Hugging Face's logo

OpenVINO
/
InternVL2-8B-fp16-ov

Image-Text-to-Text
OpenVINO
multilingual
internvl_chat
nlp
vision
internvl
conversational
custom_code
Model card Files Community
openvino-ci commited on
Commit
00b4d16
·
verified ·
1 Parent(s): 7165169

Upload folder using huggingface_hub

Browse files
Files changed (19) hide show
  1. README.md +145 -0
  2. added_tokens.json +11 -0
  3. config.json +204 -0
  4. generation_config.json +8 -0
  5. openvino_detokenizer.bin +3 -0
  6. openvino_detokenizer.xml +149 -0
  7. openvino_language_model.bin +3 -0
  8. openvino_language_model.xml +0 -0
  9. openvino_text_embeddings_model.bin +3 -0
  10. openvino_text_embeddings_model.xml +113 -0
  11. openvino_tokenizer.bin +3 -0
  12. openvino_tokenizer.xml +1013 -0
  13. openvino_vision_embeddings_model.bin +3 -0
  14. openvino_vision_embeddings_model.xml +0 -0
  15. preprocessor_config.json +27 -0
  16. special_tokens_map.json +47 -0
  17. tokenization_internlm2.py +235 -0
  18. tokenizer.model +3 -0
  19. tokenizer_config.json +180 -0
README.md ADDED
@@ -0,0 +1,145 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ language:
4
+ - multilingual
5
+ pipeline_tag: image-text-to-text
6
+ tags:
7
+ - nlp
8
+ - vision
9
+ - internvl
10
+ base_model:
11
+ - OpenGVLab/InternVL2-8B
12
+ ---
13
+
14
+ # InternVL2-8B-fp16-ov
15
+
16
+ * Model creator: [OpenGVLab](https://huggingface.co/OpenGVLab)
17
+ * Original model: [InternVL2-8B](https://huggingface.co/OpenGVLab/InternVL2-8B)
18
+
19
+ ## Description
20
+
21
+ This is [OpenGVLab/InternVL2-8B](https://huggingface.co/OpenGVLab/InternVL2-8B) model converted to the [OpenVINO™ IR](https://docs.openvino.ai/2025/documentation/openvino-ir-format.html) (Intermediate Representation) format.
22
+
23
+
24
+ ## Compatibility
25
+
26
+ The provided OpenVINO™ IR model is compatible with:
27
+
28
+ * OpenVINO version 2025.2.0 and higher
29
+ * Optimum Intel 1.26.0 and higher
30
+
31
+ ## Running Model Inference with [Optimum Intel](https://huggingface.co/docs/optimum/intel/index)
32
+
33
+ 1. Install packages required for using [Optimum Intel](https://huggingface.co/docs/optimum/intel/index) integration with the OpenVINO backend:
34
+
35
+ ```
36
+ pip install --pre -U --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/pre-release openvino_tokenizers openvino
37
+
38
+ pip install git+https://github.com/huggingface/optimum-intel.git
39
+ ```
40
+
41
+ 2. Run model inference
42
+
43
+ ```
44
+ from PIL import Image
45
+ import requests
46
+ from optimum.intel.openvino import OVModelForVisualCausalLM
47
+ from transformers import AutoTokenizer, TextStreamer
48
+
49
+ model_id = "OpenVINO/InternVL2-8B-fp16-ov"
50
+
51
+ tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
52
+
53
+ ov_model = OVModelForVisualCausalLM.from_pretrained(model_id, trust_remote_code=True)
54
+ prompt = "What is unusual on this picture?"
55
+
56
+ url = "https://github.com/openvinotoolkit/openvino_notebooks/assets/29454499/d5fbbd1a-d484-415c-88cb-9986625b7b11"
57
+ image = Image.open(requests.get(url, stream=True).raw)
58
+
59
+ inputs = ov_model.preprocess_inputs(text=prompt, image=image, tokenizer=tokenizer, config=ov_model.config)
60
+
61
+ generation_args = {
62
+ "max_new_tokens": 100,
63
+ "streamer": TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
64
+ }
65
+
66
+ generate_ids = ov_model.generate(**inputs, **generation_args)
67
+
68
+ generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
69
+ response = tokenizer.batch_decode(generate_ids, skip_special_tokens=True)[0]
70
+
71
+ ```
72
+
73
+ ## Running Model Inference with [OpenVINO GenAI](https://github.com/openvinotoolkit/openvino.genai)
74
+
75
+ 1. Install packages required for using OpenVINO GenAI.
76
+ ```
77
+ pip install --pre -U --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/pre-release openvino openvino-tokenizers openvino-genai
78
+
79
+ pip install huggingface_hub
80
+ ```
81
+
82
+ 2. Download model from HuggingFace Hub
83
+
84
+ ```
85
+ import huggingface_hub as hf_hub
86
+
87
+ model_id = "OpenVINO/InternVL2-8B-fp16-ov"
88
+ model_path = "InternVL2-8B-fp16-ov"
89
+
90
+ hf_hub.snapshot_download(model_id, local_dir=model_path)
91
+
92
+ ```
93
+
94
+ 1. Run model inference:
95
+
96
+ ```
97
+ import openvino_genai as ov_genai
98
+ import requests
99
+ from PIL import Image
100
+ from io import BytesIO
101
+ import numpy as np
102
+ import openvino as ov
103
+
104
+ device = "CPU"
105
+ pipe = ov_genai.VLMPipeline(model_path, device)
106
+
107
+ def load_image(image_file):
108
+ if isinstance(image_file, str) and (image_file.startswith("http") or image_file.startswith("https")):
109
+ response = requests.get(image_file)
110
+ image = Image.open(BytesIO(response.content)).convert("RGB")
111
+ else:
112
+ image = Image.open(image_file).convert("RGB")
113
+ image_data = np.array(image.getdata()).reshape(1, image.size[1], image.size[0], 3).astype(np.byte)
114
+ return ov.Tensor(image_data)
115
+
116
+ prompt = "What is unusual on this picture?"
117
+
118
+ url = "https://github.com/openvinotoolkit/openvino_notebooks/assets/29454499/d5fbbd1a-d484-415c-88cb-9986625b7b11"
119
+ image_tensor = load_image(url)
120
+
121
+ def streamer(subword: str) -> bool:
122
+ print(subword, end="", flush=True)
123
+ return False
124
+
125
+ pipe.start_chat()
126
+ output = pipe.generate(prompt, image=image_tensor, max_new_tokens=100, streamer=streamer)
127
+ pipe.finish_chat()
128
+ ```
129
+
130
+ More GenAI usage examples can be found in OpenVINO GenAI library [docs](https://github.com/openvinotoolkit/openvino.genai/blob/master/src/README.md) and [samples](https://github.com/openvinotoolkit/openvino.genai?tab=readme-ov-file#openvino-genai-samples)
131
+
132
+
133
+ ## Limitations
134
+
135
+
136
+ Check the original [model card](https://huggingface.co/OpenGVLab/InternVL2-8B) for limitations.
137
+
138
+ ## Legal information
139
+
140
+ The original model is distributed under [MIT](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/mit.md) license. More details can be found in [original model card](https://huggingface.co/OpenGVLab/InternVL2-8B).
141
+
142
+ ## Disclaimer
143
+
144
+ 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.
145
+
added_tokens.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</box>": 92552,
3
+ "</img>": 92545,
4
+ "</quad>": 92548,
5
+ "</ref>": 92550,
6
+ "<IMG_CONTEXT>": 92546,
7
+ "<box>": 92551,
8
+ "<img>": 92544,
9
+ "<quad>": 92547,
10
+ "<ref>": 92549
11
+ }
config.json ADDED
@@ -0,0 +1,204 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_attn_implementation_autoset": true,
3
+ "_commit_hash": "6fb9ad6924f69424e57fab2ab061d707688f0296",
4
+ "architectures": [
5
+ "InternVLChatModel"
6
+ ],
7
+ "auto_map": {
8
+ "AutoConfig": "OpenGVLab/InternVL2-8B--configuration_internvl_chat.InternVLChatConfig",
9
+ "AutoModel": "OpenGVLab/InternVL2-8B--modeling_internvl_chat.InternVLChatModel",
10
+ "AutoModelForCausalLM": "OpenGVLab/InternVL2-8B--modeling_internvl_chat.InternVLChatModel"
11
+ },
12
+ "downsample_ratio": 0.5,
13
+ "dynamic_image_size": true,
14
+ "force_image_size": 448,
15
+ "img_context_token_id": 92546,
16
+ "llm_config": {
17
+ "_attn_implementation_autoset": true,
18
+ "_name_or_path": "internlm/internlm2_5-7b-chat",
19
+ "add_cross_attention": false,
20
+ "architectures": [
21
+ "InternLM2ForCausalLM"
22
+ ],
23
+ "attn_implementation": "eager",
24
+ "auto_map": {
25
+ "AutoConfig": "configuration_internlm2.InternLM2Config",
26
+ "AutoModel": "modeling_internlm2.InternLM2ForCausalLM",
27
+ "AutoModelForCausalLM": "modeling_internlm2.InternLM2ForCausalLM"
28
+ },
29
+ "bad_words_ids": null,
30
+ "begin_suppress_tokens": null,
31
+ "bias": false,
32
+ "bos_token_id": 1,
33
+ "chunk_size_feed_forward": 0,
34
+ "cross_attention_hidden_size": null,
35
+ "decoder_start_token_id": null,
36
+ "diversity_penalty": 0.0,
37
+ "do_sample": false,
38
+ "early_stopping": false,
39
+ "encoder_no_repeat_ngram_size": 0,
40
+ "eos_token_id": 2,
41
+ "exponential_decay_length_penalty": null,
42
+ "finetuning_task": null,
43
+ "forced_bos_token_id": null,
44
+ "forced_eos_token_id": null,
45
+ "hidden_act": "silu",
46
+ "hidden_size": 4096,
47
+ "id2label": {
48
+ "0": "LABEL_0",
49
+ "1": "LABEL_1"
50
+ },
51
+ "initializer_range": 0.02,
52
+ "intermediate_size": 14336,
53
+ "is_decoder": false,
54
+ "is_encoder_decoder": false,
55
+ "label2id": {
56
+ "LABEL_0": 0,
57
