diff --git a/mace/python/tools/converter.py b/mace/python/tools/converter.py
index 241972162493ca70936b466904f3336c849ee7c9..a38bd8cc6bc109333c3c56c12e9ba4d4eca69d13 100644
--- a/mace/python/tools/converter.py
+++ b/mace/python/tools/converter.py
@@ -39,11 +39,6 @@ FLAGS = None
 device_type_map = {'cpu': mace_pb2.CPU,
                    'gpu': mace_pb2.GPU,
                    'dsp': mace_pb2.HEXAGON}
-device_data_type_map = {
-    mace_pb2.CPU: mace_pb2.DT_FLOAT,
-    mace_pb2.GPU: mace_pb2.DT_HALF,
-    mace_pb2.HEXAGON: mace_pb2.DT_UINT8
-}
 
 
 def file_checksum(fname):
@@ -128,6 +123,17 @@ def main(unused_args):
                                                        FLAGS.weight_file)
 
         output_graph_def = converter.run()
+
+        if FLAGS.gpu_data_type == 'half':
+            gpu_data_type = mace_pb2.DT_HALF
+        else:
+            gpu_data_type = mace_pb2.DT_FLOAT
+        device_data_type_map = {
+            mace_pb2.CPU: mace_pb2.DT_FLOAT,
+            mace_pb2.GPU: gpu_data_type,
+            mace_pb2.HEXAGON: mace_pb2.DT_UINT8
+        }
+
         print("Transform model to one that can better run on device")
         if not FLAGS.runtime:
             cpu_graph_def = copy.deepcopy(output_graph_def)
@@ -177,7 +183,8 @@ def main(unused_args):
         source_converter_lib.convert_to_source(
             output_graph_def, model_checksum, weight_checksum, FLAGS.template,
             FLAGS.obfuscate, FLAGS.model_tag, FLAGS.output, FLAGS.runtime,
-            FLAGS.embed_model_data, FLAGS.winograd)
+            FLAGS.embed_model_data, FLAGS.winograd,
+            FLAGS.gpu_data_type)
     else:
         with open(FLAGS.output, "wb") as f:
             f.write(output_graph_def.SerializeToString())
@@ -266,6 +273,8 @@ def parse_args():
         type=str2bool,
         default=True,
         help="embed model data.")
+    parser.add_argument(
+        "--gpu_data_type", type=str, default="half", help="half/float")
     return parser.parse_known_args()
 
 
diff --git a/mace/python/tools/source_converter_lib.py b/mace/python/tools/source_converter_lib.py
index 5b43e61b07f89e95b849fee56aea7bc3f83381af..2adcd383f515b7aae88ca81ca376e1e865855e4d 100644
--- a/mace/python/tools/source_converter_lib.py
+++ b/mace/python/tools/source_converter_lib.py
@@ -109,11 +109,11 @@ def rename_tensor(net_def):
 
 
 class TensorInfo:
-    def __init__(self, id, t, runtime):
+    def __init__(self, id, t, runtime, gpu_data_type):
         self.id = id
         self.data_type = mace_pb2.DataType.Name(t.data_type)
         if t.data_type == mace_pb2.DT_FLOAT:
-            if runtime == 'gpu':
+            if runtime == 'gpu' and gpu_data_type == 'half':
                 self.data_type = mace_pb2.DT_HALF
                 self.data = bytearray(
                     np.array(t.float_data).astype(np.float16).tobytes())
@@ -137,7 +137,7 @@ def stringfy(value):
 
 def convert_to_source(net_def, model_checksum, weight_checksum, template_dir,
                       obfuscate, model_tag, output, runtime, embed_model_data,
-                      winograd_conv):
+                      winograd_conv, gpu_data_type):
     if obfuscate:
         obfuscate_name(net_def)
     else:
@@ -157,7 +157,7 @@ def convert_to_source(net_def, model_checksum, weight_checksum, template_dir,
     offset = 0
     counter = 0
     for t in net_def.tensors:
-        tensor_info = TensorInfo(counter, t, runtime)
+        tensor_info = TensorInfo(counter, t, runtime, gpu_data_type)
         # align
         if tensor_info.data_type != 'DT_UINT8' and offset % 4 != 0:
             padding = 4 - offset % 4
@@ -208,7 +208,7 @@ def convert_to_source(net_def, model_checksum, weight_checksum, template_dir,
     build_time = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')
     template_name = 'model.jinja2'
     tensors = [
-        TensorInfo(i, net_def.tensors[i], runtime)
+        TensorInfo(i, net_def.tensors[i], runtime, gpu_data_type)
         for i in range(len(net_def.tensors))
     ]
     checksum = model_checksum
diff --git a/tools/mace_tools.py b/tools/mace_tools.py
index 2af843a63f677e0e68c6ca3845ac872f5ece1862..af0597e132256fa9389a8e9f50c31ffb49f19ccd 100644
--- a/tools/mace_tools.py
+++ b/tools/mace_tools.py
@@ -523,6 +523,11 @@ def parse_args():
         type=str,
         default="cpu",
         help="validation runtime.")
+    parser.add_argument(
+        "--gpu_data_type",
+        type=str,
+        default="half",
+        help="[half | float].")
     return parser.parse_known_args()
 
 
@@ -778,7 +783,8 @@ def main(unused_args):
                 model_config["dsp_mode"],
                 embed_model_data,
                 model_config["fast_conv"],
-                model_config["obfuscate"])
+                model_config["obfuscate"],
+                FLAGS.gpu_data_type)
 
     for target_abi in configs["target_abis"]:
         for target_soc in target_socs:
diff --git a/tools/sh_commands.py b/tools/sh_commands.py
index da93e0ec7f07912254da35d016bab2bdb3110eef..f8463cc92c77d4f07b0c264f2a71413ec4e584cd 100644
--- a/tools/sh_commands.py
+++ b/tools/sh_commands.py
@@ -471,7 +471,8 @@ def gen_model_code(model_codegen_dir,
                    dsp_mode,
                    embed_model_data,
                    fast_conv,
-                   obfuscate):
+                   obfuscate,
+                   gpu_data_type):
     print("* Genearte model code")
     bazel_build_common("//mace/python/tools:converter")
     if os.path.exists(model_codegen_dir):
@@ -497,6 +498,7 @@ def gen_model_code(model_codegen_dir,
                   "--embed_model_data=%s" % embed_model_data,
                   "--winograd=%s" % fast_conv,
                   "--obfuscate=%s" % obfuscate,
+                  "--gpu_data_type=%s" % gpu_data_type,
                   _out=process_output,
                   _bg=True,
                   _err_to_out=True)