Add delete pass, disable_trt_ops and enable_fixed_size_opt for serving

zyl75391 · Mar 9, 2023 · f88b06a · f88b06a
1 parent ad14d0e
commit f88b06a
Show file tree

Hide file tree

Showing 3 changed files with 37 additions and 2 deletions.
diff --git a/serving/src/fastdeploy_backend_utils.cc b/serving/src/fastdeploy_backend_utils.cc
@@ -144,6 +144,18 @@ TRITONSERVER_Error* FDParseShape(triton::common::TritonJson::Value& io,
   return nullptr;  // success
 }
 
+int SplitStringByDelimiter(const std::string& s, char delimiter,
+                           std::vector<std::string>* results) {
+  std::string token;
+  std::istringstream token_stream(s);
+  int number_of_tokens = 0;
+  while (std::getline(token_stream, token, delimiter)) {
+    results->push_back(token);
+    number_of_tokens += 1;
+  }
+  return number_of_tokens;
+}
+
 }  // namespace fastdeploy_runtime
 }  // namespace backend
 }  // namespace triton
diff --git a/serving/src/fastdeploy_backend_utils.h b/serving/src/fastdeploy_backend_utils.h
@@ -73,6 +73,10 @@ TRITONSERVER_Error* FDParseShape(triton::common::TritonJson::Value& io,
                                  const std::string& name,
                                  std::vector<int32_t>* shape);
 
+int SplitStringByDelimiter(const std::string& s,
+                           char delimiter,
+                           std::vector<std::string>* results);
+
 }  // namespace fastdeploy_runtime
 }  // namespace backend
 }  // namespace triton
diff --git a/serving/src/fastdeploy_runtime.cc b/serving/src/fastdeploy_runtime.cc
@@ -233,7 +233,8 @@ ModelState::ModelState(TRITONBACKEND_Model* triton_model)
               bool use_paddle_log;
               THROW_IF_BACKEND_MODEL_ERROR(
                   ParseBoolValue(value_string, &use_paddle_log));
-              runtime_options_->paddle_infer_option.enable_log_info = use_paddle_log;
+              runtime_options_->paddle_infer_option.enable_log_info =
+                  use_paddle_log;
             } else if (param_key == "num_streams") {
               int num_streams;
               THROW_IF_BACKEND_MODEL_ERROR(
@@ -320,12 +321,30 @@ ModelState::ModelState(TRITONBACKEND_Model* triton_model)
                 bool use_paddle_log;
                 THROW_IF_BACKEND_MODEL_ERROR(
                     ParseBoolValue(value_string, &use_paddle_log));
-                runtime_options_->paddle_infer_option.enable_log_info = use_paddle_log;
+                runtime_options_->paddle_infer_option.enable_log_info =
+                    use_paddle_log;
               } else if (param_key == "is_clone") {
                 THROW_IF_BACKEND_MODEL_ERROR(
                     ParseBoolValue(value_string, &is_clone_));
               } else if (param_key == "encryption_key") {
                 runtime_options_->SetEncryptionKey(value_string);
+              } else if (param_key == "disable_trt_ops") {
+                std::vector<std::string> disable_trt_ops;
+                SplitStringByDelimiter(value_string, ' ', &disable_trt_ops);
+                runtime_options_->paddle_infer_option.DisablePaddleTrtOPs(
+                    disable_trt_ops);
+              } else if (param_key == "delete_passes") {
+                std::vector<std::string> delete_passes;
+                SplitStringByDelimiter(value_string, ' ', &disable_trt_ops);
+                for (auto&& pass : delete_passes) {
+                  runtime_options_->paddle_infer_option.DeletePass(pass);
+                }
+              } else if (param_key == "enable_fixed_size_opt") {
+                bool enable_fixed_size_opt = false;
+                THROW_IF_BACKEND_MODEL_ERROR(
+                    ParseBoolValue(value_string, &enable_fixed_size_opt));
+                runtime_options_->paddle_infer_option.enable_fixed_size_opt =
+                    enable_fixed_size_opt;
               }
             }
           }