Hugging Face's logo

Wondernutts
/
Ternary-Bonsai-27B-int4-sym-ov

Image-Text-to-Text
OpenVINO
qwen3_5
int4
intel-arc
conversational
bonsai
ternary
Model card Files
xet
 Community
Wondernutts commited on
Commit
30b3750
·
verified ·
1 Parent(s): 75cd493

Upload folder using huggingface_hub

Browse files
Files changed (21) hide show
  1. .gitattributes +1 -0
  2. chat_template.jinja +154 -0
  3. config.json +142 -0
  4. generation_config.json +13 -0
  5. openvino_config.json +44 -0
  6. openvino_detokenizer.bin +3 -0
  7. openvino_detokenizer.xml +222 -0
  8. openvino_language_model.bin +3 -0
  9. openvino_language_model.xml +0 -0
  10. openvino_text_embeddings_model.bin +3 -0
  11. openvino_text_embeddings_model.xml +180 -0
  12. openvino_tokenizer.bin +3 -0
  13. openvino_tokenizer.xml +773 -0
  14. openvino_vision_embeddings_merger_model.bin +3 -0
  15. openvino_vision_embeddings_merger_model.xml +0 -0
  16. openvino_vision_embeddings_model.bin +3 -0
  17. openvino_vision_embeddings_model.xml +291 -0
  18. openvino_vision_embeddings_pos_model.bin +3 -0
  19. openvino_vision_embeddings_pos_model.xml +180 -0
  20. tokenizer.json +3 -0
  21. tokenizer_config.json +31 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is false %}
150
+ {{- '<think>\n\n</think>\n\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3_5ForConditionalGeneration"
4
+ ],
5
+ "dtype": "bfloat16",
6
+ "image_token_id": 248056,
7
+ "language_model_only": false,
8
+ "model_type": "qwen3_5",
9
+ "text_config": {
10
+ "attention_bias": false,
11
+ "attention_dropout": 0.0,
12
+ "attn_output_gate": true,
13
+ "bos_token_id": 248044,
14
+ "dtype": "bfloat16",
15
+ "eos_token_id": 248044,
16
+ "full_attention_interval": 4,
17
+ "head_dim": 256,
18
+ "hidden_act": "silu",
19
+ "hidden_size": 5120,
20
+ "initializer_range": 0.02,
21
+ "intermediate_size": 17408,
22
+ "layer_types": [
23
+ "linear_attention",
24
+ "linear_attention",
25
+ "linear_attention",
26
+ "full_attention",
27
+ "linear_attention",
28
+ "linear_attention",
29
+ "linear_attention",
30
+ "full_attention",
31
+ "linear_attention",
32
+ "linear_attention",
33
+ "linear_attention",
34
+ "full_attention",
35
+ "linear_attention",
36
+ "linear_attention",
37
+ "linear_attention",
38
+ "full_attention",
39
+ "linear_attention",
40
+ "linear_attention",
41
+ "linear_attention",
42
+ "full_attention",
43
+ "linear_attention",
44
+ "linear_attention",
45
+ "linear_attention",
46
+ "full_attention",
47
+ "linear_attention",
48
+ "linear_attention",
49
+ "linear_attention",
50
+ "full_attention",
51
+ "linear_attention",
52
+ "linear_attention",
53
+ "linear_attention",
54
+ "full_attention",
55
+ "linear_attention",
56
+ "linear_attention",
57
+ "linear_attention",
58
+ "full_attention",
59
+ "linear_attention",
60
+ "linear_attention",
61
+ "linear_attention",
62
+ "full_attention",
63
+ "linear_attention",
64
+ "linear_attention",
65
+ "linear_attention",
66
+ "full_attention",
67
+ "linear_attention",
68
+ "linear_attention",
69
+ "linear_attention",
70
+ "full_attention",
71
+ "linear_attention",
72
+ "linear_attention",
73
+ "linear_attention",
74
+ "full_attention",
75
+ "linear_attention",
76
+ "linear_attention",
77
+ "linear_attention",
78
+ "full_attention",
79
+ "linear_attention",
80
+ "linear_attention",
81
+ "linear_attention",
82
+ "full_attention",
83
+ "linear_attention",
84
+ "linear_attention",
85
+ "linear_attention",
86
+ "full_attention"
87
+ ],
88
+ "linear_conv_kernel_dim": 4,
89
+ "linear_key_head_dim": 128,
90
+ "linear_num_key_heads": 16,
91
+ "linear_num_value_heads": 48,
92
+ "linear_value_head_dim": 128,
93
+ "mamba_ssm_dtype": "float32",
94
+ "max_position_embeddings": 262144,
95
+ "model_type": "qwen3_5_text",
96
+ "mtp_num_hidden_layers": 1,