+ "LABEL_1": 1
58
+ },
59
+ "length_penalty": 1.0,
60
+ "max_length": 20,
61
+ "max_position_embeddings": 32768,
62
+ "min_length": 0,
63
+ "model_type": "internlm2",
64
+ "no_repeat_ngram_size": 0,
65
+ "num_attention_heads": 32,
66
+ "num_beam_groups": 1,
67
+ "num_beams": 1,
68
+ "num_hidden_layers": 32,
69
+ "num_key_value_heads": 8,
70
+ "num_return_sequences": 1,
71
+ "output_attentions": false,
72
+ "output_hidden_states": false,
73
+ "output_scores": false,
74
+ "pad_token_id": 2,
75
+ "prefix": null,
76
+ "pretraining_tp": 1,
77
+ "problem_type": null,
78
+ "pruned_heads": {},
79
+ "remove_invalid_values": false,
80
+ "repetition_penalty": 1.0,
81
+ "return_dict": true,
82
+ "return_dict_in_generate": false,
83
+ "rms_norm_eps": 1e-05,
84
+ "rope_scaling": {
85
+ "factor": 2.0,
86
+ "type": "dynamic"
87
+ },
88
+ "rope_theta": 1000000,
89
+ "sep_token_id": null,
90
+ "suppress_tokens": null,
91
+ "task_specific_params": null,
92
+ "temperature": 1.0,
93
+ "tf_legacy_loss": false,
94
+ "tie_encoder_decoder": false,
95
+ "tie_word_embeddings": false,
96
+ "tokenizer_class": null,
97
+ "top_k": 50,
98
+ "top_p": 1.0,
99
+ "torch_dtype": "bfloat16",
100
+ "torchscript": false,
101
+ "transformers_version": "4.51.3",
102
+ "typical_p": 1.0,
103
+ "use_bfloat16": true,
104
+ "use_cache": true,
105
+ "vocab_size": 92553
106
+ },
107
+ "max_dynamic_patch": 12,
108
+ "min_dynamic_patch": 1,
109
+ "model_type": "internvl_chat",
110
+ "ps_version": "v2",
111
+ "select_layer": -1,
112
+ "template": "internlm2-chat",
113
+ "tie_word_embeddings": false,
114
+ "torch_dtype": "float16",
115
+ "transformers_version": null,
116
+ "use_backbone_lora": 0,
117
+ "use_llm_lora": 0,
118
+ "use_thumbnail": true,
119
+ "vision_config": {
120
+ "_attn_implementation_autoset": true,
121
+ "_name_or_path": "",
122
+ "add_cross_attention": false,
123
+ "architectures": [
124
+ "InternVisionModel"
125
+ ],
126
+ "attention_dropout": 0.0,
127
+ "bad_words_ids": null,
128
+ "begin_suppress_tokens": null,
129
+ "bos_token_id": null,
130
+ "chunk_size_feed_forward": 0,
131
+ "cross_attention_hidden_size": null,
132
+ "decoder_start_token_id": null,
133
+ "diversity_penalty": 0.0,
134
+ "do_sample": false,
135
+ "drop_path_rate": 0.0,
136
+ "dropout": 0.0,
137
+ "early_stopping": false,
138
+ "encoder_no_repeat_ngram_size": 0,
139
+ "eos_token_id": null,
140
+ "exponential_decay_length_penalty": null,
141
+ "finetuning_task": null,
142
+ "forced_bos_token_id": null,
143
+ "forced_eos_token_id": null,
144
+ "hidden_act": "gelu",
145
+ "hidden_size": 1024,
146
+ "id2label": {
147
+ "0": "LABEL_0",
148
+ "1": "LABEL_1"
149
+ },
150
+ "image_size": 448,
151
+ "initializer_factor": 1.0,
152
+ "initializer_range": 0.02,
153
+ "intermediate_size": 4096,
154
+ "is_decoder": false,
155
+ "is_encoder_decoder": false,
156
+ "label2id": {
157
+ "LABEL_0": 0,
158
+ "LABEL_1": 1
159
+ },
160
+ "layer_norm_eps": 1e-06,
161
+ "length_penalty": 1.0,
162
+ "max_length": 20,
163
+ "min_length": 0,
164
+ "model_type": "intern_vit_6b",
165
+ "no_repeat_ngram_size": 0,
166
+ "norm_type": "layer_norm",
167
+ "num_attention_heads": 16,
168
+ "num_beam_groups": 1,
169
+ "num_beams": 1,
170
+ "num_channels": 3,
171
+ "num_hidden_layers": 24,
172
+ "num_return_sequences": 1,
173
+ "output_attentions": false,
174
+ "output_hidden_states": false,
175
+ "output_scores": false,
176
+ "pad_token_id": null,
177
+ "patch_size": 14,
178
+ "prefix": null,
179
+ "problem_type": null,
180
+ "pruned_heads": {},
181
+ "qk_normalization": false,
182
+ "qkv_bias": true,
183
+ "remove_invalid_values": false,
184
+ "repetition_penalty": 1.0,
185
+ "return_dict": true,
186
+ "return_dict_in_generate": false,
187
+ "sep_token_id": null,
188
+ "suppress_tokens": null,
189
+ "task_specific_params": null,
190
+ "temperature": 1.0,
191
+ "tf_legacy_loss": false,
192
+ "tie_encoder_decoder": false,
193
+ "tie_word_embeddings": true,
194
+ "tokenizer_class": null,
195
+ "top_k": 50,
196
+ "top_p": 1.0,
197
+ "torch_dtype": "bfloat16",
198
+ "torchscript": false,
199
+ "transformers_version": "4.51.3",
200
+ "typical_p": 1.0,
201
+ "use_bfloat16": true,
202
+ "use_flash_attn": false
203
+ }
204
+ }
generation_config.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 92542,
5
+ 92543
6
+ ],
7
+ "transformers_version": "4.51.3"
8
+ }
openvino_detokenizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9d99982bd38cc642f98134aabf7650a1ce0d28e7978c945c238d7620a8260d29
3
+ size 1477889
openvino_detokenizer.xml ADDED
@@ -0,0 +1,149 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="detokenizer" version="11">
3
+ <layers>
4
+ <layer id="0" name="Parameter_218308" type="Parameter" version="opset1">
5
+ <data shape="?,?" element_type="i64" />
6
+ <output>
7
+ <port id="0" precision="I64" names="Parameter_218308">
8
+ <dim>-1</dim>
9
+ <dim>-1</dim>
10
+ </port>
11
+ </output>
12
+ </layer>
13
+ <layer id="1" name="Constant_218222" type="Const" version="opset1">
14
+ <data element_type="u8" shape="1477889" offset="0" size="1477889" />
15
+ <output>
16
+ <port id="0" precision="U8">
17
+ <dim>1477889</dim>
18
+ </port>
19
+ </output>
20
+ </layer>
21
+ <layer id="2" name="Convert_218497" type="Convert" version="opset1">
22
+ <data destination_type="i32" />
23
+ <input>
24
+ <port id="0" precision="I64">
25
+ <dim>-1</dim>
26
+ <dim>-1</dim>
27
+ </port>
28
+ </input>
29
+ <output>
30
+ <port id="1" precision="I32">
31
+ <dim>-1</dim>
32
+ <dim>-1</dim>
33
+ </port>
34
+ </output>
35
+ </layer>
36
+ <layer id="3" name="SentencepieceDetokenizer_218309" type="SentencepieceDetokenizer" version="extension">
37
+ <input>
38
+ <port id="0" precision="U8">
39
+ <dim>1477889</dim>
40
+ </port>
41
+ <port id="1" precision="I32">
42
+ <dim>-1</dim>
43
+ <dim>-1</dim>
44
+ </port>
45
+ </input>
46
+ <output>
47
+ <port id="2" precision="I32">
48
+ <dim>-1</dim>
49
+ </port>
50
+ <port id="3" precision="I32">
51
+ <dim>-1</dim>
52
+ </port>
53
+ <port id="4" precision="U8">
54
+ <dim>-1</dim>
55
+ </port>
56
+ </output>
57
+ </layer>
58
+ <layer id="4" name="UTF8Validate_218310" type="UTF8Validate" version="extension">
59
+ <data replace_mode="true" />
60
+ <input>
61
+ <port id="0" precision="I32">
62
+ <dim>-1</dim>
63
+ </port>
64
+ <port id="1" precision="I32">
65
+ <dim>-1</dim>
66
+ </port>
67
+ <port id="2" precision="U8">
68
+ <dim>-1</dim>
69
+ </port>
70
+ </input>
71
+ <output>
72
+ <port id="3" precision="I32">
73
+ <dim>-1</dim>
74
+ </port>
75
+ <port id="4" precision="I32">
76
+ <dim>-1</dim>
77
+ </port>
78
+ <port id="5" precision="U8">
79
+ <dim>-1</dim>
80
+ </port>
81
+ </output>
82
+ </layer>
83
+ <layer id="5" name="StringTensorPack_218311" type="StringTensorPack" version="opset15">
84
+ <input>
85
+ <port id="0" precision="I32">
86
+ <dim>-1</dim>
87
+ </port>
88
+ <port id="1" precision="I32">
89
+ <dim>-1</dim>
90
+ </port>
91
+ <port id="2" precision="U8">
92
+ <dim>-1</dim>
93
+ </port>
94
+ </input>
95
+ <output>
96
+ <port id="3" precision="STRING" names="string_output">
97
+ <dim>-1</dim>
98
+ </port>
99
+ </output>
100
+ </layer>
101
+ <layer id="6" name="Result_218312" type="Result" version="opset1" output_names="string_output">
102
+ <input>
103
+ <port id="0" precision="STRING">
104
+ <dim>-1</dim>
105
+ </port>
106
+ </input>
107
+ </layer>
108
+ </layers>
109
+ <edges>
110
+ <edge from-layer="0" from-port="0" to-layer="2" to-port="0" />
111
+ <edge from-layer="1" from-port="0" to-layer="3" to-port="0" />
112
+ <edge from-layer="2" from-port="1" to-layer="3" to-port="1" />
113
+ <edge from-layer="3" from-port="2" to-layer="4" to-port="0" />
114
+ <edge from-layer="3" from-port="3" to-layer="4" to-port="1" />
115
+ <edge from-layer="3" from-port="4" to-layer="4" to-port="2" />
116
+ <edge from-layer="4" from-port="3" to-layer="5" to-port="0" />
117
+ <edge from-layer="4" from-port="4" to-layer="5" to-port="1" />
118
+ <edge from-layer="4" from-port="5" to-layer="5" to-port="2" />
119
+ <edge from-layer="5" from-port="3" to-layer="6" to-port="0" />
120
+ </edges>
121
+ <rt_info>
122
+ <add_attention_mask value="True" />
123
+ <add_prefix_space />
124
+ <add_special_tokens value="True" />
125
+ <bos_token_id value="1" />
126
+ <chat_template value="{{ bos_token }}{% for message in messages %}{{'&lt;|im_start|>' + message['role'] + '&#10;' + message['content'] + '&lt;|im_end|>' + '&#10;'}}{% endfor %}{% if add_generation_prompt %}{{ '&lt;|im_start|>assistant&#10;' }}{% endif %}" />
