From 8dab6d184a034ca86eafa644ea0d2100aadfe27d Mon Sep 17 00:00:00 2001
From: jmwang66 <wangjiaming.wjm@alibaba-inc.com>
Date: 星期二, 09 五月 2023 10:58:33 +0800
Subject: [PATCH] Merge pull request #473 from alibaba-damo-academy/dev_smohan

---
 funasr/runtime/onnxruntime/src/ct-transformer.cpp |   68 +++++++++++++++++----------------
 1 files changed, 35 insertions(+), 33 deletions(-)

diff --git a/funasr/runtime/onnxruntime/src/ct-transformer.cpp b/funasr/runtime/onnxruntime/src/ct-transformer.cpp
index 5698703..38a5a70 100644
--- a/funasr/runtime/onnxruntime/src/ct-transformer.cpp
+++ b/funasr/runtime/onnxruntime/src/ct-transformer.cpp
@@ -1,35 +1,36 @@
+/**
+ * Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
+ * MIT License  (https://opensource.org/licenses/MIT)
+*/
+
 #include "precomp.h"
 
-CTTransformer::CTTransformer(const char* sz_model_dir, int thread_num)
+namespace funasr {
+CTTransformer::CTTransformer()
 :env_(ORT_LOGGING_LEVEL_ERROR, ""),session_options{}
 {
+}
+
+void CTTransformer::InitPunc(const std::string &punc_model, const std::string &punc_config, int thread_num){
     session_options.SetIntraOpNumThreads(thread_num);
     session_options.SetGraphOptimizationLevel(ORT_ENABLE_ALL);
     session_options.DisableCpuMemArena();
 
-	string strModelPath = pathAppend(sz_model_dir, PUNC_MODEL_FILE);
-	string strYamlPath = pathAppend(sz_model_dir, PUNC_YAML_FILE);
-
     try{
-#ifdef _WIN32
-	std::wstring detPath = strToWstr(strModelPath);
-    m_session = std::make_unique<Ort::Session>(env_, detPath.c_str(), session_options);
-#else
-    m_session = std::make_unique<Ort::Session>(env_, strModelPath.c_str(), session_options);
-#endif
+        m_session = std::make_unique<Ort::Session>(env_, punc_model.c_str(), session_options);
     }
-    catch(exception e)
-    {
-        printf(e.what());
+    catch (std::exception const &e) {
+        LOG(ERROR) << "Error when load punc onnx model: " << e.what();
+        exit(0);
     }
-    // read inputnames outputnamess
+    // read inputnames outputnames
     string strName;
-    getInputName(m_session.get(), strName);
+    GetInputName(m_session.get(), strName);
     m_strInputNames.push_back(strName.c_str());
-    getInputName(m_session.get(), strName, 1);
+    GetInputName(m_session.get(), strName, 1);
     m_strInputNames.push_back(strName);
     
-    getOutputName(m_session.get(), strName);
+    GetOutputName(m_session.get(), strName);
     m_strOutputNames.push_back(strName);
 
     for (auto& item : m_strInputNames)
@@ -37,7 +38,7 @@
     for (auto& item : m_strOutputNames)
         m_szOutputNames.push_back(item.c_str());
 
-	m_tokenizer.OpenYaml(strYamlPath.c_str());
+	m_tokenizer.OpenYaml(punc_config.c_str());
 }
 
 CTTransformer::~CTTransformer()
@@ -54,7 +55,7 @@
     int nTotalBatch = ceil((float)InputData.size() / TOKEN_LEN);
     int nCurBatch = -1;
     int nSentEnd = -1, nLastCommaIndex = -1;
-    vector<int64_t> RemainIDs; // 
+    vector<int32_t> RemainIDs; // 
     vector<string> RemainStr; //
     vector<int> NewPunctuation; //
     vector<string> NewString; //
@@ -64,7 +65,7 @@
     for (size_t i = 0; i < InputData.size(); i += TOKEN_LEN)
     {
         nDiff = (i + TOKEN_LEN) < InputData.size() ? (0) : (i + TOKEN_LEN - InputData.size());
-        vector<int64_t> InputIDs(InputData.begin() + i, InputData.begin() + i + TOKEN_LEN - nDiff);
+        vector<int32_t> InputIDs(InputData.begin() + i, InputData.begin() + i + TOKEN_LEN - nDiff);
         vector<string> InputStr(strOut.begin() + i, strOut.begin() + i + TOKEN_LEN - nDiff);
         InputIDs.insert(InputIDs.begin(), RemainIDs.begin(), RemainIDs.end()); // RemainIDs+InputIDs;
         InputStr.insert(InputStr.begin(), RemainStr.begin(), RemainStr.end()); // RemainStr+InputStr;
@@ -77,12 +78,12 @@
             nLastCommaIndex = -1;
             for (int nIndex = Punction.size() - 2; nIndex > 0; nIndex--)
             {
-                if (m_tokenizer.ID2Punc(Punction[nIndex]) == m_tokenizer.ID2Punc(PERIOD_INDEX) || m_tokenizer.ID2Punc(Punction[nIndex]) == m_tokenizer.ID2Punc(QUESTION_INDEX))
+                if (m_tokenizer.Id2Punc(Punction[nIndex]) == m_tokenizer.Id2Punc(PERIOD_INDEX) || m_tokenizer.Id2Punc(Punction[nIndex]) == m_tokenizer.Id2Punc(QUESTION_INDEX))
                 {
                     nSentEnd = nIndex;
                     break;
                 }
-                if (nLastCommaIndex < 0 && m_tokenizer.ID2Punc(Punction[nIndex]) == m_tokenizer.ID2Punc(COMMA_INDEX))
+                if (nLastCommaIndex < 0 && m_tokenizer.Id2Punc(Punction[nIndex]) == m_tokenizer.Id2Punc(COMMA_INDEX))
                 {
                     nLastCommaIndex = nIndex;
                 }
@@ -110,7 +111,7 @@
 