97
+ "mtp_use_dedicated_embeddings": false,
98
+ "num_attention_heads": 24,
99
+ "num_hidden_layers": 64,
100
+ "num_key_value_heads": 4,
101
+ "output_gate_type": "swish",
102
+ "pad_token_id": null,
103
+ "partial_rotary_factor": 0.25,
104
+ "rms_norm_eps": 1e-06,
105
+ "rope_parameters": {
106
+ "mrope_interleaved": true,
107
+ "mrope_section": [
108
+ 11,
109
+ 11,
110
+ 10
111
+ ],
112
+ "partial_rotary_factor": 0.25,
113
+ "rope_theta": 10000000,
114
+ "rope_type": "default"
115
+ },
116
+ "tie_word_embeddings": false,
117
+ "use_cache": true,
118
+ "vocab_size": 248320
119
+ },
120
+ "tie_word_embeddings": false,
121
+ "transformers_version": "5.2.0",
122
+ "video_token_id": 248057,
123
+ "vision_config": {
124
+ "deepstack_visual_indexes": [],
125
+ "depth": 27,
126
+ "dtype": "bfloat16",
127
+ "hidden_act": "gelu_pytorch_tanh",
128
+ "hidden_size": 1152,
129
+ "in_channels": 3,
130
+ "initializer_range": 0.02,
131
+ "intermediate_size": 4304,
132
+ "model_type": "qwen3_5",
133
+ "num_heads": 16,
134
+ "num_position_embeddings": 2304,
135
+ "out_hidden_size": 5120,
136
+ "patch_size": 16,
137
+ "spatial_merge_size": 2,
138
+ "temporal_patch_size": 2
139
+ },
140
+ "vision_end_token_id": 248054,
141
+ "vision_start_token_id": 248053
142
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 248044,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 248046,
6
+ 248044
7
+ ],
8
+ "pad_token_id": 248044,
9
+ "temperature": 1.0,
10
+ "top_k": 20,
11
+ "top_p": 0.95,
12
+ "transformers_version": "5.2.0"
13
+ }
openvino_config.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dtype": "int4",
3
+ "input_info": null,
4
+ "optimum_version": "2.2.0.dev0",
5
+ "output_attentions": false,
6
+ "quantization_config": {
7
+ "_dataset_kwargs": {},
8
+ "dataset": null,
9
+ "default_config": {
10
+ "quant_method": "default"
11
+ },
12
+ "ignored_scope": null,
13
+ "num_samples": null,
14
+ "processor": "/mnt/local-scratch/work/src_model/",
15
+ "quantization_configs": {
16
+ "lm_model": {
17
+ "_dataset_kwargs": {},
18
+ "all_layers": true,
19
+ "backup_precision": "int8_sym",
20
+ "bits": 4,
21
+ "dataset": null,
22
+ "dq_group_size": null,
23
+ "dtype": "int4",
24
+ "gptq": null,
25
+ "group_size": 128,
26
+ "group_size_fallback": null,
27
+ "ignored_scope": null,
28
+ "lora_correction": null,
29
+ "num_samples": null,
30
+ "processor": "/mnt/local-scratch/work/src_model/",
31
+ "quant_method": "default",
32
+ "ratio": 1.0,
33
+ "scale_estimation": null,
34
+ "sensitivity_metric": null,
35
+ "statistics_path": null,
36
+ "sym": true,
37
+ "tokenizer": "/mnt/local-scratch/work/src_model/"
38
+ }
39
+ },
40
+ "tokenizer": "/mnt/local-scratch/work/src_model/"
41
+ },
42
+ "save_onnx_model": false,
43
+ "transformers_version": "5.2.0"
44
+ }
openvino_detokenizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48fc2c6a7a1056994982911eb9695826acef933c3d742364c56d16317a0b2a79
3
+ size 3828184
openvino_detokenizer.xml ADDED
@@ -0,0 +1,222 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="detokenizer" version="11">
3
+ <layers>
4
+ <layer id="0" name="Parameter_4168748" type="Parameter" version="opset1">
5
+ <data shape="?,?" element_type="i64" />
6
+ <output>
7
+ <port id="0" precision="I64" names="Parameter_4168748">
8
+ <dim>-1</dim>
9
+ <dim>-1</dim>
10
+ </port>
11
+ </output>
12
+ </layer>
13
+ <layer id="1" name="Convert_4168783" 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_4168750" type="Const" version="opset1">
29
+ <data element_type="i32" shape="248077" offset="0" size="992308" />
30
+ <output>
31
+ <port id="0" precision="I32">
32
+ <dim>248077</dim>
33
+ </port>
34
+ </output>
35
+ </layer>
36
+ <layer id="3" name="Constant_4168752" type="Const" version="opset1">