127
+ <clean_up_tokenization_spaces value="False" />
128
+ <detokenizer_input_type value="i64" />
129
+ <eos_token_id value="2" />
130
+ <handle_special_tokens_with_re value="True" />
131
+ <max_length />
132
+ <number_of_inputs value="1" />
133
+ <openvino_tokenizers_version value="2025.2.0.1-567-7885335c24b" />
134
+ <openvino_version value="2025.2.0-19140-c01cd93e24d-releases/2025/2" />
135
+ <original_tokenizer_class value="&lt;class 'transformers_modules.OpenGVLab.InternVL2-8B.6fb9ad6924f69424e57fab2ab061d707688f0296.tokenization_internlm2.InternLM2Tokenizer'>" />
136
+ <pad_token_id value="2" />
137
+ <sentencepiece_version value="0.2.1" />
138
+ <skip_special_tokens value="True" />
139
+ <streaming_detokenizer value="False" />
140
+ <tiktoken_version value="0.9.0" />
141
+ <tokenizer_output_type value="i64" />
142
+ <tokenizers_version value="0.21.4" />
143
+ <transformers_version value="4.51.3" />
144
+ <use_max_padding value="False" />
145
+ <use_sentencepiece_backend value="False" />
146
+ <utf8_replace_mode value="replace" />
147
+ <with_detokenizer value="True" />
148
+ </rt_info>
149
+ </net>
openvino_language_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d091a2235f2c316e2f8b8575a73b89bb73ce28f484d3f426f88c8a14d373d76
3
+ size 14734147850
openvino_language_model.xml ADDED
The diff for this file is too large to render. See raw diff
 
openvino_text_embeddings_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a5a801bc9bd0441d6ad2a954084942f5276248ab6d4c5d2e83c578c8757bdb3
3
+ size 758194180
openvino_text_embeddings_model.xml ADDED
@@ -0,0 +1,113 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="Model3" version="11">
3
+ <layers>
4
+ <layer id="0" name="input" type="Parameter" version="opset1">
5
+ <data shape="?,?" element_type="i64" />
6
+ <output>
7
+ <port id="0" precision="I64" names="input">
8
+ <dim>-1</dim>
9
+ <dim>-1</dim>
10
+ </port>
11
+ </output>
12
+ </layer>
13
+ <layer id="1" name="self.weight" type="Const" version="opset1">
14
+ <data element_type="f16" shape="92553, 4096" offset="0" size="758194176" />
15
+ <output>
16
+ <port id="0" precision="FP16" names="self.weight">
17
+ <dim>92553</dim>
18
+ <dim>4096</dim>
19
+ </port>
20
+ </output>
21
+ </layer>
22
+ <layer id="2" name="ov_ext::embedding/Convert" type="Convert" version="opset1">
23
+ <data destination_type="f32" />
24
+ <rt_info>
25
+ <attribute name="decompression" version="0" />
26
+ </rt_info>
27
+ <input>
28
+ <port id="0" precision="FP16">
29
+ <dim>92553</dim>
30
+ <dim>4096</dim>
31
+ </port>
32
+ </input>
33
+ <output>
34
+ <port id="1" precision="FP32">
35
+ <dim>92553</dim>
36
+ <dim>4096</dim>
37
+ </port>
38
+ </output>
39
+ </layer>
40
+ <layer id="3" name="ov_ext::embedding/Convert_1" type="Convert" version="opset1">
41
+ <data destination_type="i32" />
42
+ <input>
43
+ <port id="0" precision="I64">
44
+ <dim>-1</dim>
45
+ <dim>-1</dim>
46
+ </port>
47
+ </input>
48
+ <output>
49
+ <port id="1" precision="I32">
50
+ <dim>-1</dim>
51
+ <dim>-1</dim>
52
+ </port>
53
+ </output>
54
+ </layer>
55
+ <layer id="4" name="ov_ext::embedding/Constant" type="Const" version="opset1">
56
+ <data element_type="i32" shape="" offset="758194176" size="4" />
57
+ <output>
58
+ <port id="0" precision="I32" />
59
+ </output>
60
+ </layer>
61
+ <layer id="5" name="ov_ext::embedding/Gather" type="Gather" version="opset8">
62
+ <data batch_dims="0" />
63
+ <input>
64
+ <port id="0" precision="FP32">
65
+ <dim>92553</dim>
66
+ <dim>4096</dim>
67
+ </port>
68
+ <port id="1" precision="I32">
69
+ <dim>-1</dim>
70
+ <dim>-1</dim>
71
+ </port>
72
+ <port id="2" precision="I32" />
73
+ </input>
74
+ <output>
75
+ <port id="3" precision="FP32" names="inputs_embeds">
76
+ <dim>-1</dim>
77
+ <dim>-1</dim>
78
+ <dim>4096</dim>
79
+ </port>
80
+ </output>
81
+ </layer>
82
+ <layer id="6" name="Result_25601" type="Result" version="opset1" output_names="inputs_embeds">
83
+ <input>
84
+ <port id="0" precision="FP32">
85
+ <dim>-1</dim>
86
+ <dim>-1</dim>
87
+ <dim>4096</dim>
88
+ </port>
89
+ </input>
90
+ </layer>
91
+ </layers>
92
+ <edges>
93
+ <edge from-layer="0" from-port="0" to-layer="3" to-port="0" />
94
+ <edge from-layer="1" from-port="0" to-layer="2" to-port="0" />
95
+ <edge from-layer="2" from-port="1" to-layer="5" to-port="0" />
96
+ <edge from-layer="3" from-port="1" to-layer="5" to-port="1" />
97
+ <edge from-layer="4" from-port="0" to-layer="5" to-port="2" />
98
+ <edge from-layer="5" from-port="3" to-layer="6" to-port="0" />
99
+ </edges>
100
+ <rt_info>
101
+ <Runtime_version value="2025.2.0-19140-c01cd93e24d-releases/2025/2" />
102
+ <conversion_parameters>
103
+ <framework value="pytorch" />
104
+ <is_python_object value="True" />
105
+ </conversion_parameters>
106
+ <optimum>
107
+ <optimum_intel_version value="1.26.0.dev0+e9c57b9" />
108
+ <optimum_version value="1.27.0" />
109
+ <pytorch_version value="2.8.0+cpu" />
110
+ <transformers_version value="4.51.3" />
111
+ </optimum>
112
+ </rt_info>
113
+ </net>
openvino_tokenizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bd1ac23b5528df5dc2926bd19ff8d352d00b20fcaa778948306e70591ce6f153
3
+ size 1478353
openvino_tokenizer.xml ADDED
@@ -0,0 +1,1013 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="tokenizer" version="11">
3
+ <layers>
4
+ <layer id="0" name="string_input" type="Parameter" version="opset1">
5
+ <data shape="?" element_type="string" />
6
+ <output>
7
+ <port id="0" precision="STRING" names="string_input">
8
+ <dim>-1</dim>
9
+ </port>
10
+ </output>
11
+ </layer>
12
+ <layer id="1" name="Constant_218282" 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_218221" type="Const" version="opset1">
19
+ <data element_type="u8" shape="1477889" offset="4" size="1477889" />
20
+ <output>
21
+ <port id="0" precision="U8">
22
+ <dim>1477889</dim>
23
+ </port>
24
+ </output>
25
+ </layer>
26
+ <layer id="3" name="Constant_218225" type="Const" version="opset1">
27
+ <data element_type="i32" shape="18" offset="1477893" size="72" />
28
+ <output>
29
+ <port id="0" precision="I32">
30
+ <dim>18</dim>
31
+ </port>
32
+ </output>
33
+ </layer>
34
+ <layer id="4" name="Constant_218227" type="Const" version="opset1">
35
+ <data element_type="i32" shape="18" offset="1477965" size="72" />
36
+ <output>
37
+ <port id="0" precision="I32">
38
+ <dim>18</dim>
39
+ </port>
40
+ </output>
41
+ </layer>
42
+ <layer id="5" name="Constant_218229" type="Const" version="opset1">
43
+ <data element_type="u8" shape="148" offset="1478037" size="148" />
44
+ <output>
45
+ <port id="0" precision="U8">
46
+ <dim>148</dim>
47
+ </port>
48
+ </output>
49
+ </layer>
50
+ <layer id="6" name="Constant_218230" type="Const" version="opset1">
51
+ <data element_type="i32" shape="18" offset="1478185" size="72" />
52
+ <output>
53
+ <port id="0" precision="I32">
54
+ <dim>18</dim>
55
+ </port>
56
+ </output>
57
+ </layer>
58
+ <layer id="7" name="SentencepieceTokenizer_218231" type="SentencepieceTokenizer" version="extension">
59
+ <data nbest_size="1" alpha="1" add_bos="false" add_eos="false" reverse="true" />
60
+ <input>
61
+ <port id="0" precision="U8">
62
+ <dim>1477889</dim>
63
+ </port>
64
+ <port id="1" precision="STRING">
65
+ <dim>-1</dim>
66
+ </port>
67
+ <port id="2" precision="I32">
68
+ <dim>18</dim>
69
+ </port>
70
+ <port id="3" precision="I32">
71
+ <dim>18</dim>
72
+ </port>
73
+ <port id="4" precision="U8">
74
+ <dim>148</dim>
75
+ </port>
76
+ <port id="5" precision="I32">
77
+ <dim>18</dim>
78
+ </port>
79
+ </input>
80
+ <output>
81
+ <port id="6" precision="I64">
82
+ <dim>-1</dim>
83
+ <dim>2</dim>
84
+ </port>
85
+ <port id="7" precision="I32">
86
+ <dim>-1</dim>
87
+ </port>
88
+ <port id="8" precision="I64">
89
+ <dim>2</dim>
90
+ </port>
91
+ </output>
92
+ </layer>
93
+ <layer id="8" name="Convert_218241" type="Const" version="opset1">
94
+ <data element_type="i64" shape="2" offset="1478257" size="16" />
95
+ <output>
96
+ <port id="0" precision="I64">
97
+ <dim>2</dim>
98
+ </port>
99
+ </output>
100
+ </layer>
101
+ <layer id="9" name="Add_218242" type="Add" version="opset1">
102
+ <data auto_broadcast="numpy" />
103
+ <input>
104
+ <port id="0" precision="I64">
105
+ <dim>2</dim>
106
+ </port>
107
+ <port id="1" precision="I64">
108
+ <dim>2</dim>
109
+ </port>
110
+ </input>
111
+ <output>
112
+ <port id="2" precision="I64">
113
+ <dim>2</dim>
114
+ </port>
115
+ </output>
116
+ </layer>
117
