Hugging Face's logo

OpenVINO
/
gemma-2-9b-it-int8-ov

OpenVINO
gemma2
Model card Files
xet
Community
openvino-ci commited on
Commit
5c4d529
·
verified ·
1 Parent(s): 2b1b324

Upload folder using huggingface_hub

Browse files
Files changed (9) hide show
  1. README.md +5 -6
  2. config.json +3 -2
  3. generation_config.json +1 -1
  4. openvino_detokenizer.bin +3 -0
  5. openvino_detokenizer.xml +263 -0
  6. openvino_model.bin +2 -2
  7. openvino_model.xml +0 -0
  8. openvino_tokenizer.bin +3 -0
  9. openvino_tokenizer.xml +837 -0
README.md CHANGED
@@ -1,12 +1,11 @@
1
  ---
2
  license: gemma
3
  license_link: https://choosealicense.com/licenses/gemma/
4
- base_model:
5
- - google/gemma-2-9b-it
6
  base_model_relation: quantized
7
  ---
8
  # gemma-2-9b-it-int8-ov
9
- * Model creator: [Google](https://huggingface.co/google)
10
  * Original model: [gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it)
11
 
12
  ## Description
@@ -26,8 +25,8 @@ For more information on quantization, check the [OpenVINO model optimization gui
26
 
27
  The provided OpenVINO™ IR model is compatible with:
28
 
29
- * OpenVINO version 2024.4.0 and higher
30
- * Optimum Intel 1.20.0 and higher
31
 
32
  ## Running Model Inference
33
 
@@ -66,4 +65,4 @@ The original model is distributed under [gemma](https://ai.google.dev/gemma/term
66
 
67
  ## Disclaimer
68
 
69
- Intel is committed to respecting human rights and avoiding causing or contributing to adverse impacts on human rights. See [Intel’s Global Human Rights Principles](https://www.intel.com/content/dam/www/central-libraries/us/en/documents/policy-human-rights.pdf). Intel’s products and software are intended only to be used in applications that do not cause or contribute to adverse impacts on human rights.
 
1
  ---
2
  license: gemma
3
  license_link: https://choosealicense.com/licenses/gemma/
4
+ base_model: google/gemma-2-9b-it
 
5
  base_model_relation: quantized
6
  ---
7
  # gemma-2-9b-it-int8-ov
8
+ * Model creator: [google](https://huggingface.co/google)
9
  * Original model: [gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it)
10
 
11
  ## Description
 
25
 
26
  The provided OpenVINO™ IR model is compatible with:
27
 
28
+ * OpenVINO version 2024.5.0 and higher
29
+ * Optimum Intel 1.21.0 and higher
30
 
31
  ## Running Model Inference
32
 
 
65
 
66
  ## Disclaimer
67
 
68
+ Intel is committed to respecting human rights and avoiding causing or contributing to adverse impacts on human rights. See [Intel’s Global Human Rights Principles](https://www.intel.com/content/dam/www/central-libraries/us/en/documents/policy-human-rights.pdf). Intel’s products and software are intended only to be used in applications that do not cause or contribute to adverse impacts on human rights.
config.json CHANGED
@@ -1,5 +1,6 @@
1
  {
2
- "_name_or_path": "google/gemma-2-9b-it",
 
3
  "architectures": [
4
  "Gemma2ForCausalLM"
5
  ],
@@ -28,7 +29,7 @@
28
  "sliding_window": 4096,
29
  "sliding_window_size": 4096,
30
  "torch_dtype": "bfloat16",
31
- "transformers_version": "4.45.2",
32
  "use_cache": true,
33
  "vocab_size": 256000
34
  }
 
1
  {
2
+ "_attn_implementation_autoset": true,
3
+ "_name_or_path": "OpenVINO/gemma-2-9b-it-int8-ov",
4
  "architectures": [
5
  "Gemma2ForCausalLM"
6
  ],
 
29
  "sliding_window": 4096,
30
  "sliding_window_size": 4096,
31
  "torch_dtype": "bfloat16",
32
+ "transformers_version": "4.46.3",
33
  "use_cache": true,
34
  "vocab_size": 256000
35
  }
generation_config.json CHANGED
@@ -4,5 +4,5 @@
4
  "cache_implementation": "hybrid",
5
  "eos_token_id": 1,
6
  "pad_token_id": 0,
7
- "transformers_version": "4.45.2"
8
  }
 
