oga5 commited on
Commit
21e5a6f
·
1 Parent(s): 8856b35

update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -40
README.md CHANGED
@@ -66,44 +66,7 @@ print(enc_path, tte_path, tok_path)
66
  Required libraries:
67
  - `onnxruntime`
68
  - `onnxruntime-extensions` (build with `-DOCOS_ENABLE_SPM_TOKENIZER=ON`)
69
- - `pg_onnx` (PostgreSQL extension)
70
-
71
- Note: To enable SentencePiece tokenizer support, ensure `libortextensions.so` is registered via `RegisterCustomOpsLibrary` in `onnxruntime-server/src/onnx/session.cpp`.
72
-
73
- ``` C++
74
- Orts::onnx::session::session(session_key key, const std::string &path, const json &option)
75
- : session(std::move(key), option) {
76
- #ifdef _WIN32
77
- int size_needed = MultiByteToWideChar(CP_ACP, 0, path.c_str(), -1, NULL, 0);
78
- std::wstring wstr(size_needed, 0);
79
- MultiByteToWideChar(CP_ACP, 0, path.c_str(), -1, &wstr[0], size_needed);
80
-
81
- auto model_path = wstr.c_str();
82
- #else
83
- auto model_path = path.c_str();
84
- #endif
85
-
86
- void* handle = nullptr;
87
- const char* ext_path = "libortextensions.so";
88
- OrtStatus* status = Ort::GetApi().RegisterCustomOpsLibrary(session_options, ext_path, &handle);
89
-
90
- ort_session = new Ort::Session(env, model_path, session_options);
91
- init();
92
- }
93
-
94
- Orts::onnx::session::session(
95
- session_key key, const char *model_data, const size_t model_data_length, const json &option
96
- )
97
- : session(std::move(key), option) {
98
-
99
- void* handle = nullptr;
100
- const char* ext_path = "libortextensions.so";
101
- OrtStatus* status = Ort::GetApi().RegisterCustomOpsLibrary(session_options, ext_path, &handle);
102
-
103
- ort_session = new Ort::Session(env, model_data, model_data_length, session_options);
104
- init();
105
- }
106
- ```
107
 
108
  -- After building pg_onnx, install the pg_onnx extension
109
  ``` SQL
@@ -115,8 +78,8 @@ create extension pg_onnx
115
  ```sql
116
 
117
  -- Register models
118
- select pg_onnx_import_model('e5-tok', 'v1', pg_read_binary_file('/PATH/tokenizer.onnx')::bytea, jsonb('{}'), 'e5 tokenizer');
119
- select pg_onnx_import_model('e5-embedding', 'v1', pg_read_binary_file('/PATH/text_to_embedding.onnx')::bytea, jsonb('{}'), 'e5 text to embedding');
120
 
121
  -- Create functions
122
  create or replace function e5_tok(input_text text)
 
66
  Required libraries:
67
  - `onnxruntime`
68
  - `onnxruntime-extensions` (build with `-DOCOS_ENABLE_SPM_TOKENIZER=ON`)
69
+ - `pg_onnx` (v1.23.1b or later)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
70
 
71
  -- After building pg_onnx, install the pg_onnx extension
72
  ``` SQL
 
78
  ```sql
79
 
80
  -- Register models
81
+ select pg_onnx_import_model('e5-tok', 'v1', pg_read_binary_file('/PATH/tokenizer.onnx')::bytea, '{"ortextensions_path": "libortextensions.so"}'::jsonb, 'e5 tokenizer');
82
+ select pg_onnx_import_model('e5-embedding', 'v1', pg_read_binary_file('/PATH/text_to_embedding.onnx')::bytea, '{"ortextensions_path": "libortextensions.so"}'::jsonb, 'e5 text to embedding');
83
 
84
  -- Create functions
85
  create or replace function e5_tok(input_text text)