+ <layer id="10" name="Constant_218247" type="Const" version="opset1">
118
+ <data element_type="i64" shape="1" offset="1478273" size="8" />
119
+ <output>
120
+ <port id="0" precision="I64">
121
+ <dim>1</dim>
122
+ </port>
123
+ </output>
124
+ </layer>
125
+ <layer id="11" name="Constant_218248" type="Const" version="opset1">
126
+ <data element_type="i64" shape="1" offset="1478281" size="8" />
127
+ <output>
128
+ <port id="0" precision="I64">
129
+ <dim>1</dim>
130
+ </port>
131
+ </output>
132
+ </layer>
133
+ <layer id="12" name="Constant_218249" type="Const" version="opset1">
134
+ <data element_type="i64" shape="1" offset="1478281" size="8" />
135
+ <output>
136
+ <port id="0" precision="I64">
137
+ <dim>1</dim>
138
+ </port>
139
+ </output>
140
+ </layer>
141
+ <layer id="13" name="Constant_218251" type="Const" version="opset1">
142
+ <data element_type="i64" shape="1" offset="1478273" size="8" />
143
+ <output>
144
+ <port id="0" precision="I64">
145
+ <dim>1</dim>
146
+ </port>
147
+ </output>
148
+ </layer>
149
+ <layer id="14" name="Slice_218250" type="Slice" version="opset8">
150
+ <input>
151
+ <port id="0" precision="I64">
152
+ <dim>2</dim>
153
+ </port>
154
+ <port id="1" precision="I64">
155
+ <dim>1</dim>
156
+ </port>
157
+ <port id="2" precision="I64">
158
+ <dim>1</dim>
159
+ </port>
160
+ <port id="3" precision="I64">
161
+ <dim>1</dim>
162
+ </port>
163
+ <port id="4" precision="I64">
164
+ <dim>1</dim>
165
+ </port>
166
+ </input>
167
+ <output>
168
+ <port id="5" precision="I64">
169
+ <dim>1</dim>
170
+ </port>
171
+ </output>
172
+ </layer>
173
+ <layer id="15" name="Constant_218283" type="Const" version="opset1">
174
+ <data element_type="i64" shape="1" offset="1478281" size="8" />
175
+ <output>
176
+ <port id="0" precision="I64">
177
+ <dim>1</dim>
178
+ </port>
179
+ </output>
180
+ </layer>
181
+ <layer id="16" name="Concat_218284" type="Concat" version="opset1">
182
+ <data axis="0" />
183
+ <input>
184
+ <port id="0" precision="I64">
185
+ <dim>1</dim>
186
+ </port>
187
+ <port id="1" precision="I64">
188
+ <dim>1</dim>
189
+ </port>
190
+ </input>
191
+ <output>
192
+ <port id="2" precision="I64">
193
+ <dim>2</dim>
194
+ </port>
195
+ </output>
196
+ </layer>
197
+ <layer id="17" name="Broadcast_218285" type="Broadcast" version="opset3">
198
+ <data mode="numpy" />
199
+ <input>
200
+ <port id="0" precision="I32" />
201
+ <port id="1" precision="I64">
202
+ <dim>2</dim>
203
+ </port>
204
+ </input>
205
+ <output>
206
+ <port id="2" precision="I32">
207
+ <dim>-1</dim>
208
+ <dim>1</dim>
209
+ </port>
210
+ </output>
211
+ </layer>
212
+ <layer id="18" name="Constant_218232" type="Const" version="opset1">
213
+ <data element_type="i32" shape="" offset="1478289" size="4" />
214
+ <output>
215
+ <port id="0" precision="I32" />
216
+ </output>
217
+ </layer>
218
+ <layer id="19" name="Broadcast_218233" type="Broadcast" version="opset3">
219
+ <data mode="numpy" />
220
+ <input>
221
+ <port id="0" precision="I32" />
222
+ <port id="1" precision="I64">
223
+ <dim>2</dim>
224
+ </port>
225
+ </input>
226
+ <output>
227
+ <port id="2" precision="I32">
228
+ <dim>-1</dim>
229
+ <dim>-1</dim>
230
+ </port>
231
+ </output>
232
+ </layer>
233
+ <layer id="20" name="Constant_218234" type="Const" version="opset1">
234
+ <data element_type="i32" shape="" offset="0" size="4" />
235
+ <output>
236
+ <port id="0" precision="I32" />
237
+ </output>
238
+ </layer>
239
+ <layer id="21" name="ShapeOf_218235" type="ShapeOf" version="opset3">
240
+ <data output_type="i64" />
241
+ <input>
242
+ <port id="0" precision="I32">
243
+ <dim>-1</dim>
244
+ </port>
245
+ </input>
246
+ <output>
247
+ <port id="1" precision="I64">
248
+ <dim>1</dim>
249
+ </port>
250
+ </output>
251
+ </layer>
252
+ <layer id="22" name="Broadcast_218236" type="Broadcast" version="opset3">
253
+ <data mode="numpy" />
254
+ <input>
255
+ <port id="0" precision="I32" />
256
+ <port id="1" precision="I64">
257
+ <dim>1</dim>
258
+ </port>
259
+ </input>
260
+ <output>
261
+ <port id="2" precision="I32">
262
+ <dim>-1</dim>
263
+ </port>
264
+ </output>
265
+ </layer>
266
+ <layer id="23" name="ScatterNDUpdate_218239" type="ScatterNDUpdate" version="opset4">
267
+ <input>
268
+ <port id="0" precision="I32">
269
+ <dim>-1</dim>
270
+ <dim>-1</dim>
271
+ </port>
272
+ <port id="1" precision="I64">
273
+ <dim>-1</dim>
274
+ <dim>2</dim>
275
+ </port>
276
+ <port id="2" precision="I32">
277
+ <dim>-1</dim>
278
+ </port>
279
+ </input>
280
+ <output>
281
+ <port id="3" precision="I32">
282
+ <dim>-1</dim>
283
+ <dim>-1</dim>
284
+ </port>
285
+ </output>
286
+ </layer>
287
+ <layer id="24" name="Concat_218286" type="Concat" version="opset1">
288
+ <data axis="1" />
289
+ <input>
290
+ <port id="0" precision="I32">
291
+ <dim>-1</dim>
292
+ <dim>1</dim>
293
+ </port>
294
+ <port id="1" precision="I32">
295
+ <dim>-1</dim>
296
+ <dim>-1</dim>
297
+ </port>
298
+ </input>
299
+ <output>
300
+ <port id="2" precision="I32">
301
+ <dim>-1</dim>
302
+ <dim>-1</dim>
303
+ </port>
304
+ </output>
305
+ </layer>
306
+ <layer id="25" name="Constant_218290" type="Const" version="opset1">
307
+ <data element_type="i64" shape="1" offset="1478293" size="8" />
308
+ <output>
309
+ <port id="0" precision="I64">
310
+ <dim>1</dim>
311
+ </port>
312
+ </output>
313
+ </layer>
314
+ <layer id="26" name="Reverse_218291" type="Reverse" version="opset1">
315
+ <data mode="index" />
316
+ <input>
317
+ <port id="0" precision="I32">
318
+ <dim>-1</dim>
319
+ <dim>-1</dim>
320
+ </port>
321
+ <port id="1" precision="I64">
322
+ <dim>1</dim>
323
+ </port>
324
+ </input>
325
+ <output>
326
+ <port id="2" precision="I32">
327
+ <dim>-1</dim>
328
+ <dim>-1</dim>
329
+ </port>
330
+ </output>
331
+ </layer>
332
+ <layer id="27" name="Constant_218299" type="Const" version="opset1">
333
+ <data element_type="i64" shape="1" offset="1478301" size="8" />
334
+ <output>
335
+ <port id="0" precision="I64">
336
+ <dim>1</dim>
337
+ </port>
338
+ </output>
339
+ </layer>
340
+ <layer id="28" name="Constant_218300" type="Const" version="opset1">
341
+ <data element_type="i64" shape="1" offset="1478309" size="8" />
342
+ <output>
343
+ <port id="0" precision="I64">
344
+ <dim>1</dim>
345
+ </port>
346
+ </output>
347
+ </layer>
348
+ <layer id="29" name="Constant_218301" type="Const" version="opset1">
349
+ <data element_type="i64" shape="1" offset="1478281" size="8" />
350
+ <output>
351
+ <port id="0" precision="I64">
352
+ <dim>1</dim>
353
+ </port>
354
+ </output>
355
+ </layer>
356
+ <layer id="30" name="Constant_218302" type="Const" version="opset1">
357
+ <data element_type="i64" shape="1" offset="1478293" size="8" />
358
+ <output>
359
+ <port id="0" precision="I64">
360
+ <dim>1</dim>
361
+ </port>
362
+ </output>
363
+ </layer>
364
+ <layer id="31" name="Slice_218303" type="Slice" version="opset8">
365
+ <input>
366
+ <port id="0" precision="I32">
367
+ <dim>-1</dim>
368
+ <dim>-1</dim>
369
+ </port>
370
+ <port id="1" precision="I64">
371
+ <dim>1</dim>
372
+ </port>
373
+ <port id="2" precision="I64">
374
+ <dim>1</dim>
375
+ </port>
376
+ <port id="3" precision="I64">
377
+ <dim>1</dim>
378
+ </port>
379
+ <port id="4" precision="I64">
380
+ <dim>1</dim>
381
+ </port>
382
+ </input>
383
+ <output>
384
+ <port id="5" precision="I32">
385
+ <dim>-1</dim>
386
+ <dim>-1</dim>
387
+ </port>
388
+ </output>
389
+ </layer>
390
+ <layer id="32" name="Slice_218303.0" type="Convert" version="opset1">
391
+ <data destination_type="i64" />
392
+ <input>
393
+ <port id="0" precision="I32">
394
+ <dim>-1</dim>
395
+ <dim>-1</dim>
396
+ </port>
397
+ </input>
398
+ <output>
399
+ <port id="1" precision="I64" names="attention_mask">
400
+ <dim>-1</dim>
401
+ <dim>-1</dim>
402
+ </port>
403
+ </output>
404
+ </layer>
405
+ <layer id="34" name="Constant_218287" type="Const" version="opset1">
406
+ <data element_type="i32" shape="" offset="1478317" size="4" />
407
+ <output>
408
+ <port id="0" precision="I32" />
409
+ </output>
410
+ </layer>
411
+ <layer id="35" name="Broadcast_218288" type="Broadcast" version="opset3">
412
+ <data mode="bidirectional" />
413
+ <input>
414
+ <port id="0" precision="I32" />
415
+ <port id="1" precision="I64">
416
+ <dim>2</dim>
417
+ </port>
418
+ </input>
419
+ <output>
420
+ <port id="2" precision="I32">
421
+ <dim>-1</dim>
422
+ <dim>-1</dim>
423
+ </port>
424
+ </output>
425
+ </layer>
426
+ <layer id="36" name="Constant_218262" type="Const" version="opset1">