4
  "cache_implementation": "hybrid",
5
  "eos_token_id": 1,
6
  "pad_token_id": 0,
7
+ "transformers_version": "4.46.3"
8
  }
openvino_detokenizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a8708fde74785835df6bc8f1a119d713e8bb70fd99c6df7c47a3934b52de781
3
+ size 2955914
openvino_detokenizer.xml ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="detokenizer" version="11">
3
+ <layers>
4
+ <layer id="0" name="Parameter_126327" type="Parameter" version="opset1">
5
+ <data shape="?,?" element_type="i64" />
6
+ <output>
7
+ <port id="0" precision="I64" names="Parameter_126327">
8
+ <dim>-1</dim>
9
+ <dim>-1</dim>
10
+ </port>
11
+ </output>
12
+ </layer>
13
+ <layer id="1" name="Convert_126344" type="Convert" version="opset1">
14
+ <data destination_type="i32" />
15
+ <input>
16
+ <port id="0" precision="I64">
17
+ <dim>-1</dim>
18
+ <dim>-1</dim>
19
+ </port>
20
+ </input>
21
+ <output>
22
+ <port id="1" precision="I32">
23
+ <dim>-1</dim>
24
+ <dim>-1</dim>
25
+ </port>
26
+ </output>
27
+ </layer>
28
+ <layer id="2" name="Constant_126294" type="Const" version="opset1">
29
+ <data element_type="u8" shape="2955910" offset="0" size="2955910" />
30
+ <output>
31
+ <port id="0" precision="U8">
32
+ <dim>2955910</dim>
33
+ </port>
34
+ </output>
35
+ </layer>
36
+ <layer id="3" name="StringTensorUnpack_126295" type="StringTensorUnpack" version="extension">
37
+ <data mode="begins_ends" />
38
+ <input>
39
+ <port id="0" precision="U8">
40
+ <dim>2955910</dim>
41
+ </port>
42
+ </input>
43
+ <output>
44
+ <port id="1" precision="I32">
45
+ <dim>-1</dim>
46
+ </port>
47
+ <port id="2" precision="I32">
48
+ <dim>-1</dim>
49
+ </port>
50
+ <port id="3" precision="U8">
51
+ <dim>-1</dim>
52
+ </port>
53
+ </output>
54
+ </layer>
55
+ <layer id="4" name="VocabDecoder_126328" type="VocabDecoder" version="extension">
56
+ <data skip_tokens="0, 1, 2, 3, 106, 107" />
57
+ <input>
58
+ <port id="0" precision="I32">
59
+ <dim>-1</dim>
60
+ <dim>-1</dim>
61
+ </port>
62
+ <port id="1" precision="I32">
63
+ <dim>-1</dim>
64
+ </port>
65
+ <port id="2" precision="I32">
66
+ <dim>-1</dim>
67
+ </port>
68
+ <port id="3" precision="U8">
69
+ <dim>-1</dim>
70
+ </port>
71
+ </input>
72
+ <output>
73
+ <port id="4" precision="I32">
74
+ <dim>-1</dim>
75
+ </port>
76
+ <port id="5" precision="I32">
77
+ <dim>-1</dim>
78
+ </port>
79
+ <port id="6" precision="I32">
80
+ <dim>-1</dim>
81
+ </port>
82
+ <port id="7" precision="I32">
83
+ <dim>-1</dim>
84
+ </port>
85
+ <port id="8" precision="U8">
86
+ <dim>-1</dim>
87
+ </port>
88
+ </output>
89
+ </layer>
90
+ <layer id="5" name="Constant_126330" type="Const" version="opset1">
91
+ <data element_type="u8" shape="3" offset="2955910" size="3" />
92
+ <output>
93
+ <port id="0" precision="U8">
94
+ <dim>3</dim>
95
+ </port>
96
+ </output>
97
+ </layer>
98
+ <layer id="6" name="Constant_126332" type="Const" version="opset1">
99
+ <data element_type="u8" shape="1" offset="2955913" size="1" />
100
+ <output>
101
+ <port id="0" precision="U8">
102
+ <dim>1</dim>
103
+ </port>
104
+ </output>
105
+ </layer>
106
+ <layer id="7" name="RegexNormalization_126333" type="RegexNormalization" version="extension">
107
+ <data global_replace="true" />
108
+ <input>
109
+ <port id="0" precision="I32">
110
+ <dim>-1</dim>
111
+ </port>
112
+ <port id="1" precision="I32">
113
+ <dim>-1</dim>
114
+ </port>
115
+ <port id="2" precision="U8">
116
+ <dim>-1</dim>
117
+ </port>
118
+ <port id="3" precision="U8">
119
+ <dim>3</dim>
120
+ </port>
121
+ <port id="4" precision="U8">
122
+ <dim>1</dim>
123
+ </port>
124
+ </input>
125
+ <output>
126
+ <port id="5" precision="I32">
127
+ <dim>-1</dim>
128
+ </port>
129
+ <port id="6" precision="I32">
130
+ <dim>-1</dim>
131
+ </port>
132
+ <port id="7" precision="U8">
133
+ <dim>-1</dim>
134
+ </port>
135
+ </output>
136
+ </layer>
137
+ <layer id="8" name="ByteFallback_126334" type="ByteFallback" version="extension">
138
+ <input>
139
+ <port id="0" precision="I32">
140
+ <dim>-1</dim>
141
+ </port>
142
+ <port id="1" precision="I32">
143
+ <dim>-1</dim>
144
+ </port>
145
+ <port id="2" precision="U8">
146
+ <dim>-1</dim>
147
+ </port>
148
+ </input>
149
+ <output>
150
+ <port id="3" precision="I32">
151
+ <dim>-1</dim>
152
+ </port>
153
+ <port id="4" precision="I32">
154
+ <dim>-1</dim>
155
+ </port>
156
+ <port id="5" precision="U8">
157
+ <dim>-1</dim>
158
+ </port>
159
+ </output>
160
+ </layer>
161
+ <layer id="9" name="FuzeRagged_126335" type="FuzeRagged" version="extension">
162
+ <input>
163
+ <port id="0" precision="I32">