             if (Punction[i] != NOTPUNC_INDEX) // 锟铰伙拷锟斤拷
             {
-                WordWithPunc.push_back(m_tokenizer.ID2Punc(Punction[i]));
+                WordWithPunc.push_back(m_tokenizer.Id2Punc(Punction[i]));
             }
         }
 
@@ -120,17 +121,17 @@
         // last mini sentence
         if(nCurBatch == nTotalBatch - 1)
         {
-            if (NewString[NewString.size() - 1] == m_tokenizer.ID2Punc(COMMA_INDEX) || NewString[NewString.size() - 1] == m_tokenizer.ID2Punc(DUN_INDEX))
+            if (NewString[NewString.size() - 1] == m_tokenizer.Id2Punc(COMMA_INDEX) || NewString[NewString.size() - 1] == m_tokenizer.Id2Punc(DUN_INDEX))
             {
                 NewSentenceOut.assign(NewString.begin(), NewString.end() - 1);
-                NewSentenceOut.push_back(m_tokenizer.ID2Punc(PERIOD_INDEX));
+                NewSentenceOut.push_back(m_tokenizer.Id2Punc(PERIOD_INDEX));
                 NewPuncOut.assign(NewPunctuation.begin(), NewPunctuation.end() - 1);
                 NewPuncOut.push_back(PERIOD_INDEX);
             }
-            else if (NewString[NewString.size() - 1] == m_tokenizer.ID2Punc(PERIOD_INDEX) && NewString[NewString.size() - 1] == m_tokenizer.ID2Punc(QUESTION_INDEX))
+            else if (NewString[NewString.size() - 1] == m_tokenizer.Id2Punc(PERIOD_INDEX) && NewString[NewString.size() - 1] == m_tokenizer.Id2Punc(QUESTION_INDEX))
             {
                 NewSentenceOut = NewString;
-                NewSentenceOut.push_back(m_tokenizer.ID2Punc(PERIOD_INDEX));
+                NewSentenceOut.push_back(m_tokenizer.Id2Punc(PERIOD_INDEX));
                 NewPuncOut = NewPunctuation;
                 NewPuncOut.push_back(PERIOD_INDEX);
             }
@@ -141,12 +142,13 @@
     return strResult;
 }
 
-vector<int> CTTransformer::Infer(vector<int64_t> input_data)
+vector<int> CTTransformer::Infer(vector<int32_t> input_data)
 {
     Ort::MemoryInfo m_memoryInfo = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault);
     vector<int> punction;
     std::array<int64_t, 2> input_shape_{ 1, (int64_t)input_data.size()};
-    Ort::Value onnx_input = Ort::Value::CreateTensor<int64_t>(m_memoryInfo,
+    Ort::Value onnx_input = Ort::Value::CreateTensor<int32_t>(
+        m_memoryInfo,
         input_data.data(),
         input_data.size(),
         input_shape_.data(),
@@ -173,16 +175,16 @@
 
         for (int i = 0; i < outputCount; i += CANDIDATE_NUM)
         {
-            int index = argmax(floatData + i, floatData + i + CANDIDATE_NUM-1);
+            int index = Argmax(floatData + i, floatData + i + CANDIDATE_NUM-1);
             punction.push_back(index);
         }
     }
     catch (std::exception const &e)
     {
-        printf(e.what());
+        LOG(ERROR) << "Error when run punc onnx forword: " << (e.what());
+        exit(0);
     }
     return punction;
 }
 
-
-
+} // namespace funasr
\ No newline at end of file

--
Gitblit v1.9.1