427
+ <data element_type="i64" shape="" offset="1478273" size="8" />
428
+ <output>
429
+ <port id="0" precision="I64" />
430
+ </output>
431
+ </layer>
432
+ <layer id="37" name="Constant_218244" type="Const" version="opset1">
433
+ <data element_type="i64" shape="" offset="1478273" size="8" />
434
+ <output>
435
+ <port id="0" precision="I64" />
436
+ </output>
437
+ </layer>
438
+ <layer id="38" name="Constant_218245" type="Const" version="opset1">
439
+ <data element_type="i64" shape="" offset="1478273" size="8" />
440
+ <output>
441
+ <port id="0" precision="I64" />
442
+ </output>
443
+ </layer>
444
+ <layer id="39" name="Gather_218246" type="Gather" version="opset8">
445
+ <data batch_dims="0" />
446
+ <input>
447
+ <port id="0" precision="I64">
448
+ <dim>2</dim>
449
+ </port>
450
+ <port id="1" precision="I64" />
451
+ <port id="2" precision="I64" />
452
+ </input>
453
+ <output>
454
+ <port id="3" precision="I64" />
455
+ </output>
456
+ </layer>
457
+ <layer id="40" name="Constant_218263" type="Const" version="opset1">
458
+ <data element_type="i64" shape="" offset="1478281" size="8" />
459
+ <output>
460
+ <port id="0" precision="I64" />
461
+ </output>
462
+ </layer>
463
+ <layer id="41" name="Range_218264" type="Range" version="opset4">
464
+ <data output_type="i64" />
465
+ <input>
466
+ <port id="0" precision="I64" />
467
+ <port id="1" precision="I64" />
468
+ <port id="2" precision="I64" />
469
+ </input>
470
+ <output>
471
+ <port id="3" precision="I64">
472
+ <dim>-1</dim>
473
+ </port>
474
+ </output>
475
+ </layer>
476
+ <layer id="42" name="Constant_218265" type="Const" version="opset1">
477
+ <data element_type="i64" shape="1" offset="1478281" size="8" />
478
+ <output>
479
+ <port id="0" precision="I64">
480
+ <dim>1</dim>
481
+ </port>
482
+ </output>
483
+ </layer>
484
+ <layer id="43" name="Concat_218266" type="Concat" version="opset1">
485
+ <data axis="0" />
486
+ <input>
487
+ <port id="0" precision="I64">
488
+ <dim>1</dim>
489
+ </port>
490
+ <port id="1" precision="I64">
491
+ <dim>1</dim>
492
+ </port>
493
+ </input>
494
+ <output>
495
+ <port id="2" precision="I64">
496
+ <dim>2</dim>
497
+ </port>
498
+ </output>
499
+ </layer>
500
+ <layer id="44" name="Broadcast_218267" type="Broadcast" version="opset3">
501
+ <data mode="bidirectional" />
502
+ <input>
503
+ <port id="0" precision="I64">
504
+ <dim>-1</dim>
505
+ </port>
506
+ <port id="1" precision="I64">
507
+ <dim>2</dim>
508
+ </port>
509
+ </input>
510
+ <output>
511
+ <port id="2" precision="I64">
512
+ <dim>1</dim>
513
+ <dim>-1</dim>
514
+ </port>
515
+ </output>
516
+ </layer>
517
+ <layer id="45" name="Constant_218268" type="Const" version="opset1">
518
+ <data element_type="i64" shape="2" offset="1478321" size="16" />
519
+ <output>
520
+ <port id="0" precision="I64">
521
+ <dim>2</dim>
522
+ </port>
523
+ </output>
524
+ </layer>
525
+ <layer id="46" name="Transpose_218269" type="Transpose" version="opset1">
526
+ <input>
527
+ <port id="0" precision="I64">
528
+ <dim>1</dim>
529
+ <dim>-1</dim>
530
+ </port>
531
+ <port id="1" precision="I64">
532
+ <dim>2</dim>
533
+ </port>
534
+ </input>
535
+ <output>
536
+ <port id="2" precision="I64">
537
+ <dim>-1</dim>
538
+ <dim>1</dim>
539
+ </port>
540
+ </output>
541
+ </layer>
542
+ <layer id="47" name="Constant_218270" type="Const" version="opset1">
543
+ <data element_type="i64" shape="2" offset="1478337" size="16" />
544
+ <output>
545
+ <port id="0" precision="I64">
546
+ <dim>2</dim>
547
+ </port>
548
+ </output>
549
+ </layer>
550
+ <layer id="48" name="Reshape_218271" type="Reshape" version="opset1">
551
+ <data special_zero="false" />
552
+ <input>
553
+ <port id="0" precision="I64">
554
+ <dim>-1</dim>
555
+ <dim>1</dim>
556
+ </port>
557
+ <port id="1" precision="I64">
558
+ <dim>2</dim>
559
+ </port>
560
+ </input>
561
+ <output>
562
+ <port id="2" precision="I64">
563
+ <dim>-1</dim>
564
+ <dim>1</dim>
565
+ </port>
566
+ </output>
567
+ </layer>
568
+ <layer id="49" name="Constant_218272" type="Const" version="opset1">
569
+ <data element_type="i64" shape="" offset="1478293" size="8" />
570
+ <output>
571
+ <port id="0" precision="I64" />
572
+ </output>
573
+ </layer>
574
+ <layer id="50" name="ReduceSum_218273" type="ReduceSum" version="opset1">
575
+ <data keep_dims="true" />
576
+ <input>
577
+ <port id="0" precision="I32">
578
+ <dim>-1</dim>
579
+ <dim>-1</dim>
580
+ </port>
581
+ <port id="1" precision="I64" />
582
+ </input>
583
+ <output>
584
+ <port id="2" precision="I32">
585
+ <dim>-1</dim>
586
+ <dim>1</dim>
587
+ </port>
588
+ </output>
589
+ </layer>
590
+ <layer id="51" name="Convert_218274" type="Convert" version="opset1">
591
+ <data destination_type="i64" />
592
+ <input>
593
+ <port id="0" precision="I32">
594
+ <dim>-1</dim>
595
+ <dim>1</dim>
596
+ </port>
597
+ </input>
598
+ <output>
599
+ <port id="1" precision="I64">
600
+ <dim>-1</dim>
601
+ <dim>1</dim>
602
+ </port>
603
+ </output>
604
+ </layer>
605
+ <layer id="52" name="Reshape_218276" type="Const" version="opset1">
606
+ <data element_type="i64" shape="1, 1" offset="1478273" size="8" />
607
+ <output>
608
+ <port id="0" precision="I64">
609
+ <dim>1</dim>
610
+ <dim>1</dim>
611
+ </port>
612
+ </output>
613
+ </layer>
614
+ <layer id="53" name="Add_218277" type="Add" version="opset1">
615
+ <data auto_broadcast="numpy" />
616
+ <input>
617
+ <port id="0" precision="I64">
618
+ <dim>-1</dim>
619
+ <dim>1</dim>
620
+ </port>
621
+ <port id="1" precision="I64">
622
+ <dim>1</dim>
623
+ <dim>1</dim>
624
+ </port>
625
+ </input>
626
+ <output>
627
+ <port id="2" precision="I64">
628
+ <dim>-1</dim>
629
+ <dim>1</dim>
630
+ </port>
631
+ </output>
632
+ </layer>
633
+ <layer id="54" name="Constant_218278" type="Const" version="opset1">
634
+ <data element_type="i64" shape="2" offset="1478337" size="16" />
635
+ <output>
636
+ <port id="0" precision="I64">
637
+ <dim>2</dim>
638
+ </port>
639
+ </output>
640
+ </layer>
641
+ <layer id="55" name="Reshape_218279" type="Reshape" version="opset1">
642
+ <data special_zero="false" />
643
+ <input>
644
+ <port id="0" precision="I64">
645
+ <dim>-1</dim>
646
+ <dim>1</dim>
647
+ </port>
648
+ <port id="1" precision="I64">
649
+ <dim>2</dim>
650
+ </port>
651
+ </input>
652
+ <output>
653
+ <port id="2" precision="I64">
654
+ <dim>-1</dim>
655
+ <dim>1</dim>
656
+ </port>
657
+ </output>
658
+ </layer>
659
+ <layer id="56" name="Concat_218280" type="Concat" version="opset1">
660
+ <data axis="1" />
661
+ <input>
662
+ <port id="0" precision="I64">
663
+ <dim>-1</dim>
664
+ <dim>1</dim>
665
+ </port>
666
+ <port id="1" precision="I64">
667
+ <dim>-1</dim>
668
+ <dim>1</dim>
669
+ </port>
670
+ </input>
671
+ <output>
672
+ <port id="2" precision="I64">
673
+ <dim>-1</dim>
674
+ <dim>2</dim>
675
+ </port>
676
+ </output>
677
+ </layer>
678
+ <layer id="57" name="Concat_218281" type="Concat" version="opset1">
679
+ <data axis="0" />
680
+ <input>
681
+ <port id="0" precision="I64">
682
+ <dim>-1</dim>
683
+ <dim>2</dim>
684
+ </port>
685
+ <port id="1" precision="I64">
686
+ <dim>-1</dim>
687
+ <dim>2</dim>
688
+ </port>
689
+ </input>
690
+ <output>
691
+ <port id="2" precision="I64">
692
+ <dim>-1</dim>
693
+ <dim>2</dim>
694
+ </port>
695
+ </output>
696
+ </layer>
697
+ <layer id="58" name="Constant_218243" type="Const" version="opset1">
698
+ <data element_type="i32" shape="1, 1" offset="0" size="4" />
699
+ <output>
700
+ <port id="0" precision="I32">
701
+ <dim>1</dim>
702
+ <dim>1</dim>
703
+ </port>
704
+ </output>
705
+ </layer>
706
+ <layer id="59" name="Broadcast_218254" type="Broadcast" version="opset3">
707
+ <data mode="bidirectional" />
708
+ <input>
709
+ <port id="0" precision="I32">
710
+ <dim>1</dim>
711
+ <dim>1</dim>
712
+ </port>
713
+ <port id="1" precision="I64">
714
+ <dim>2</dim>
715
+ </port>
716
+ </input>
717
+ <output>
718
+ <port id="2" precision="I32">
719
+ <dim>-1</dim>
720
+ <dim>1</dim>
721
+ </port>
722
+ </output>
723
+ </layer>
724
+ <layer id="60" name="Constant_218255" type="Const" version="opset1">
725
+ <data element_type="i64" shape="1" offset="1478293" size="8" />
726
+ <output>
727
+ <port id="0" precision="I64">
728
+ <dim>1</dim>
729
+ </port>
730
+ </output>
731
+ </layer>
732
+ <layer id="61" name="Reshape_218256" type="Reshape" version="opset1">
733
+ <data special_zero="false" />
734
+ <input>
735
+ <port id="0" precision="I32">
736