164
+ <dim>-1</dim>
165
+ </port>
166
+ <port id="1" precision="I32">
167
+ <dim>-1</dim>
168
+ </port>
169
+ <port id="2" precision="I32">
170
+ <dim>-1</dim>
171
+ </port>
172
+ <port id="3" precision="I32">
173
+ <dim>-1</dim>
174
+ </port>
175
+ </input>
176
+ <output>
177
+ <port id="4" precision="I32">
178
+ <dim>-1</dim>
179
+ </port>
180
+ <port id="5" precision="I32">
181
+ <dim>-1</dim>
182
+ </port>
183
+ </output>
184
+ </layer>
185
+ <layer id="10" name="StringTensorPack_126336" type="StringTensorPack" version="extension">
186
+ <data mode="begins_ends" />
187
+ <input>
188
+ <port id="0" precision="I32">
189
+ <dim>-1</dim>
190
+ </port>
191
+ <port id="1" precision="I32">
192
+ <dim>-1</dim>
193
+ </port>
194
+ <port id="2" precision="U8">
195
+ <dim>-1</dim>
196
+ </port>
197
+ </input>
198
+ <output>
199
+ <port id="3" precision="STRING" names="string_output">
200
+ <dim>-1</dim>
201
+ </port>
202
+ </output>
203
+ </layer>
204
+ <layer id="11" name="Result_126337" type="Result" version="opset1">
205
+ <input>
206
+ <port id="0" precision="STRING">
207
+ <dim>-1</dim>
208
+ </port>
209
+ </input>
210
+ </layer>
211
+ </layers>
212
+ <edges>
213
+ <edge from-layer="0" from-port="0" to-layer="1" to-port="0" />
214
+ <edge from-layer="1" from-port="1" to-layer="4" to-port="0" />
215
+ <edge from-layer="2" from-port="0" to-layer="3" to-port="0" />
216
+ <edge from-layer="3" from-port="1" to-layer="4" to-port="1" />
217
+ <edge from-layer="3" from-port="2" to-layer="4" to-port="2" />
218
+ <edge from-layer="3" from-port="3" to-layer="4" to-port="3" />
219
+ <edge from-layer="4" from-port="6" to-layer="7" to-port="0" />
220
+ <edge from-layer="4" from-port="7" to-layer="7" to-port="1" />
221
+ <edge from-layer="4" from-port="8" to-layer="7" to-port="2" />
222
+ <edge from-layer="4" from-port="5" to-layer="9" to-port="1" />
223
+ <edge from-layer="4" from-port="4" to-layer="9" to-port="0" />
224
+ <edge from-layer="5" from-port="0" to-layer="7" to-port="3" />
225
+ <edge from-layer="6" from-port="0" to-layer="7" to-port="4" />
226
+ <edge from-layer="7" from-port="5" to-layer="8" to-port="0" />
227
+ <edge from-layer="7" from-port="7" to-layer="8" to-port="2" />
228
+ <edge from-layer="7" from-port="6" to-layer="8" to-port="1" />
229
+ <edge from-layer="8" from-port="3" to-layer="9" to-port="2" />
230
+ <edge from-layer="8" from-port="4" to-layer="9" to-port="3" />
231
+ <edge from-layer="8" from-port="5" to-layer="10" to-port="2" />
232
+ <edge from-layer="9" from-port="4" to-layer="10" to-port="0" />
233
+ <edge from-layer="9" from-port="5" to-layer="10" to-port="1" />
234
+ <edge from-layer="10" from-port="3" to-layer="11" to-port="0" />
235
+ </edges>
236
+ <rt_info>
237
+ <add_attention_mask value="True" />
238
+ <add_prefix_space />
239
+ <add_special_tokens value="True" />
240
+ <bos_token_id value="2" />
241
+ <chat_template value="{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ raise_exception('System role not supported') }}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '&lt;start_of_turn>' + role + '&#10;' + message['content'] | trim + '&lt;end_of_turn>&#10;' }}{% endfor %}{% if add_generation_prompt %}{{'&lt;start_of_turn>model&#10;'}}{% endif %}" />
242
+ <clean_up_tokenization_spaces />
243
+ <detokenizer_input_type value="i64" />
244
+ <eos_token_id value="1" />
245
+ <handle_special_tokens_with_re />
246
+ <number_of_inputs value="1" />
247
+ <openvino_tokenizers_version value="2024.5.0.0" />
248
+ <openvino_version value="2024.5.0" />
249
+ <original_tokenizer_class value="&lt;class 'transformers.models.gemma.tokenization_gemma_fast.GemmaTokenizerFast'>" />
250
+ <pad_token_id value="0" />
251
+ <sentencepiece_version value="0.2.0" />
252
+ <skip_special_tokens value="True" />
253
+ <streaming_detokenizer value="False" />
254
+ <tiktoken_version value="0.8.0" />
255
+ <tokenizer_output_type value="i64" />
256
+ <tokenizers_version value="0.20.3" />
257
+ <transformers_version value="4.46.3" />
258
+ <use_max_padding value="False" />
259
+ <use_sentencepiece_backend value="False" />
260
+ <utf8_replace_mode />
261
+ <with_detokenizer value="True" />
262
+ </rt_info>
263
+ </net>
openvino_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d93f8692e53fff202da3e865b9d615542d4c53f1402293b7bbc7df33cf9f559a
3
- size 9249839914
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8874c9c799db44db2dcc27d2211b8318491b85b21359c7a0044557f57ca8c6c
3
+ size 9249839910
openvino_model.xml CHANGED
The diff for this file is too large to render. See raw diff
 
