381b0100b3bcfb701f6c5c2ca6 100644
--- a/lite/backends/fpga/KD/llapi/zynqmp_api.cpp
+++ b/lite/backends/fpga/KD/llapi/zynqmp_api.cpp
@@ -187,7 +187,7 @@ int get_device_info(const struct DeviceInfo &args) {
 int perform_bypass(const struct BypassArgs &args) {
   int ret = -1;
   int size = args.image.channels * args.image.width * args.image.height;
-  int max_size = 1 << 21;
+  int max_size = 1 << 22;
 
   float times = 1.0 * size / max_size;
   int count = static_cast<int>(times);
diff --git a/lite/backends/fpga/KD/pes/input_pe.hpp b/lite/backends/fpga/KD/pes/input_pe.hpp
index abae094d97e66f1ecff78f1e41199893456e6722..d8f9a15c6a94f6869a588f758982800b35eecc40 100755
--- a/lite/backends/fpga/KD/pes/input_pe.hpp
+++ b/lite/backends/fpga/KD/pes/input_pe.hpp
@@ -29,28 +29,20 @@ class InputPE : public PE {
   }
 
   bool dispatch() {
-    std::cout << "input_dispatch()\n";
+    // std::cout << "input_dispatch()\n";
     Tensor* input = param_.input;
     Tensor* output = param_.output;
 
     Tensor* src = input;
-    // std::cout << "input:" << input << std::endl;
     input->flush();
-    // std::cout << "input_flush()\n";
     Tensor half_tensor;
     if (input->dataType() == DataType::FP32) {
-      // std::cout << "2()\n";
       half_tensor.mutableData<void*>(DataType::FP16, input->shape());
-      // std::cout << "3()\n";
       half_tensor.copyFrom(input);
-      // std::cout << "4()\n";
       src = &half_tensor;
     }
-    // std::cout << "5()\n";
     output->mutableData<void>();
-    // std::cout << "6()\n";
     src->alignImage(output, true);
-    // std::cout << "7()\n";
     return true;
   }
 
diff --git a/lite/backends/fpga/KD/pes/output_pe.hpp b/lite/backends/fpga/KD/pes/output_pe.hpp
index 53da0c5be768d05d1ca58abf26a4a4ab0425cc9f..2d02d30fbae12efc372e58c2ad80348356a8f22d 100755
--- a/lite/backends/fpga/KD/pes/output_pe.hpp
+++ b/lite/backends/fpga/KD/pes/output_pe.hpp
@@ -54,8 +54,10 @@ class OutputPE : public PE {
            output->data<void>(),
            output->shape().numel() * sizeof(float));
 
-    // auto max = fpga_get_memory_size_max();
-    // std::cout << "===== Max: ===== :: " << max << std::endl;
+    fpga_reset();
+
+    auto max = fpga_get_memory_size_max();
+    std::cout << "PL ===== Max: ===== :: " << max << std::endl;
 
     return true;
   }
diff --git a/lite/backends/fpga/KD/pes/pooling_pe.hpp b/lite/backends/fpga/KD/pes/pooling_pe.hpp
index 84ed4f946e1a394cb0fc40d7c156faf534e1f8db..bec99a539490513d79ad86a4d5e3b2f4168c24e4 100755
--- a/lite/backends/fpga/KD/pes/pooling_pe.hpp
+++ b/lite/backends/fpga/KD/pes/pooling_pe.hpp
@@ -67,7 +67,7 @@ class PoolingPE : public PE {
 
     use_cpu_ = output->shape().width() == 1 && output->shape().height() == 1 &&
                (k_width > 255 || k_height > 255);
-    use_cpu_ = param_.type == AVERAGE;
+    // use_cpu_ = param_.type == AVERAGE;
   }
 