+ <dim>-1</dim>
737
+ <dim>1</dim>
738
+ </port>
739
+ <port id="1" precision="I64">
740
+ <dim>1</dim>
741
+ </port>
742
+ </input>
743
+ <output>
744
+ <port id="2" precision="I32">
745
+ <dim>-1</dim>
746
+ </port>
747
+ </output>
748
+ </layer>
749
+ <layer id="62" name="Concat_218257" type="Concat" version="opset1">
750
+ <data axis="0" />
751
+ <input>
752
+ <port id="0" precision="I32">
753
+ <dim>-1</dim>
754
+ </port>
755
+ <port id="1" precision="I32">
756
+ <dim>-1</dim>
757
+ </port>
758
+ </input>
759
+ <output>
760
+ <port id="2" precision="I32">
761
+ <dim>-1</dim>
762
+ </port>
763
+ </output>
764
+ </layer>
765
+ <layer id="63" name="ScatterNDUpdate_218289" type="ScatterNDUpdate" version="opset4">
766
+ <input>
767
+ <port id="0" precision="I32">
768
+ <dim>-1</dim>
769
+ <dim>-1</dim>
770
+ </port>
771
+ <port id="1" precision="I64">
772
+ <dim>-1</dim>
773
+ <dim>2</dim>
774
+ </port>
775
+ <port id="2" precision="I32">
776
+ <dim>-1</dim>
777
+ </port>
778
+ </input>
779
+ <output>
780
+ <port id="3" precision="I32">
781
+ <dim>-1</dim>
782
+ <dim>-1</dim>
783
+ </port>
784
+ </output>
785
+ </layer>
786
+ <layer id="64" name="Constant_218292" type="Const" version="opset1">
787
+ <data element_type="i64" shape="1" offset="1478293" size="8" />
788
+ <output>
789
+ <port id="0" precision="I64">
790
+ <dim>1</dim>
791
+ </port>
792
+ </output>
793
+ </layer>
794
+ <layer id="65" name="Reverse_218293" type="Reverse" version="opset1">
795
+ <data mode="index" />
796
+ <input>
797
+ <port id="0" precision="I32">
798
+ <dim>-1</dim>
799
+ <dim>-1</dim>
800
+ </port>
801
+ <port id="1" precision="I64">
802
+ <dim>1</dim>
803
+ </port>
804
+ </input>
805
+ <output>
806
+ <port id="2" precision="I32">
807
+ <dim>-1</dim>
808
+ <dim>-1</dim>
809
+ </port>
810
+ </output>
811
+ </layer>
812
+ <layer id="66" name="Constant_218294" type="Const" version="opset1">
813
+ <data element_type="i64" shape="1" offset="1478301" size="8" />
814
+ <output>
815
+ <port id="0" precision="I64">
816
+ <dim>1</dim>
817
+ </port>
818
+ </output>
819
+ </layer>
820
+ <layer id="67" name="Constant_218295" type="Const" version="opset1">
821
+ <data element_type="i64" shape="1" offset="1478309" size="8" />
822
+ <output>
823
+ <port id="0" precision="I64">
824
+ <dim>1</dim>
825
+ </port>
826
+ </output>
827
+ </layer>
828
+ <layer id="68" name="Constant_218296" type="Const" version="opset1">
829
+ <data element_type="i64" shape="1" offset="1478281" size="8" />
830
+ <output>
831
+ <port id="0" precision="I64">
832
+ <dim>1</dim>
833
+ </port>
834
+ </output>
835
+ </layer>
836
+ <layer id="69" name="Constant_218297" type="Const" version="opset1">
837
+ <data element_type="i64" shape="1" offset="1478293" size="8" />
838
+ <output>
839
+ <port id="0" precision="I64">
840
+ <dim>1</dim>
841
+ </port>
842
+ </output>
843
+ </layer>
844
+ <layer id="70" name="Slice_218298" type="Slice" version="opset8">
845
+ <input>
846
+ <port id="0" precision="I32">
847
+ <dim>-1</dim>
848
+ <dim>-1</dim>
849
+ </port>
850
+ <port id="1" precision="I64">
851
+ <dim>1</dim>
852
+ </port>
853
+ <port id="2" precision="I64">
854
+ <dim>1</dim>
855
+ </port>
856
+ <port id="3" precision="I64">
857
+ <dim>1</dim>
858
+ </port>
859
+ <port id="4" precision="I64">
860
+ <dim>1</dim>
861
+ </port>
862
+ </input>
863
+ <output>
864
+ <port id="5" precision="I32">
865
+ <dim>-1</dim>
866
+ <dim>-1</dim>
867
+ </port>
868
+ </output>
869
+ </layer>
870
+ <layer id="71" name="Slice_218298.0" type="Convert" version="opset1">
871
+ <data destination_type="i64" />
872
+ <input>
873
+ <port id="0" precision="I32">
874
+ <dim>-1</dim>
875
+ <dim>-1</dim>
876
+ </port>
877
+ </input>
878
+ <output>
879
+ <port id="1" precision="I64" names="input_ids">
880
+ <dim>-1</dim>
881
+ <dim>-1</dim>
882
+ </port>
883
+ </output>
884
+ </layer>
885
+ <layer id="72" name="Result_218304" type="Result" version="opset1" output_names="input_ids">
886
+ <input>
887
+ <port id="0" precision="I64">
888
+ <dim>-1</dim>
889
+ <dim>-1</dim>
890
+ </port>
891
+ </input>
892
+ </layer>
893
+ <layer id="33" name="Result_218305" type="Result" version="opset1" output_names="attention_mask">
894
+ <input>
895
+ <port id="0" precision="I64">
896
+ <dim>-1</dim>
897
+ <dim>-1</dim>
898
+ </port>
899
+ </input>
900
+ </layer>
901
+ </layers>
902
+ <edges>
903
+ <edge from-layer="0" from-port="0" to-layer="7" to-port="1" />
904
+ <edge from-layer="1" from-port="0" to-layer="17" to-port="0" />
905
+ <edge from-layer="2" from-port="0" to-layer="7" to-port="0" />
906
+ <edge from-layer="3" from-port="0" to-layer="7" to-port="2" />
907
+ <edge from-layer="4" from-port="0" to-layer="7" to-port="3" />
908
+ <edge from-layer="5" from-port="0" to-layer="7" to-port="4" />
909
+ <edge from-layer="6" from-port="0" to-layer="7" to-port="5" />
910
+ <edge from-layer="7" from-port="8" to-layer="9" to-port="0" />
911
+ <edge from-layer="7" from-port="7" to-layer="62" to-port="0" />
912
+ <edge from-layer="7" from-port="6" to-layer="57" to-port="0" />
913
+ <edge from-layer="7" from-port="6" to-layer="23" to-port="1" />
914
+ <edge from-layer="7" from-port="7" to-layer="21" to-port="0" />
915
+ <edge from-layer="7" from-port="8" to-layer="19" to-port="1" />
916
+ <edge from-layer="8" from-port="0" to-layer="9" to-port="1" />
917
+ <edge from-layer="9" from-port="2" to-layer="35" to-port="1" />
918
+ <edge from-layer="9" from-port="2" to-layer="39" to-port="0" />
919
+ <edge from-layer="9" from-port="2" to-layer="14" to-port="0" />
920
+ <edge from-layer="10" from-port="0" to-layer="14" to-port="1" />
921
+ <edge from-layer="11" from-port="0" to-layer="14" to-port="2" />
922
+ <edge from-layer="12" from-port="0" to-layer="14" to-port="3" />
923
+ <edge from-layer="13" from-port="0" to-layer="14" to-port="4" />
924
+ <edge from-layer="14" from-port="5" to-layer="16" to-port="0" />
925
+ <edge from-layer="14" from-port="5" to-layer="43" to-port="1" />
926
+ <edge from-layer="15" from-port="0" to-layer="16" to-port="1" />
927
+ <edge from-layer="16" from-port="2" to-layer="17" to-port="1" />
928
+ <edge from-layer="16" from-port="2" to-layer="59" to-port="1" />
929
+ <edge from-layer="17" from-port="2" to-layer="24" to-port="0" />
930
+ <edge from-layer="18" from-port="0" to-layer="19" to-port="0" />
931
+ <edge from-layer="19" from-port="2" to-layer="23" to-port="0" />
932
+ <edge from-layer="20" from-port="0" to-layer="22" to-port="0" />
933
+ <edge from-layer="21" from-port="1" to-layer="22" to-port="1" />
934
+ <edge from-layer="22" from-port="2" to-layer="23" to-port="2" />
935
+ <edge from-layer="23" from-port="3" to-layer="24" to-port="1" />
936
+ <edge from-layer="23" from-port="3" to-layer="50" to-port="0" />
937
+ <edge from-layer="24" from-port="2" to-layer="26" to-port="0" />
938
+ <edge from-layer="25" from-port="0" to-layer="26" to-port="1" />
939
+ <edge from-layer="26" from-port="2" to-layer="31" to-port="0" />
940
+ <edge from-layer="27" from-port="0" to-layer="31" to-port="1" />
941
+ <edge from-layer="28" from-port="0" to-layer="31" to-port="2" />
942
+ <edge from-layer="29" from-port="0" to-layer="31" to-port="3" />
943
+ <edge from-layer="30" from-port="0" to-layer="31" to-port="4" />
944
+ <edge from-layer="31" from-port="5" to-layer="32" to-port="0" />
945
+ <edge from-layer="32" from-port="1" to-layer="33" to-port="0" />
946
+ <edge from-layer="34" from-port="0" to-layer="35" to-port="0" />
947
+ <edge from-layer="35" from-port="2" to-layer="63" to-port="0" />
948
+ <edge from-layer="36" from-port="0" to-layer="41" to-port="0" />
949
+ <edge from-layer="37" from-port="0" to-layer="39" to-port="1" />
950
+ <edge from-layer="38" from-port="0" to-layer="39" to-port="2" />
951
+ <edge from-layer="39" from-port="3" to-layer="41" to-port="1" />
952
+ <edge from-layer="40" from-port="0" to-layer="41" to-port="2" />
953
+ <edge from-layer="41" from-port="3" to-layer="44" to-port="0" />
954
+ <edge from-layer="42" from-port="0" to-layer="43" to-port="0" />
955
+ <edge from-layer="43" from-port="2" to-layer="44" to-port="1" />
956
+ <edge from-layer="44" from-port="2" to-layer="46" to-port="0" />
957
+ <edge from-layer="45" from-port="0" to-layer="46" to-port="1" />
958
+ <edge from-layer="46" from-port="2" to-layer="48" to-port="0" />
959
+ <edge from-layer="47" from-port="0" to-layer="48" to-port="1" />
960
+ <edge from-layer="48" from-port="2" to-layer="56" to-port="0" />