openvino_tokenizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f92947ab004d23330300509b3a98e4e7252e7516ed2b67cd41e9d8f4cc92f6b
3
+ size 12245627
openvino_tokenizer.xml ADDED
@@ -0,0 +1,837 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="tokenizer" version="11">
3
+ <layers>
4
+ <layer id="0" name="Parameter_126208" type="Parameter" version="opset1">
5
+ <data shape="?" element_type="string" />
6
+ <output>
7
+ <port id="0" precision="STRING" names="Parameter_126208">
8
+ <dim>-1</dim>
9
+ </port>
10
+ </output>
11
+ </layer>
12
+ <layer id="1" name="Constant_126311" type="Const" version="opset1">
13
+ <data element_type="i32" shape="" offset="0" size="4" />
14
+ <output>
15
+ <port id="0" precision="I32" />
16
+ </output>
17
+ </layer>
18
+ <layer id="2" name="Constant_126312" type="Const" version="opset1">
19
+ <data element_type="i32" shape="" offset="4" size="4" />
20
+ <output>
21
+ <port id="0" precision="I32" />
22
+ </output>
23
+ </layer>
24
+ <layer id="3" name="Constant_126313" type="Const" version="opset1">
25
+ <data element_type="i32" shape="1" offset="8" size="4" />
26
+ <output>
27
+ <port id="0" precision="I32">
28
+ <dim>1</dim>
29
+ </port>
30
+ </output>
31
+ </layer>
32
+ <layer id="4" name="Constant_126214" type="Const" version="opset1">
33
+ <data element_type="i64" shape="" offset="12" size="8" />
34
+ <output>
35
+ <port id="0" precision="I64" />
36
+ </output>
37
+ </layer>
38
+ <layer id="5" name="StringTensorUnpack_126209" type="StringTensorUnpack" version="extension">
39
+ <data mode="begins_ends" />
40
+ <input>
41
+ <port id="0" precision="STRING">
42
+ <dim>-1</dim>
43
+ </port>
44
+ </input>
45
+ <output>
46
+ <port id="1" precision="I32">
47
+ <dim>-1</dim>
48
+ </port>
49
+ <port id="2" precision="I32">
50
+ <dim>-1</dim>
51
+ </port>
52
+ <port id="3" precision="U8">
53
+ <dim>-1</dim>
54
+ </port>
55
+ </output>
56
+ </layer>
57
+ <layer id="6" name="ShapeOf_126210" type="ShapeOf" version="opset3">
58
+ <data output_type="i64" />
59
+ <input>
60
+ <port id="0" precision="I32">
61
+ <dim>-1</dim>
62
+ </port>
63
+ </input>
64
+ <output>
65
+ <port id="1" precision="I64">
66
+ <dim>1</dim>
67
+ </port>
68
+ </output>
69
+ </layer>
70
+ <layer id="7" name="Constant_126211" type="Const" version="opset1">
71
+ <data element_type="i64" shape="" offset="12" size="8" />
72
+ <output>
73
+ <port id="0" precision="I64" />
74
+ </output>
75
+ </layer>
76
+ <layer id="8" name="Constant_126212" type="Const" version="opset1">
77
+ <data element_type="i64" shape="" offset="12" size="8" />
78
+ <output>
79
+ <port id="0" precision="I64" />
80
+ </output>
81
+ </layer>
82
+ <layer id="9" name="Gather_126213" type="Gather" version="opset8">
83
+ <data batch_dims="0" />
84
+ <input>
85
+ <port id="0" precision="I64">
86
+ <dim>1</dim>
87
+ </port>
88
+ <port id="1" precision="I64" />
89
+ <port id="2" precision="I64" />
90
+ </input>
91
+ <output>
92
+ <port id="3" precision="I64" />
93
+ </output>
94
+ </layer>
95
+ <layer id="10" name="Constant_126215" type="Const" version="opset1">
96
+ <data element_type="i64" shape="" offset="20" size="8" />
97
+ <output>
98
+ <port id="0" precision="I64" />
99
+ </output>
100
+ </layer>
101
+ <layer id="11" name="Range_126216" type="Range" version="opset4">
102
+ <data output_type="i32" />
103
+ <input>
104
+ <port id="0" precision="I64" />
105
+ <port id="1" precision="I64" />
106
+ <port id="2" precision="I64" />
107
+ </input>
108
+ <output>
109
+ <port id="3" precision="I32">
110
+ <dim>-1</dim>
111
+ </port>
112
+ </output>
113
+ </layer>
114
+ <layer id="12" name="Constant_126217" type="Const" version="opset1">
115
+ <data element_type="i64" shape="" offset="20" size="8" />
116
+ <output>
117
+ <port id="0" precision="I64" />
118
+ </output>
119
+ </layer>
120
+ <layer id="13" name="Constant_126218" type="Const" version="opset1">
121
+ <data element_type="i64" shape="" offset="20" size="8" />
122
+ <output>
123
+ <port id="0" precision="I64" />
124
+ </output>
125
+ </layer>
126
+ <layer id="14" name="Add_126219" type="Add" version="opset1">
127
+ <data auto_broadcast="numpy" />
128
+ <input>
129
+ <port id="0" precision="I64" />
130
+ <port id="1" precision="I64" />
131
+ </input>
132
+ <output>
133
+ <port id="2" precision="I64" />
134
+ </output>
135
+ </layer>
136
+ <layer id="15" name="Constant_126220" type="Const" version="opset1">
137
+ <data element_type="i64" shape="" offset="20" size="8" />
138
+ <output>
139
+ <port id="0" precision="I64" />
140
+ </output>
141
+ </layer>
142
+ <layer id="16" name="Range_126221" type="Range" version="opset4">
143
+ <data output_type="i32" />
144
+ <input>
145
+ <port id="0" precision="I64" />
146
+ <port id="1" precision="I64" />
147
+ <port id="2" precision="I64" />
148
+ </input>
149
+ <output>
150
+ <port id="3" precision="I32">
151
+ <dim>-1</dim>
152
+ </port>
153
+ </output>
154
+ </layer>
155
+ <layer id="17" name="Constant_126283" type="Const" version="opset1">
156
+ <data element_type="u8" shape="6395" offset="28" size="6395" />
157
+ <output>
158
+ <port id="0" precision="U8">
159
+ <dim>6395</dim>
160
+ </port>
161
+ </output>
162
+ </layer>
163
+ <layer id="18" name="SpecialTokensSplit_126284" type="SpecialTokensSplit" version="extension">
164
+ <input>
165
+ <port id="0" precision="I32">
166
+ <dim>-1</dim>
167
+ </port>
168
+ <port id="1" precision="I32">
169
+ <dim>-1</dim>
170
+ </port>
171
+ <port id="2" precision="I32">
172
+ <dim>-1</dim>
173
+ </port>
