apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-11-03 02:53:26 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

Merge branch 'develop' of https://github.com/paddlepaddle/fastdeploy into set_stream_infer-shareExData

Browse Source
This commit is contained in:
wwbitejotunn
2023-02-13 03:14:18 +00:00
parent abfa9fd850 e63f5f369e
commit 289d353d99
50 changed files with 2809 additions and 395 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
fastdeploy/runtime/backends/lite/configure_hardware.cc
View File

@@ -51,8 +51,9 @@ void LiteBackend::ConfigureCpu(const LiteBackendOption& option) {
void LiteBackend::ConfigureKunlunXin(const LiteBackendOption& option) {
std::vector<paddle::lite_api::Place> valid_places;
valid_places.push_back(
paddle::lite_api::Place{TARGET(kXPU), PRECISION(kInt8)});
// TODO(yeliang): Placing kInt8 first may cause accuracy issues of some model
// valid_places.push_back(
// paddle::lite_api::Place{TARGET(kXPU), PRECISION(kInt8)});
if (option.enable_fp16) {
valid_places.push_back(
paddle::lite_api::Place{TARGET(kXPU), PRECISION(kFP16)});

2
fastdeploy/runtime/backends/lite/lite_backend.h
View File

@@ -32,7 +32,7 @@ class LiteBackend : public BaseBackend {
LiteBackend() {}
virtual ~LiteBackend() = default;
bool Init(const RuntimeOption& option);
bool Init(const RuntimeOption& option) override;
bool Infer(std::vector<FDTensor>& inputs,
std::vector<FDTensor>* outputs,

10
fastdeploy/runtime/backends/paddle/option.h
View File

@@ -75,6 +75,16 @@ struct PaddleBackendOption {
delete_pass_names.push_back(pass_name);
}
void SetIpuConfig(bool enable_fp16, int replica_num,
float available_memory_proportion,
bool enable_half_partial) {
ipu_option.ipu_enable_fp16 = enable_fp16;
ipu_option.ipu_replica_num = replica_num;
ipu_option.ipu_available_memory_proportion =
available_memory_proportion;
ipu_option.ipu_enable_half_partial = enable_half_partial;
}
// The belowing parameters may be removed, please do not
// read or write them directly
TrtBackendOption trt_option;

3
fastdeploy/runtime/backends/paddle/option_pybind.cc
View File

@@ -47,7 +47,8 @@ void BindPaddleOption(pybind11::module& m) {
.def_readwrite("gpu_mem_init_size",
&PaddleBackendOption::gpu_mem_init_size)
.def("disable_trt_ops", &PaddleBackendOption::DisableTrtOps)
.def("delete_pass", &PaddleBackendOption::DeletePass);
.def("delete_pass", &PaddleBackendOption::DeletePass)
.def("set_ipu_config", &PaddleBackendOption::SetIpuConfig);
}
} // namespace fastdeploy

5
fastdeploy/runtime/backends/paddle/paddle_backend.cc
View File

@@ -45,7 +45,10 @@ void PaddleBackend::BuildOption(const PaddleBackendOption& option) {
"file will save to the directory where paddle model saved."
<< std::endl;
use_static = true;
config_.SetOptimCacheDir(option.trt_option.serialize_file);
std::string opt_cache_dir =
GetDirFromPath(option.trt_option.serialize_file);
config_.SetOptimCacheDir(opt_cache_dir);
}
config_.EnableTensorRtEngine(option.trt_option.max_workspace_size,
option.trt_option.max_batch_size, 20,