961
+ <edge from-layer="49" from-port="0" to-layer="50" to-port="1" />
962
+ <edge from-layer="50" from-port="2" to-layer="51" to-port="0" />
963
+ <edge from-layer="51" from-port="1" to-layer="53" to-port="0" />
964
+ <edge from-layer="52" from-port="0" to-layer="53" to-port="1" />
965
+ <edge from-layer="53" from-port="2" to-layer="55" to-port="0" />
966
+ <edge from-layer="54" from-port="0" to-layer="55" to-port="1" />
967
+ <edge from-layer="55" from-port="2" to-layer="56" to-port="1" />
968
+ <edge from-layer="56" from-port="2" to-layer="57" to-port="1" />
969
+ <edge from-layer="57" from-port="2" to-layer="63" to-port="1" />
970
+ <edge from-layer="58" from-port="0" to-layer="59" to-port="0" />
971
+ <edge from-layer="59" from-port="2" to-layer="61" to-port="0" />
972
+ <edge from-layer="60" from-port="0" to-layer="61" to-port="1" />
973
+ <edge from-layer="61" from-port="2" to-layer="62" to-port="1" />
974
+ <edge from-layer="62" from-port="2" to-layer="63" to-port="2" />
975
+ <edge from-layer="63" from-port="3" to-layer="65" to-port="0" />
976
+ <edge from-layer="64" from-port="0" to-layer="65" to-port="1" />
977
+ <edge from-layer="65" from-port="2" to-layer="70" to-port="0" />
978
+ <edge from-layer="66" from-port="0" to-layer="70" to-port="1" />
979
+ <edge from-layer="67" from-port="0" to-layer="70" to-port="2" />
980
+ <edge from-layer="68" from-port="0" to-layer="70" to-port="3" />
981
+ <edge from-layer="69" from-port="0" to-layer="70" to-port="4" />
982
+ <edge from-layer="70" from-port="5" to-layer="71" to-port="0" />
983
+ <edge from-layer="71" from-port="1" to-layer="72" to-port="0" />
984
+ </edges>
985
+ <rt_info>
986
+ <add_attention_mask value="True" />
987
+ <add_prefix_space />
988
+ <add_special_tokens value="True" />
989
+ <bos_token_id value="1" />
990
+ <chat_template value="{{ bos_token }}{% for message in messages %}{{'&lt;|im_start|>' + message['role'] + '&#10;' + message['content'] + '&lt;|im_end|>' + '&#10;'}}{% endfor %}{% if add_generation_prompt %}{{ '&lt;|im_start|>assistant&#10;' }}{% endif %}" />
991
+ <clean_up_tokenization_spaces value="False" />
992
+ <detokenizer_input_type value="i64" />
993
+ <eos_token_id value="2" />
994
+ <handle_special_tokens_with_re value="True" />
995
+ <max_length />
996
+ <number_of_inputs value="1" />
997
+ <openvino_tokenizers_version value="2025.2.0.1-567-7885335c24b" />
998
+ <openvino_version value="2025.2.0-19140-c01cd93e24d-releases/2025/2" />
999
+ <original_tokenizer_class value="&lt;class 'transformers_modules.OpenGVLab.InternVL2-8B.6fb9ad6924f69424e57fab2ab061d707688f0296.tokenization_internlm2.InternLM2Tokenizer'>" />
1000
+ <pad_token_id value="2" />
1001
+ <sentencepiece_version value="0.2.1" />
1002
+ <skip_special_tokens value="True" />
1003
+ <streaming_detokenizer value="False" />
1004
+ <tiktoken_version value="0.9.0" />
1005
+ <tokenizer_output_type value="i64" />
1006
+ <tokenizers_version value="0.21.4" />
1007
+ <transformers_version value="4.51.3" />
1008
+ <use_max_padding value="False" />
1009
+ <use_sentencepiece_backend value="False" />
1010
+ <utf8_replace_mode value="replace" />
1011
+ <with_detokenizer value="True" />
1012
+ </rt_info>
1013
+ </net>
openvino_vision_embeddings_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c4cf4961af6a361c06b10215eedd6564d8482b21c6cbb2abeb3e8ce5141879b
3
+ size 675166622
openvino_vision_embeddings_model.xml ADDED
The diff for this file is too large to render. See raw diff
 
preprocessor_config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "crop_size": {
3
+ "height": 448,
4
+ "width": 448
5
+ },
6
+ "do_center_crop": true,
7
+ "do_convert_rgb": true,
8
+ "do_normalize": true,
9
+ "do_rescale": true,
10
+ "do_resize": true,
11
+ "image_mean": [
12
+ 0.485,
13
+ 0.456,
14
+ 0.406
15
+ ],
16
+ "image_processor_type": "CLIPImageProcessor",
17
+ "image_std": [
18
+ 0.229,
19
+ 0.224,
20
+ 0.225
21
+ ],
22
+ "resample": 3,
23
+ "rescale_factor": 0.00392156862745098,
24
+ "size": {
25
+ "shortest_edge": 448
26
+ }
27
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|action_start|>",
6
+ "<|action_end|>",
7
+ "<|interpreter|>",
8
+ "<|plugin|>",
9
+ "<img>",
10
+ "</img>",
11
+ "<IMG_CONTEXT>",
12
+ "<quad>",
13
+ "</quad>",
14
+ "<ref>",
15
+ "</ref>",
16
+ "<box>",
17
+ "</box>"
18
+ ],
19
+ "bos_token": {
20
+ "content": "<s>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "eos_token": {
27
+ "content": "</s>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ },
33
+ "pad_token": {
34
+ "content": "</s>",
35
+ "lstrip": false,
36
+ "normalized": false,
37
+ "rstrip": false,
38
+ "single_word": false
39
+ },
40
+ "unk_token": {
41
+ "content": "<unk>",
42
+ "lstrip": false,
43
+ "normalized": false,
44
+ "rstrip": false,
45
+ "single_word": false
46
+ }
47
+ }
tokenization_internlm2.py ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) The InternLM team and The HuggingFace Inc. team. All rights reserved.
2
+ #
3
+ # This code is based on transformers/src/transformers/models/llama/tokenization_llama.py
4
+ #
5
+ # Licensed under the Apache License, Version 2.0 (the "License");
6
+ # you may not use this file except in compliance with the License.
7
+ # You may obtain a copy of the License at
8
+ #
9
+ # http://www.apache.org/licenses/LICENSE-2.0
10
+ #
11
+ # Unless required by applicable law or agreed to in writing, software
12
+ # distributed under the License is distributed on an "AS IS" BASIS,
13
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14
+ # See the License for the specific language governing permissions and
15
+ # limitations under the License.
16
+
17
+ """Tokenization classes for InternLM."""
18
+ import os
19
+ from shutil import copyfile
20
+ from typing import Any, Dict, List, Optional, Tuple
21
+
22
+ import sentencepiece as spm
23
+ from transformers.tokenization_utils import PreTrainedTokenizer
24
+ from transformers.utils import logging
25
+
26
+ logger = logging.get_logger(__name__)
27
+
28
+ VOCAB_FILES_NAMES = {'vocab_file': './tokenizer.model'}
29
+
30
+ PRETRAINED_VOCAB_FILES_MAP = {}
31
+
32
+
33
+ # Modified from transformers.model.llama.tokenization_llama.LlamaTokenizer
34
+ class InternLM2Tokenizer(PreTrainedTokenizer):
35
+ """
36
+ Construct a InternLM2 tokenizer. Based on byte-level Byte-Pair-Encoding.
37
+
38
+ Args:
39
+ vocab_file (`str`):
40
+ Path to the vocabulary file.
41
+ """
42
+
43
+ vocab_files_names = VOCAB_FILES_NAMES
44
+ pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
45
+ model_input_names = ['input_ids', 'attention_mask']
46
+ _auto_class = 'AutoTokenizer'
47
+
48
+ def __init__(
49
+ self,
50
+ vocab_file,
51
+ unk_token='<unk>',
52
+ bos_token='<s>',
53
+ eos_token='</s>',
54
+ pad_token='</s>',
55
+ sp_model_kwargs: Optional[Dict[str, Any]] = None,
56
+ add_bos_token=True,
57
+ add_eos_token=False,
58
+ decode_with_prefix_space=False,
59
+ clean_up_tokenization_spaces=False,
60
+ **kwargs,
61
+ ):
62
+ self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
63
+ self.vocab_file = vocab_file
64
+ self.add_bos_token = add_bos_token
65
+ self.add_eos_token = add_eos_token
66
+ self.decode_with_prefix_space = decode_with_prefix_space
67
+ self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
68
+ self.sp_model.Load(vocab_file)
69
+ self._no_prefix_space_tokens = None
70
+ super().__init__(
71
+ bos_token=bos_token,
72
+ eos_token=eos_token,
73
+ unk_token=unk_token,
74
+ pad_token=pad_token,
75
+ clean_up_tokenization_spaces=clean_up_tokenization_spaces,
76
+ **kwargs,
77
+ )
78
+
79
+ @property
80
+ def no_prefix_space_tokens(self):
81
+ if self._no_prefix_space_tokens is None:
82
+ vocab = self.convert_ids_to_tokens(list(range(self.vocab_size)))
83
+ self._no_prefix_space_tokens = {i for i, tok in enumerate(vocab) if not tok.startswith('▁')}
84
+ return self._no_prefix_space_tokens
85
+
86
+ @property
87
+ def vocab_size(self):
88
+ """Returns vocab size"""
89
+ return self.sp_model.get_piece_size()
90
+
91
+ @property
92
+ def bos_token_id(self) -> Optional[int]:
93
+ return self.sp_model.bos_id()
94
+
95
+ @property
96
+ def eos_token_id(self) -> Optional[int]:
97
+ return self.sp_model.eos_id()
98
+
99
+ def get_vocab(self):
100
+ """Returns vocab as a dict"""
101
+ vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
102
+ vocab.update(self.added_tokens_encoder)
103
+ return vocab
104
+
105
+ def _tokenize(self, text):
106
+ """Returns a tokenized string."""