174
+ <port id="3" precision="I32">
175
+ <dim>-1</dim>
176
+ </port>
177
+ <port id="4" precision="U8">
178
+ <dim>-1</dim>
179
+ </port>
180
+ <port id="5" precision="U8">
181
+ <dim>6395</dim>
182
+ </port>
183
+ </input>
184
+ <output>
185
+ <port id="6" precision="I32">
186
+ <dim>-1</dim>
187
+ </port>
188
+ <port id="7" precision="I32">
189
+ <dim>-1</dim>
190
+ </port>
191
+ <port id="8" precision="I32">
192
+ <dim>-1</dim>
193
+ </port>
194
+ <port id="9" precision="I32">
195
+ <dim>-1</dim>
196
+ </port>
197
+ <port id="10" precision="U8">
198
+ <dim>-1</dim>
199
+ </port>
200
+ <port id="11" precision="BOOL">
201
+ <dim>-1</dim>
202
+ </port>
203
+ </output>
204
+ </layer>
205
+ <layer id="19" name="Constant_126286" type="Const" version="opset1">
206
+ <data element_type="u8" shape="1" offset="6423" size="1" />
207
+ <output>
208
+ <port id="0" precision="U8">
209
+ <dim>1</dim>
210
+ </port>
211
+ </output>
212
+ </layer>
213
+ <layer id="20" name="Constant_126288" type="Const" version="opset1">
214
+ <data element_type="u8" shape="3" offset="6424" size="3" />
215
+ <output>
216
+ <port id="0" precision="U8">
217
+ <dim>3</dim>
218
+ </port>
219
+ </output>
220
+ </layer>
221
+ <layer id="21" name="RegexNormalization_126289" type="RegexNormalization" version="extension">
222
+ <data global_replace="true" />
223
+ <input>
224
+ <port id="0" precision="I32">
225
+ <dim>-1</dim>
226
+ </port>
227
+ <port id="1" precision="I32">
228
+ <dim>-1</dim>
229
+ </port>
230
+ <port id="2" precision="U8">
231
+ <dim>-1</dim>
232
+ </port>
233
+ <port id="3" precision="BOOL">
234
+ <dim>-1</dim>
235
+ </port>
236
+ <port id="4" precision="U8">
237
+ <dim>1</dim>
238
+ </port>
239
+ <port id="5" precision="U8">
240
+ <dim>3</dim>
241
+ </port>
242
+ </input>
243
+ <output>
244
+ <port id="6" precision="I32">
245
+ <dim>-1</dim>
246
+ </port>
247
+ <port id="7" precision="I32">
248
+ <dim>-1</dim>
249
+ </port>
250
+ <port id="8" precision="U8">
251
+ <dim>-1</dim>
252
+ </port>
253
+ <port id="9" precision="BOOL">
254
+ <dim>-1</dim>
255
+ </port>
256
+ </output>
257
+ </layer>
258
+ <layer id="22" name="Constant_126291" type="Const" version="opset1">
259
+ <data element_type="u8" shape="1" offset="6423" size="1" />
260
+ <output>
261
+ <port id="0" precision="U8">
262
+ <dim>1</dim>
263
+ </port>
264
+ </output>
265
+ </layer>
266
+ <layer id="23" name="RegexSplit_126292" type="RegexSplit" version="extension">
267
+ <data behaviour="mergedwithprevious" invert="false" max_splits="-1" />
268
+ <input>
269
+ <port id="0" precision="I32">
270
+ <dim>-1</dim>
271
+ </port>
272
+ <port id="1" precision="I32">
273
+ <dim>-1</dim>
274
+ </port>
275
+ <port id="2" precision="I32">
276
+ <dim>-1</dim>
277
+ </port>
278
+ <port id="3" precision="I32">
279
+ <dim>-1</dim>
280
+ </port>
281
+ <port id="4" precision="U8">
282
+ <dim>-1</dim>
283
+ </port>
284
+ <port id="5" precision="BOOL">
285
+ <dim>-1</dim>
286
+ </port>
287
+ <port id="6" precision="U8">
288
+ <dim>1</dim>
289
+ </port>
290
+ </input>
291
+ <output>
292
+ <port id="7" precision="I32">
293
+ <dim>-1</dim>
294
+ </port>
295
+ <port id="8" precision="I32">
296
+ <dim>-1</dim>
297
+ </port>
298
+ <port id="9" precision="I32">
299
+ <dim>-1</dim>
300
+ </port>
301
+ <port id="10" precision="I32">
302
+ <dim>-1</dim>
303
+ </port>
304
+ <port id="11" precision="U8">
305
+ <dim>-1</dim>
306
+ </port>
307
+ <port id="12" precision="BOOL">
308
+ <dim>-1</dim>
309
+ </port>
310
+ </output>
311
+ </layer>
312
+ <layer id="24" name="Constant_126294" type="Const" version="opset1">
313
+ <data element_type="u8" shape="2955910" offset="6427" size="2955910" />
314
+ <output>
315
+ <port id="0" precision="U8">
316
+ <dim>2955910</dim>
317
+ </port>
318
+ </output>
319
+ </layer>
320
+ <layer id="25" name="StringTensorUnpack_126295" type="StringTensorUnpack" version="extension">
321
+ <data mode="begins_ends" />
322
+ <input>
323
+ <port id="0" precision="U8">
324
+ <dim>2955910</dim>
325
+ </port>
326
+ </input>
327
+ <output>
328
+ <port id="1" precision="I32">
329
+ <dim>-1</dim>
330
+ </port>
331
+ <port id="2" precision="I32">
332
+ <dim>-1</dim>
333
+ </port>
334
+ <port id="3" precision="U8">
335
+ <dim>-1</dim>
336
+ </port>
337
+ </output>
338
+ </layer>
339
+ <layer id="26" name="Constant_126300" type="Const" version="opset1">
340
+ <data element_type="u8" shape="5031736" offset="2962337" size="5031736" />
341
+ <output>
342
+ <port id="0" precision="U8">
343
+ <dim>5031736</dim>
344
+ </port>
345
+ </output>
346
+ </layer>
347
+ <layer id="27" name="StringTensorUnpack_126301" type="StringTensorUnpack" version="extension">
348
+ <data mode="begins_ends" />
349
+ <input>
350
+ <port id="0" precision="U8">
351
+ <dim>5031736</dim>
352
+ </port>
353
+ </input>
354
+ <output>
355
+ <port id="1" precision="I32">
356
+ <dim>-1</dim>
357
+ </port>
358
+ <port id="2" precision="I32">
359
+ <dim>-1</dim>
360
+ </port>
361
+ <port id="3" precision="U8">
362
+ <dim>-1</dim>
363
+ </port>
364
+ </output>
365
+ </layer>
366
+ <layer id="28" name="Constant_126303" type="Const" version="opset1">
367
+ <data element_type="u8" shape="4245743" offset="7994073" size="4245743" />
368
+ <output>
369
+ <port id="0" precision="U8">
370
+ <dim>4245743</dim>
371
+ </port>
372
+ </output>
373
+ </layer>
374
+ <layer id="29" name="StringTensorUnpack_126304" type="StringTensorUnpack" version="extension">
375
+ <data mode="begins_ends" />
376
+ <input>
377
+ <port id="0" precision="U8">
