1 /* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 #include "tensorflow_lite_support/cc/text/tokenizers/tokenizer_utils.h"
17
18 #include "absl/status/status.h"
19 #include "tensorflow_lite_support/cc/common.h"
20 #include "tensorflow_lite_support/cc/port/status_macros.h"
21 #include "tensorflow_lite_support/cc/text/tokenizers/bert_tokenizer.h"
22 #include "tensorflow_lite_support/cc/text/tokenizers/regex_tokenizer.h"
23 #include "tensorflow_lite_support/cc/text/tokenizers/sentencepiece_tokenizer.h"
24 #include "tensorflow_lite_support/metadata/metadata_schema_generated.h"
25
26 namespace tflite {
27 namespace support {
28 namespace text {
29 namespace tokenizer {
30
31
32 using ::tflite::ProcessUnit;
33 using ::tflite::SentencePieceTokenizerOptions;
34 using ::tflite::support::CreateStatusWithPayload;
35 using ::tflite::support::StatusOr;
36 using ::tflite::support::TfLiteSupportStatus;
37
38 namespace {
39
CheckAndLoadFirstAssociatedFile(const flatbuffers::Vector<flatbuffers::Offset<tflite::AssociatedFile>> * associated_files,const tflite::metadata::ModelMetadataExtractor * metadata_extractor)40 StatusOr<absl::string_view> CheckAndLoadFirstAssociatedFile(
41 const flatbuffers::Vector<flatbuffers::Offset<tflite::AssociatedFile>>*
42 associated_files,
43 const tflite::metadata::ModelMetadataExtractor* metadata_extractor) {
44 if (associated_files == nullptr || associated_files->size() < 1 ||
45 associated_files->Get(0)->name() == nullptr) {
46 return CreateStatusWithPayload(
47 absl::StatusCode::kInvalidArgument,
48 "Invalid vocab_file from input process unit.",
49 TfLiteSupportStatus::kMetadataInvalidTokenizerError);
50 }
51 ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
52 metadata_extractor->GetAssociatedFile(
53 associated_files->Get(0)->name()->str()));
54 return vocab_buffer;
55 }
56 } // namespace
57
CreateTokenizerFromProcessUnit(const tflite::ProcessUnit * tokenizer_process_unit,const tflite::metadata::ModelMetadataExtractor * metadata_extractor)58 StatusOr<std::unique_ptr<Tokenizer>> CreateTokenizerFromProcessUnit(
59 const tflite::ProcessUnit* tokenizer_process_unit,
60 const tflite::metadata::ModelMetadataExtractor* metadata_extractor) {
61 if (metadata_extractor == nullptr || tokenizer_process_unit == nullptr) {
62 return CreateStatusWithPayload(
63 absl::StatusCode::kInvalidArgument,
64 "No metadata or input process unit found.",
65 TfLiteSupportStatus::kMetadataInvalidTokenizerError);
66 }
67 switch (tokenizer_process_unit->options_type()) {
68 case ProcessUnitOptions_BertTokenizerOptions: {
69 const tflite::BertTokenizerOptions* options =
70 tokenizer_process_unit->options_as<tflite::BertTokenizerOptions>();
71 ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
72 CheckAndLoadFirstAssociatedFile(options->vocab_file(),
73 metadata_extractor));
74 return absl::make_unique<BertTokenizer>(vocab_buffer.data(),
75 vocab_buffer.size());
76 }
77 case ProcessUnitOptions_SentencePieceTokenizerOptions: {
78 const tflite::SentencePieceTokenizerOptions* options =
79 tokenizer_process_unit->options_as<SentencePieceTokenizerOptions>();
80 ASSIGN_OR_RETURN(absl::string_view model_buffer,
81 CheckAndLoadFirstAssociatedFile(
82 options->sentencePiece_model(), metadata_extractor));
83 // TODO(b/160647204): Extract sentence piece model vocabulary
84 return absl::make_unique<SentencePieceTokenizer>(model_buffer.data(),
85 model_buffer.size());
86 }
87 case ProcessUnitOptions_RegexTokenizerOptions: {
88 const tflite::RegexTokenizerOptions* options =
89 tokenizer_process_unit->options_as<RegexTokenizerOptions>();
90 ASSIGN_OR_RETURN(absl::string_view vocab_buffer,
91 CheckAndLoadFirstAssociatedFile(options->vocab_file(),
92 metadata_extractor));
93 if (options->delim_regex_pattern() == nullptr) {
94 return CreateStatusWithPayload(
95 absl::StatusCode::kInvalidArgument,
96 "Invalid delim_regex_pattern from input process unit.",
97 TfLiteSupportStatus::kMetadataInvalidTokenizerError);
98 }
99
100 std::unique_ptr<RegexTokenizer> regex_tokenizer =
101 absl::make_unique<RegexTokenizer>(
102 options->delim_regex_pattern()->str(), vocab_buffer.data(),
103 vocab_buffer.size());
104
105 int unknown_token_id = 0;
106 if (!regex_tokenizer->GetUnknownToken(&unknown_token_id)) {
107 return CreateStatusWithPayload(
108 absl::StatusCode::kInvalidArgument,
109 "RegexTokenizer doesn't have <UNKNOWN> token.",
110 TfLiteSupportStatus::kMetadataInvalidTokenizerError);
111 }
112
113 int pad_token_id = 0;
114 if (!regex_tokenizer->GetPadToken(&pad_token_id)) {
115 return CreateStatusWithPayload(
116 absl::StatusCode::kInvalidArgument,
117 "RegexTokenizer doesn't have <PAD> token.",
118 TfLiteSupportStatus::kMetadataInvalidTokenizerError);
119 }
120
121 return regex_tokenizer;
122 }
123 default:
124 return CreateStatusWithPayload(
125 absl::StatusCode::kNotFound,
126 absl::StrCat("Incorrect options_type:",
127 tokenizer_process_unit->options_type()),
128 TfLiteSupportStatus::kMetadataInvalidTokenizerError);
129 }
130 }
131
132 } // namespace tokenizer
133 } // namespace text
134 } // namespace support
135 } // namespace tflite
136
137