p60975123
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!13987 [MS][LITE]fix a bug of the fc op in the fp16 subgraph and add models to the entrance guard 

From: @probiotics_53
Reviewed-by: @zhanghaibo5,@zhang_xue_tong
Signed-off-by: @zhanghaibo5
This commit is contained in:
mindspore-ci-bot 2021-03-25 09:01:41 +08:00 committed by Gitee
parent 4f0ad9398e 7aeeed7e4d
commit b808876234
2 changed files with 11 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
mindspore/lite/src/lite_kernel.cc
View File

@ -105,6 +105,9 @@ int LiteKernel::PreProcess() {
for (auto *output : this->out_tensors()) {
MS_ASSERT(output != nullptr);
if (desc_.data_type == kNumberTypeFloat16 && output->data_type() == kNumberTypeFloat32) {
output->set_data_type(kNumberTypeFloat16);
}
if (output->ElementsNum() >= MAX_MALLOC_SIZE / static_cast<int>(sizeof(int64_t))) {
MS_LOG(ERROR) << "The size of output tensor is too big";
return RET_ERROR;

8
mindspore/lite/test/models_tflite_fp16.cfg
View File

@ -149,3 +149,11 @@ scan_hms_detect_pb2tflite.tflite 1.5
ml_location.tflite 0.5
ml_face_openclose_tflite.tflite 0.5
ml_object_detect_pb2tflite.tflite 1.5
# lite-model_on_device_vision_classifier_landmarks_classifier* models' bias are caused by error accumulation and small
# output value
lite-model_on_device_vision_classifier_landmarks_classifier_africa_V1_1.tflite 10
lite-model_on_device_vision_classifier_landmarks_classifier_north_america_V1_1.tflite 18
lite-model_on_device_vision_classifier_landmarks_classifier_asia_V1_1.tflite 25
lite-model_on_device_vision_classifier_landmarks_classifier_oceania_antarctica_V1_1.tflite 10
lite-model_on_device_vision_classifier_landmarks_classifier_europe_V1_1.tflite 32
lite-model_on_device_vision_classifier_landmarks_classifier_south_america_V1_1.tflite 14