378
+ <dim>4245743</dim>
379
+ </port>
380
+ </input>
381
+ <output>
382
+ <port id="1" precision="I32">
383
+ <dim>-1</dim>
384
+ </port>
385
+ <port id="2" precision="I32">
386
+ <dim>-1</dim>
387
+ </port>
388
+ <port id="3" precision="U8">
389
+ <dim>-1</dim>
390
+ </port>
391
+ </output>
392
+ </layer>
393
+ <layer id="30" name="Constant_126297" type="Const" version="opset1">
394
+ <data element_type="u8" shape="4815" offset="12239816" size="4815" />
395
+ <output>
396
+ <port id="0" precision="U8">
397
+ <dim>4815</dim>
398
+ </port>
399
+ </output>
400
+ </layer>
401
+ <layer id="31" name="StringTensorUnpack_126298" type="StringTensorUnpack" version="extension">
402
+ <data mode="begins_ends" />
403
+ <input>
404
+ <port id="0" precision="U8">
405
+ <dim>4815</dim>
406
+ </port>
407
+ </input>
408
+ <output>
409
+ <port id="1" precision="I32">
410
+ <dim>-1</dim>
411
+ </port>
412
+ <port id="2" precision="I32">
413
+ <dim>-1</dim>
414
+ </port>
415
+ <port id="3" precision="U8">
416
+ <dim>-1</dim>
417
+ </port>
418
+ </output>
419
+ </layer>
420
+ <layer id="32" name="Constant_126305" type="Const" version="opset1">
421
+ <data element_type="i32" shape="248" offset="12244631" size="992" />
422
+ <output>
423
+ <port id="0" precision="I32">
424
+ <dim>248</dim>
425
+ </port>
426
+ </output>
427
+ </layer>
428
+ <layer id="33" name="BPETokenizer_126306" type="BPETokenizer" version="extension">
429
+ <data unk_token="&lt;unk>" fuse_unk="true" suffix_indicator="" end_suffix="" byte_fallback="true" cache_capacity="51200" />
430
+ <input>
431
+ <port id="0" precision="I32">
432
+ <dim>-1</dim>
433
+ </port>
434
+ <port id="1" precision="I32">
435
+ <dim>-1</dim>
436
+ </port>
437
+ <port id="2" precision="I32">
438
+ <dim>-1</dim>
439
+ </port>
440
+ <port id="3" precision="I32">
441
+ <dim>-1</dim>
442
+ </port>
443
+ <port id="4" precision="U8">
444
+ <dim>-1</dim>
445
+ </port>
446
+ <port id="5" precision="I32">
447
+ <dim>-1</dim>
448
+ </port>
449
+ <port id="6" precision="I32">
450
+ <dim>-1</dim>
451
+ </port>
452
+ <port id="7" precision="U8">
453
+ <dim>-1</dim>
454
+ </port>
455
+ <port id="8" precision="I32">
456
+ <dim>-1</dim>
457
+ </port>
458
+ <port id="9" precision="I32">
459
+ <dim>-1</dim>
460
+ </port>
461
+ <port id="10" precision="U8">
462
+ <dim>-1</dim>
463
+ </port>
464
+ <port id="11" precision="I32">
465
+ <dim>-1</dim>
466
+ </port>
467
+ <port id="12" precision="I32">
468
+ <dim>-1</dim>
469
+ </port>
470
+ <port id="13" precision="U8">
471
+ <dim>-1</dim>
472
+ </port>
473
+ <port id="14" precision="I32">
474
+ <dim>-1</dim>
475
+ </port>
476
+ <port id="15" precision="I32">
477
+ <dim>-1</dim>
478
+ </port>
479
+ <port id="16" precision="U8">
480
+ <dim>-1</dim>
481
+ </port>
482
+ <port id="17" precision="I32">
483
+ <dim>248</dim>
484
+ </port>
485
+ </input>
486
+ <output>
487
+ <port id="18" precision="I32">
488
+ <dim>-1</dim>
489
+ </port>
490
+ <port id="19" precision="I32">
491
+ <dim>-1</dim>
492
+ </port>
493
+ <port id="20" precision="I32">
494
+ <dim>-1</dim>
495
+ </port>
496
+ </output>
497
+ </layer>
498
+ <layer id="34" name="Subtract_126307" type="Subtract" version="opset1">
499
+ <data auto_broadcast="numpy" />
500
+ <input>
501
+ <port id="0" precision="I32">
502
+ <dim>-1</dim>
503
+ </port>
504
+ <port id="1" precision="I32">
505
+ <dim>-1</dim>
506
+ </port>
507
+ </input>
508
+ <output>
509
+ <port id="2" precision="I32">
510
+ <dim>-1</dim>
511
+ </port>
512
+ </output>
513
+ </layer>
514
+ <layer id="35" name="Constant_126308" type="Const" version="opset1">
515
+ <data element_type="i32" shape="" offset="12245623" size="4" />
516
+ <output>
517
+ <port id="0" precision="I32" />
518
+ </output>
519
+ </layer>
520
+ <layer id="36" name="Minimum_126309" type="Minimum" version="opset1">
521
+ <data auto_broadcast="numpy" />
522
+ <input>
523
+ <port id="0" precision="I32">
524
+ <dim>-1</dim>
525
+ </port>
526
+ <port id="1" precision="I32" />
527
+ </input>
528
+ <output>
529
+ <port id="2" precision="I32">
530
+ <dim>-1</dim>
531
+ </port>
532
+ </output>
533
+ </layer>
534
+ <layer id="37" name="Subtract_126310" type="Subtract" version="opset1">
535
+ <data auto_broadcast="numpy" />
536
+ <input>
537
+ <port id="0" precision="I32">
538
+ <dim>-1</dim>
539
+ </port>
540
+ <port id="1" precision="I32">
541
+ <dim>-1</dim>
542
+ </port>
543
+ </input>
544
+ <output>
545
+ <port id="2" precision="I32">
546
+ <dim>-1</dim>
547
+ </port>
548
+ </output>
549
+ </layer>
550
+ <layer id="38" name="Constant_126314" type="Const" version="opset1">
551
+ <data element_type="i32" shape="2" offset="12" size="8" />
552
+ <output>
553
+ <port id="0" precision="I32">
554
+ <dim>2</dim>
555
+ </port>
556
+ </output>
557
+ </layer>
558
+ <layer id="39" name="CombineSegments_126315" type="CombineSegments" version="extension">
559
+ <input>
560
+ <port id="0" precision="I32" />
561
+ <port id="1" precision="I32" />
562
+ <port id="2" precision="I32">
563
+ <dim>1</dim>
564
+ </port>
565
+ <port id="3" precision="I32">
566
+ <dim>-1</dim>
567
+ </port>
568
+ <port id="4" precision="I32">
569
+ <dim>-1</dim>
570
+ </port>
571
+ <port id="5" precision="I32">
572
+ <dim>-1</dim>
573
+ </port>
574
+ <port id="6" precision="I32">
575
+ <dim>2</dim>
576
+ </port>
577
+ </input>
578
+ <output>
579
+ <port id="7" precision="I32">
580
+ <dim>-1</dim>
581
+ </port>
582
+ <port id="8" precision="I32">
583
+ <dim>-1</dim>
584
+ </port>
585