37
+ <data element_type="i32" shape="248077" offset="992308" size="992308" />
38
+ <output>
39
+ <port id="0" precision="I32">
40
+ <dim>248077</dim>
41
+ </port>
42
+ </output>
43
+ </layer>
44
+ <layer id="4" name="Constant_4168754" type="Const" version="opset1">
45
+ <data element_type="u8" shape="1843484" offset="1984616" size="1843484" />
46
+ <output>
47
+ <port id="0" precision="U8">
48
+ <dim>1843484</dim>
49
+ </port>
50
+ </output>
51
+ </layer>
52
+ <layer id="5" name="Slice_4168759" type="Const" version="opset1">
53
+ <data element_type="i32" shape="21" offset="3828100" size="84" />
54
+ <output>
55
+ <port id="0" precision="I32">
56
+ <dim>21</dim>
57
+ </port>
58
+ </output>
59
+ </layer>
60
+ <layer id="6" name="VocabDecoder_4168761" type="VocabDecoder" version="extension">
61
+ <data skip_tokens="" />
62
+ <input>
63
+ <port id="0" precision="I32">
64
+ <dim>-1</dim>
65
+ <dim>-1</dim>
66
+ </port>
67
+ <port id="1" precision="I32">
68
+ <dim>248077</dim>
69
+ </port>
70
+ <port id="2" precision="I32">
71
+ <dim>248077</dim>
72
+ </port>
73
+ <port id="3" precision="U8">
74
+ <dim>1843484</dim>
75
+ </port>
76
+ <port id="4" precision="I32">
77
+ <dim>21</dim>
78
+ </port>
79
+ </input>
80
+ <output>
81
+ <port id="5" precision="I32">
82
+ <dim>-1</dim>
83
+ </port>
84
+ <port id="6" precision="I32">
85
+ <dim>-1</dim>
86
+ </port>
87
+ <port id="7" precision="I32">
88
+ <dim>-1</dim>
89
+ </port>
90
+ <port id="8" precision="I32">
91
+ <dim>-1</dim>
92
+ </port>
93
+ <port id="9" precision="U8">
94
+ <dim>-1</dim>
95
+ </port>
96
+ </output>
97
+ </layer>
98
+ <layer id="7" name="FuzeRagged_4168762" type="FuzeRagged" version="extension">
99
+ <input>
100
+ <port id="0" precision="I32">
101
+ <dim>-1</dim>
102
+ </port>
103
+ <port id="1" precision="I32">
104
+ <dim>-1</dim>
105
+ </port>
106
+ <port id="2" precision="I32">
107
+ <dim>-1</dim>
108
+ </port>
109
+ <port id="3" precision="I32">
110
+ <dim>-1</dim>
111
+ </port>
112
+ </input>
113
+ <output>
114
+ <port id="4" precision="I32">
115
+ <dim>-1</dim>
116
+ </port>
117
+ <port id="5" precision="I32">
118
+ <dim>-1</dim>
119
+ </port>
120
+ </output>
121
+ </layer>
122
+ <layer id="8" name="UTF8Validate_4168763" type="UTF8Validate" version="extension">
123
+ <data replace_mode="true" />
124
+ <input>
125
+ <port id="0" precision="I32">
126
+ <dim>-1</dim>
127
+ </port>
128
+ <port id="1" precision="I32">
129
+ <dim>-1</dim>
130
+ </port>
131
+ <port id="2" precision="U8">
132
+ <dim>-1</dim>
133
+ </port>
134
+ </input>
135
+ <output>
136
+ <port id="3" precision="I32">
137
+ <dim>-1</dim>
138
+ </port>
139
+ <port id="4" precision="I32">
140
+ <dim>-1</dim>
141
+ </port>
142
+ <port id="5" precision="U8">
143
+ <dim>-1</dim>
144
+ </port>
145
+ </output>
146
+ </layer>
147
+ <layer id="9" name="StringTensorPack_4168764" type="StringTensorPack" version="opset15">
148
+ <input>
149
+ <port id="0" precision="I32">
150
+ <dim>-1</dim>
151
+ </port>
152
+ <port id="1" precision="I32">
153
+ <dim>-1</dim>
154
+ </port>
155
+ <port id="2" precision="U8">
156
+ <dim>-1</dim>
157
+ </port>
158
+ </input>
159
+ <output>
160
+ <port id="3" precision="STRING" names="Result_4168765,string_output">
161
+ <dim>-1</dim>
162
+ </port>
163
+ </output>
164
+ </layer>
165
+ <layer id="10" name="Result_4168765" type="Result" version="opset1" output_names="Result_4168765,string_output">
166
+ <input>
167
+ <port id="0" precision="STRING">
168
+ <dim>-1</dim>
169
+ </port>
170
+ </input>
171
+ </layer>
172
+ </layers>
173
+ <edges>
174
+ <edge from-layer="0" from-port="0" to-layer="1" to-port="0" />
175
+ <edge from-layer="1" from-port="1" to-layer="6" to-port="0" />
176
+ <edge from-layer="2" from-port="0" to-layer="6" to-port="1" />