   void compute() {
diff --git a/lite/kernels/arm/sequence_pool_compute.cc b/lite/kernels/arm/sequence_pool_compute.cc
index 8fcbb8cffe72935e4df503c3c1748ddb68247fb7..93072fe499eed296d6e31d87ee9b74494de07aa1 100644
--- a/lite/kernels/arm/sequence_pool_compute.cc
+++ b/lite/kernels/arm/sequence_pool_compute.cc
@@ -59,6 +59,7 @@ void SequencePoolCompute::Run() {
   for (int i = 0; i <= batch_size; i++) {
     offset_new[i] = i;
   }
+  (output->mutable_lod())->clear();
   (output->mutable_lod())->push_back(offset_new);
 }
 
diff --git a/lite/kernels/fpga/CMakeLists.txt b/lite/kernels/fpga/CMakeLists.txt
index e71e5255ca6daa0c86c7f1b1c3d9174df66cac25..90cd2bfd3653a876f6568ecc351464b9d37e3af3 100755
--- a/lite/kernels/fpga/CMakeLists.txt
+++ b/lite/kernels/fpga/CMakeLists.txt
@@ -14,7 +14,6 @@ add_kernel(conv_compute_fpga FPGA basic SRCS conv_compute.cc DEPS ${fpga_deps})
 # add_kernel(density_prior_box_compute_fpga FPGA basic SRCS density_prior_box_compute.cc DEPS ${fpga_deps})
 add_kernel(dropout_compute_fpga FPGA basic SRCS dropout_compute.cc DEPS ${fpga_deps})
 add_kernel(elementwise_compute_fpga FPGA basic SRCS elementwise_compute.cc DEPS ${fpga_deps})
-# add_kernel(feed_compute_fpga FPGA basic SRCS fc_compute.cc DEPS ${fpga_deps})
 
 add_kernel(fc_compute_fpga FPGA basic SRCS fc_compute.cc DEPS ${fpga_deps})
 add_kernel(gru_compute_fpga FPGA extra SRCS gru_compute.cc DEPS ${fpga_deps})
diff --git a/lite/kernels/fpga/fetch_compute.cc b/lite/kernels/fpga/fetch_compute.cc
index 2d296f4d4a89b1fd86e5b2330d3caf44fbad0903..71ec37a64d94bcbef00d7e3c2a187bdb28c47935 100755
--- a/lite/kernels/fpga/fetch_compute.cc
+++ b/lite/kernels/fpga/fetch_compute.cc
@@ -82,6 +82,6 @@ REGISTER_LITE_KERNEL(fetch,
                      kNHWC,
                      paddle::lite::kernels::fpga::FetchCompute,
                      host_host)
-    .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost))})
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kHost))})
     .Finalize();
diff --git a/lite/kernels/host/one_hot_compute.cc b/lite/kernels/host/one_hot_compute.cc
index e0af6f5173f367bb9b2e06de10499ee36806379c..e1bf4c103bcb59c59617d8c5d1ce10ae8780e403 100755
--- a/lite/kernels/host/one_hot_compute.cc
+++ b/lite/kernels/host/one_hot_compute.cc
@@ -16,7 +16,7 @@
 #include <utility>
 #include <vector>
 
-#include "lite/backends/fpga/KD/debugger.hpp"
+// #include "lite/backends/fpga/KD/debugger.hpp"
 #include "lite/kernels/host/one_hot_compute.h"
 #include "lite/utils/paddle_enforce.h"
 
diff --git a/lite/operators/one_hot_op.cc b/lite/operators/one_hot_op.cc
index 023cdc23aeb8329736b7438af2c52cbfa899c75c..ebab9e20679038546611b8dc3221f4ecba1bbe21 100644
--- a/lite/operators/one_hot_op.cc
+++ b/lite/operators/one_hot_op.cc
@@ -15,7 +15,7 @@
 #include "lite/operators/one_hot_op.h"
 #include "lite/core/op_registry.h"
 
-#include "lite/backends/fpga/KD/debugger.hpp"
+// #include "lite/backends/fpga/KD/debugger.hpp"
 
 namespace paddle {
 namespace lite {