+ <port id="9" precision="I32">
586
+ <dim>-1</dim>
587
+ </port>
588
+ <port id="10" precision="I32">
589
+ <dim>-1</dim>
590
+ </port>
591
+ <port id="11" precision="I32">
592
+ <dim>-1</dim>
593
+ </port>
594
+ <port id="12" precision="I32">
595
+ <dim>-1</dim>
596
+ </port>
597
+ </output>
598
+ </layer>
599
+ <layer id="40" name="Subtract_126316" type="Subtract" version="opset1">
600
+ <data auto_broadcast="numpy" />
601
+ <input>
602
+ <port id="0" precision="I32">
603
+ <dim>-1</dim>
604
+ </port>
605
+ <port id="1" precision="I32">
606
+ <dim>-1</dim>
607
+ </port>
608
+ </input>
609
+ <output>
610
+ <port id="2" precision="I32">
611
+ <dim>-1</dim>
612
+ </port>
613
+ </output>
614
+ </layer>
615
+ <layer id="41" name="Constant_126317" type="Const" version="opset1">
616
+ <data element_type="i32" shape="" offset="0" size="4" />
617
+ <output>
618
+ <port id="0" precision="I32" />
619
+ </output>
620
+ </layer>
621
+ <layer id="42" name="ReduceMax_126318" type="ReduceMax" version="opset1">
622
+ <data keep_dims="false" />
623
+ <input>
624
+ <port id="0" precision="I32">
625
+ <dim>-1</dim>
626
+ </port>
627
+ <port id="1" precision="I32" />
628
+ </input>
629
+ <output>
630
+ <port id="2" precision="I32" />
631
+ </output>
632
+ </layer>
633
+ <layer id="43" name="Constant_126319" type="Const" version="opset1">
634
+ <data element_type="i32" shape="" offset="0" size="4" />
635
+ <output>
636
+ <port id="0" precision="I32" />
637
+ </output>
638
+ </layer>
639
+ <layer id="44" name="RaggedToDense_126320" type="RaggedToDense" version="extension">
640
+ <data pad_right="false" />
641
+ <input>
642
+ <port id="0" precision="I32">
643
+ <dim>-1</dim>
644
+ </port>
645
+ <port id="1" precision="I32">
646
+ <dim>-1</dim>
647
+ </port>
648
+ <port id="2" precision="I32">
649
+ <dim>-1</dim>
650
+ </port>
651
+ <port id="3" precision="I32" />
652
+ <port id="4" precision="I32" />
653
+ </input>
654
+ <output>
655
+ <port id="5" precision="I32">
656
+ <dim>-1</dim>
657
+ <dim>-1</dim>
658
+ </port>
659
+ <port id="6" precision="BOOL">
660
+ <dim>-1</dim>
661
+ <dim>-1</dim>
662
+ </port>
663
+ </output>
664
+ </layer>
665
+ <layer id="45" name="Convert_126321" type="Convert" version="opset1">
666
+ <data destination_type="i32" />
667
+ <input>
668
+ <port id="0" precision="BOOL">
669
+ <dim>-1</dim>
670
+ <dim>-1</dim>
671
+ </port>
672
+ </input>
673
+ <output>
674
+ <port id="1" precision="I32">
675
+ <dim>-1</dim>
676
+ <dim>-1</dim>
677
+ </port>
678
+ </output>
679
+ </layer>
680
+ <layer id="46" name="Convert_126321" type="Convert" version="opset1">
681
+ <data destination_type="i64" />
682
+ <input>
683
+ <port id="0" precision="I32">
684
+ <dim>-1</dim>
685
+ <dim>-1</dim>
686
+ </port>
687
+ </input>
688
+ <output>
689
+ <port id="1" precision="I64" names="attention_mask">
690
+ <dim>-1</dim>
691
+ <dim>-1</dim>
692
+ </port>
693
+ </output>
694
+ </layer>
695
+ <layer id="48" name="RaggedToDense_126320.0" type="Convert" version="opset1">
696
+ <data destination_type="i64" />
697
+ <input>
698
+ <port id="0" precision="I32">
699
+ <dim>-1</dim>
700
+ <dim>-1</dim>
701
+ </port>
702
+ </input>
703
+ <output>
704
+ <port id="1" precision="I64" names="input_ids">
705
+ <dim>-1</dim>
706
+ <dim>-1</dim>
707
+ </port>
708
+ </output>
709
+ </layer>
710
+ <layer id="49" name="Result_126324" type="Result" version="opset1">
711
+ <input>
712
+ <port id="0" precision="I64">
713
+ <dim>-1</dim>
714
+ <dim>-1</dim>
715
+ </port>
716
+ </input>
717
+ </layer>
718
+ <layer id="47" name="Result_126326" type="Result" version="opset1">
719
+ <input>
720
+ <port id="0" precision="I64">
721
+ <dim>-1</dim>
722
+ <dim>-1</dim>
723
+ </port>
724
+ </input>
725
+ </layer>
726
+ </layers>
727
+ <edges>
728
+ <edge from-layer="0" from-port="0" to-layer="5" to-port="0" />
729
+ <edge from-layer="1" from-port="0" to-layer="39" to-port="0" />
730
+ <edge from-layer="2" from-port="0" to-layer="39" to-port="1" />
731
+ <edge from-layer="3" from-port="0" to-layer="39" to-port="2" />
732
+ <edge from-layer="4" from-port="0" to-layer="11" to-port="0" />
733
+ <edge from-layer="5" from-port="1" to-layer="6" to-port="0" />
734
+ <edge from-layer="5" from-port="3" to-layer="18" to-port="4" />
735
+ <edge from-layer="5" from-port="2" to-layer="18" to-port="3" />
736
+ <edge from-layer="5" from-port="1" to-layer="18" to-port="2" />
737
+ <edge from-layer="6" from-port="1" to-layer="9" to-port="0" />
738
+ <edge from-layer="7" from-port="0" to-layer="9" to-port="1" />
739
+ <edge from-layer="8" from-port="0" to-layer="9" to-port="2" />
740
+ <edge from-layer="9" from-port="3" to-layer="14" to-port="0" />
741
+ <edge from-layer="9" from-port="3" to-layer="11" to-port="1" />
742
+ <edge from-layer="10" from-port="0" to-layer="11" to-port="2" />
743
+ <edge from-layer="11" from-port="3" to-layer="18" to-port="0" />
744
+ <edge from-layer="12" from-port="0" to-layer="16" to-port="0" />
745
+ <edge from-layer="13" from-port="0" to-layer="14" to-port="1" />
746
+ <edge from-layer="14" from-port="2" to-layer="16" to-port="1" />
747
+ <edge from-layer="15" from-port="0" to-layer="16" to-port="2" />
748
+ <edge from-layer="16" from-port="3" to-layer="18" to-port="1" />
749
+ <edge from-layer="17" from-port="0" to-layer="18" to-port="5" />
750
+ <edge from-layer="18" from-port="8" to-layer="21" to-port="0" />
751
+ <edge from-layer="18" from-port="7" to-layer="23" to-port="1" />
752
+ <edge from-layer="18" from-port="6" to-layer="23" to-port="0" />