107
+ return self.sp_model.encode(text, out_type=str)
108
+
109
+ def _convert_token_to_id(self, token):
110
+ """Converts a token (str) in an id using the vocab."""
111
+ return self.sp_model.piece_to_id(token)
112
+
113
+ def _convert_id_to_token(self, index):
114
+ """Converts an index (integer) in a token (str) using the vocab."""
115
+ token = self.sp_model.IdToPiece(index)
116
+ return token
117
+
118
+ def _maybe_add_prefix_space(self, tokens, decoded):
119
+ if tokens and tokens[0] not in self.no_prefix_space_tokens:
120
+ return ' ' + decoded
121
+ else:
122
+ return decoded
123
+
124
+ def convert_tokens_to_string(self, tokens):
125
+ """Converts a sequence of tokens (string) in a single string."""
126
+ current_sub_tokens = []
127
+ out_string = ''
128
+ prev_is_special = False
129
+ for token in tokens:
130
+ # make sure that special tokens are not decoded using sentencepiece model
131
+ if token in self.all_special_tokens:
132
+ if not prev_is_special:
133
+ out_string += ' '
134
+ out_string += self.sp_model.decode(current_sub_tokens) + token
135
+ prev_is_special = True
136
+ current_sub_tokens = []
137
+ else:
138
+ current_sub_tokens.append(token)
139
+ prev_is_special = False
140
+ out_string += self.sp_model.decode(current_sub_tokens)
141
+ out_string = self.clean_up_tokenization(out_string)
142
+ out_string = self._maybe_add_prefix_space(tokens=tokens, decoded=out_string)
143
+ return out_string[1:]
144
+
145
+ def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None) -> Tuple[str]:
146
+ """
147
+ Save the vocabulary and special tokens file to a directory.
148
+
149
+ Args:
150
+ save_directory (`str`):
151
+ The directory in which to save the vocabulary.
152
+
153
+ Returns:
154
+ `Tuple(str)`: Paths to the files saved.
155
+ """
156
+ if not os.path.isdir(save_directory):
157
+ logger.error(f'Vocabulary path ({save_directory}) should be a directory')
158
+ return
159
+ out_vocab_file = os.path.join(
160
+ save_directory, (filename_prefix + '-' if filename_prefix else '') + VOCAB_FILES_NAMES['vocab_file']
161
+ )
162
+
163
+ if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
164
+ copyfile(self.vocab_file, out_vocab_file)
165
+ elif not os.path.isfile(self.vocab_file):
166
+ with open(out_vocab_file, 'wb') as fi:
167
+ content_spiece_model = self.sp_model.serialized_model_proto()
168
+ fi.write(content_spiece_model)
169
+
170
+ return (out_vocab_file,)
171
+
172
+ def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
173
+ if self.add_bos_token:
174
+ bos_token_ids = [self.bos_token_id]
175
+ else:
176
+ bos_token_ids = []
177
+
178
+ output = bos_token_ids + token_ids_0
179
+
180
+ if token_ids_1 is not None:
181
+ output = output + token_ids_1
182
+
183
+ if self.add_eos_token:
184
+ output = output + [self.eos_token_id]
185
+
186
+ return output
187
+
188
+ def get_special_tokens_mask(
189
+ self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
190
+ ) -> List[int]:
191
+ """
192
+ Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
193
+ special tokens using the tokenizer `prepare_for_model` method.
194
+
195
+ Args:
196
+ token_ids_0 (`List[int]`):
197
+ List of IDs.
198
+ token_ids_1 (`List[int]`, *optional*):
199
+ Optional second list of IDs for sequence pairs.
200
+ already_has_special_tokens (`bool`, *optional*, defaults to `False`):
201
+ Whether or not the token list is already formatted with special tokens for the model.
202
+
203
+ Returns:
204
+ `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
205
+ """
206
+ if already_has_special_tokens:
207
+ return super().get_special_tokens_mask(
208
+ token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
209
+ )
210
+
211
+ if token_ids_1 is None:
212
+ return [1] + ([0] * len(token_ids_0)) + [1]
213
+ return [1] + ([0] * len(token_ids_0)) + [1, 1] + ([0] * len(token_ids_1)) + [1]
214
+
215
+ def create_token_type_ids_from_sequences(
216
+ self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
217
+ ) -> List[int]:
218
+ """
219
+ Create a mask from the two sequences passed to be used in a sequence-pair classification task. T5 does not make
220
+ use of token type ids, therefore a list of zeros is returned.
221
+
222
+ Args:
223
+ token_ids_0 (`List[int]`):
224
+ List of IDs.
225
+ token_ids_1 (`List[int]`, *optional*):
226
+ Optional second list of IDs for sequence pairs.
227
+
228
+ Returns:
229
+ `List[int]`: List of zeros.
230
+ """
231
+ eos = [self.eos_token_id]
232
+
233
+ if token_ids_1 is None:
234
+ return len(token_ids_0 + eos) * [0]
235
+ return len(token_ids_0 + eos + token_ids_1 + eos) * [0]
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f868398fc4e05ee1e8aeba95ddf18ddcc45b8bce55d5093bead5bbf80429b48b
3
+ size 1477754
tokenizer_config.json ADDED
@@ -0,0 +1,180 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "<unk>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<s>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "2": {
20
+ "content": "</s>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "92538": {
28
+ "content": "<|plugin|>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ },
35
+ "92539": {
36
+ "content": "<|interpreter|>",
37
+ "lstrip": false,
38
+ "normalized": false,
39
+ "rstrip": false,
40
+ "single_word": false,
41
+ "special": true
42
+ },
43
+ "92540": {
44
+ "content": "<|action_end|>",
45
+ "lstrip": false,
46
+ "normalized": false,
47
+ "rstrip": false,
48
+ "single_word": false,
49
+ "special": true
50
+ },
51
+ "92541": {
52
+ "content": "<|action_start|>",
53
+ "lstrip": false,
54
+ "normalized": false,
55
+ "rstrip": false,
56
+ "single_word": false,
57
+ "special": true
58
+ },
59
+ "92542": {
60
+ "content": "<|im_end|>",
61
+ "lstrip": false,
62
+ "normalized": false,
63
+ "rstrip": false,
64
+ "single_word": false,
65
+ "special": true
66
+ },
67
+ "92543": {
68
+ "content": "<|im_start|>",
69
+ "lstrip": false,
70
+ "normalized": false,
71
+ "rstrip": false,
72
+ "single_word": false,
73
+ "special": true
74
+ },
75
+ "92544": {
76
+ "content": "<img>",
77
+ "lstrip": false,
78
+ "normalized": false,
79
+ "rstrip": false,
80
+ "single_word": false,
81
+ "special": true
82
+ },
83
+ "92545": {
84
+ "content": "</img>",
85
+ "lstrip": false,
86
+ "normalized": false,
87
+ "rstrip": false,
88
+ "single_word": false,
89
+ "special": true
90
+ },
91
+ "92546": {
92
+ "content": "<IMG_CONTEXT>",
93
+ "lstrip": false,
94
+ "normalized": false,
95
+ "rstrip": false,
96
+ "single_word": false,
97
+ "special": true
98
+ },
99
+ "92547": {
100
+ "content": "<quad>",
101
+ "lstrip": false,
102
+ "normalized": false,
103
+ "rstrip": false,
104
+ "single_word": false,
105
+ "special": true
106
+ },
107
+ "92548": {
108
+ "content": "</quad>",
109
+ "lstrip": false,
110
+ "normalized": false,
111
+ "rstrip": false,
112
+ "single_word": false,
113
+ "special": true
114
+ },
115
+ "92549": {
116
+ "content": "<ref>",
117
+ "lstrip": false,
118
+ "normalized": false,
119
+ "rstrip": false,
120
+ "single_word": false,
121
+ "special": true
122
+ },
123
+ "92550": {
124
+ "content": "</ref>",
125
+ "lstrip": false,
126
+ "normalized": false,
127
+ "rstrip": false,
128
+ "single_word": false,
129
+ "special": true
130
+ },
131
+ "92551": {
132
+ "content": "<box>",
133
+ "lstrip": false,
134
+ "normalized": false,
135
+ "rstrip": false,
136
+ "single_word": false,
137
+ "special": true
138
+ },
139
+ "92552": {
140
+ "content": "</box>",
141
+ "lstrip": false,
142
+ "normalized": false,
143
+ "rstrip": false,
144
+ "single_word": false,
145
+ "special": true
146
+ }
147
+ },
148
+ "additional_special_tokens": [
149
+ "<|im_start|>",
150
+ "<|im_end|>",
151
+ "<|action_start|>",
152
+ "<|action_end|>",
153
+ "<|interpreter|>",
154
+ "<|plugin|>",
155
+ "<img>",
156
+ "</img>",
157
+ "<IMG_CONTEXT>",
158
+ "<quad>",
159
+ "</quad>",
160
+ "<ref>",
161
+ "</ref>",
162
+ "<box>",
163
+ "</box>"
164
+ ],
165
+ "auto_map": {
166
+ "AutoTokenizer": [
167
+ "tokenization_internlm2.InternLM2Tokenizer",
168
+ null
169
+ ]
170
+ },
171
+ "bos_token": "<s>",
172
+ "chat_template": "{{ bos_token }}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
173
+ "clean_up_tokenization_spaces": false,
174
+ "eos_token": "</s>",
175
+ "extra_special_tokens": {},
176
+ "model_max_length": 8192,
177
+ "pad_token": "</s>",
178
+ "tokenizer_class": "InternLM2Tokenizer",
179
+ "unk_token": "<unk>"
180
+ }