753
+ <edge from-layer="18" from-port="11" to-layer="21" to-port="3" />
754
+ <edge from-layer="18" from-port="10" to-layer="21" to-port="2" />
755
+ <edge from-layer="18" from-port="9" to-layer="21" to-port="1" />
756
+ <edge from-layer="19" from-port="0" to-layer="21" to-port="4" />
757
+ <edge from-layer="20" from-port="0" to-layer="21" to-port="5" />
758
+ <edge from-layer="21" from-port="6" to-layer="23" to-port="2" />
759
+ <edge from-layer="21" from-port="7" to-layer="23" to-port="3" />
760
+ <edge from-layer="21" from-port="8" to-layer="23" to-port="4" />
761
+ <edge from-layer="21" from-port="9" to-layer="23" to-port="5" />
762
+ <edge from-layer="22" from-port="0" to-layer="23" to-port="6" />
763
+ <edge from-layer="23" from-port="11" to-layer="33" to-port="4" />
764
+ <edge from-layer="23" from-port="10" to-layer="33" to-port="3" />
765
+ <edge from-layer="23" from-port="9" to-layer="33" to-port="2" />
766
+ <edge from-layer="23" from-port="7" to-layer="33" to-port="0" />
767
+ <edge from-layer="23" from-port="8" to-layer="33" to-port="1" />
768
+ <edge from-layer="24" from-port="0" to-layer="25" to-port="0" />
769
+ <edge from-layer="25" from-port="1" to-layer="33" to-port="5" />
770
+ <edge from-layer="25" from-port="2" to-layer="33" to-port="6" />
771
+ <edge from-layer="25" from-port="3" to-layer="33" to-port="7" />
772
+ <edge from-layer="26" from-port="0" to-layer="27" to-port="0" />
773
+ <edge from-layer="27" from-port="1" to-layer="33" to-port="8" />
774
+ <edge from-layer="27" from-port="2" to-layer="33" to-port="9" />
775
+ <edge from-layer="27" from-port="3" to-layer="33" to-port="10" />
776
+ <edge from-layer="28" from-port="0" to-layer="29" to-port="0" />
777
+ <edge from-layer="29" from-port="1" to-layer="33" to-port="11" />
778
+ <edge from-layer="29" from-port="2" to-layer="33" to-port="12" />
779
+ <edge from-layer="29" from-port="3" to-layer="33" to-port="13" />
780
+ <edge from-layer="30" from-port="0" to-layer="31" to-port="0" />
781
+ <edge from-layer="31" from-port="2" to-layer="33" to-port="15" />
782
+ <edge from-layer="31" from-port="3" to-layer="33" to-port="16" />
783
+ <edge from-layer="31" from-port="1" to-layer="33" to-port="14" />
784
+ <edge from-layer="32" from-port="0" to-layer="33" to-port="17" />
785
+ <edge from-layer="33" from-port="19" to-layer="34" to-port="0" />
786
+ <edge from-layer="33" from-port="18" to-layer="34" to-port="1" />
787
+ <edge from-layer="33" from-port="19" to-layer="37" to-port="0" />
788
+ <edge from-layer="33" from-port="20" to-layer="39" to-port="5" />
789
+ <edge from-layer="33" from-port="19" to-layer="39" to-port="4" />
790
+ <edge from-layer="34" from-port="2" to-layer="36" to-port="0" />
791
+ <edge from-layer="35" from-port="0" to-layer="36" to-port="1" />
792
+ <edge from-layer="36" from-port="2" to-layer="37" to-port="1" />
793
+ <edge from-layer="37" from-port="2" to-layer="39" to-port="3" />
794
+ <edge from-layer="38" from-port="0" to-layer="39" to-port="6" />
795
+ <edge from-layer="39" from-port="8" to-layer="40" to-port="0" />
796
+ <edge from-layer="39" from-port="7" to-layer="40" to-port="1" />
797
+ <edge from-layer="39" from-port="7" to-layer="44" to-port="0" />
798
+ <edge from-layer="39" from-port="8" to-layer="44" to-port="1" />
799
+ <edge from-layer="39" from-port="9" to-layer="44" to-port="2" />
800
+ <edge from-layer="40" from-port="2" to-layer="42" to-port="0" />
801
+ <edge from-layer="41" from-port="0" to-layer="42" to-port="1" />
802
+ <edge from-layer="42" from-port="2" to-layer="44" to-port="3" />
803
+ <edge from-layer="43" from-port="0" to-layer="44" to-port="4" />
804
+ <edge from-layer="44" from-port="6" to-layer="45" to-port="0" />
805
+ <edge from-layer="44" from-port="5" to-layer="48" to-port="0" />
806
+ <edge from-layer="45" from-port="1" to-layer="46" to-port="0" />
807
+ <edge from-layer="46" from-port="1" to-layer="47" to-port="0" />
808
+ <edge from-layer="48" from-port="1" to-layer="49" to-port="0" />
809
+ </edges>
810
+ <rt_info>
811
+ <add_attention_mask value="True" />
812
+ <add_prefix_space />
813
+ <add_special_tokens value="True" />
814
+ <bos_token_id value="2" />
815
+ <chat_template value="{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ raise_exception('System role not supported') }}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '&lt;start_of_turn>' + role + '&#10;' + message['content'] | trim + '&lt;end_of_turn>&#10;' }}{% endfor %}{% if add_generation_prompt %}{{'&lt;start_of_turn>model&#10;'}}{% endif %}" />
816
+ <clean_up_tokenization_spaces />
817
+ <detokenizer_input_type value="i64" />
818
+ <eos_token_id value="1" />
819
+ <handle_special_tokens_with_re />
820
+ <number_of_inputs value="1" />
821
+ <openvino_tokenizers_version value="2024.5.0.0" />
822
+ <openvino_version value="2024.5.0" />
823
+ <original_tokenizer_class value="&lt;class 'transformers.models.gemma.tokenization_gemma_fast.GemmaTokenizerFast'>" />
824
+ <pad_token_id value="0" />
825
+ <sentencepiece_version value="0.2.0" />
826
+ <skip_special_tokens value="True" />
827
+ <streaming_detokenizer value="False" />
828
+ <tiktoken_version value="0.8.0" />
829
+ <tokenizer_output_type value="i64" />
830
+ <tokenizers_version value="0.20.3" />
831
+ <transformers_version value="4.46.3" />
832
+ <use_max_padding value="False" />
833
+ <use_sentencepiece_backend value="False" />
834
+ <utf8_replace_mode />
835
+ <with_detokenizer value="True" />
836
+ </rt_info>
837
+ </net>