177
+ <edge from-layer="3" from-port="0" to-layer="6" to-port="2" />
178
+ <edge from-layer="4" from-port="0" to-layer="6" to-port="3" />
179
+ <edge from-layer="5" from-port="0" to-layer="6" to-port="4" />
180
+ <edge from-layer="6" from-port="7" to-layer="7" to-port="2" />
181
+ <edge from-layer="6" from-port="9" to-layer="8" to-port="2" />
182
+ <edge from-layer="6" from-port="8" to-layer="7" to-port="3" />
183
+ <edge from-layer="6" from-port="6" to-layer="7" to-port="1" />
184
+ <edge from-layer="6" from-port="5" to-layer="7" to-port="0" />
185
+ <edge from-layer="7" from-port="4" to-layer="8" to-port="0" />
186
+ <edge from-layer="7" from-port="5" to-layer="8" to-port="1" />
187
+ <edge from-layer="8" from-port="3" to-layer="9" to-port="0" />
188
+ <edge from-layer="8" from-port="4" to-layer="9" to-port="1" />
189
+ <edge from-layer="8" from-port="5" to-layer="9" to-port="2" />
190
+ <edge from-layer="9" from-port="3" to-layer="10" to-port="0" />
191
+ </edges>
192
+ <rt_info>
193
+ <info name="OpenVINO Runtime" value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
194
+ <add_attention_mask value="True" />
195
+ <add_prefix_space />
196
+ <add_special_tokens value="True" />
197
+ <chat_template value="{%- set image_count = namespace(value=0) %}&#10;{%- set video_count = namespace(value=0) %}&#10;{%- macro render_content(content, do_vision_count, is_system_content=false) %}&#10; {%- if content is string %}&#10; {{- content }}&#10; {%- elif content is iterable and content is not mapping %}&#10; {%- for item in content %}&#10; {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}&#10; {%- if is_system_content %}&#10; {{- raise_exception('System message cannot contain images.') }}&#10; {%- endif %}&#10; {%- if do_vision_count %}&#10; {%- set image_count.value = image_count.value + 1 %}&#10; {%- endif %}&#10; {%- if add_vision_id %}&#10; {{- 'Picture ' ~ image_count.value ~ ': ' }}&#10; {%- endif %}&#10; {{- '&lt;|vision_start|>&lt;|image_pad|>&lt;|vision_end|>' }}&#10; {%- elif 'video' in item or item.type == 'video' %}&#10; {%- if is_system_content %}&#10; {{- raise_exception('System message cannot contain videos.') }}&#10; {%- endif %}&#10; {%- if do_vision_count %}&#10; {%- set video_count.value = video_count.value + 1 %}&#10; {%- endif %}&#10; {%- if add_vision_id %}&#10; {{- 'Video ' ~ video_count.value ~ ': ' }}&#10; {%- endif %}&#10; {{- '&lt;|vision_start|>&lt;|video_pad|>&lt;|vision_end|>' }}&#10; {%- elif 'text' in item %}&#10; {{- item.text }}&#10; {%- else %}&#10; {{- raise_exception('Unexpected item type in content.') }}&#10; {%- endif %}&#10; {%- endfor %}&#10; {%- elif content is none or content is undefined %}&#10; {{- '' }}&#10; {%- else %}&#10; {{- raise_exception('Unexpected content type.') }}&#10; {%- endif %}&#10;{%- endmacro %}&#10;{%- if not messages %}&#10; {{- raise_exception('No messages provided.') }}&#10;{%- endif %}&#10;{%- if tools and tools is iterable and tools is not mapping %}&#10; {{- '&lt;|im_start|>system\n' }}&#10; {{- &quot;# Tools\n\nYou have access to the following functions:\n\n&lt;tools>&quot; }}&#10; {%- for tool in tools %}&#10; {{- &quot;\n&quot; }}&#10; {{- tool | tojson }}&#10; {%- endfor %}&#10; {{- &quot;\n&lt;/tools>&quot; }}&#10; {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n&lt;tool_call>\n&lt;function=example_function_name>\n&lt;parameter=example_parameter_1>\nvalue_1\n&lt;/parameter>\n&lt;parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n&lt;/parameter>\n&lt;/function>\n&lt;/tool_call>\n\n&lt;IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner &lt;function=...>&lt;/function> block must be nested within &lt;tool_call>&lt;/tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n&lt;/IMPORTANT>' }}&#10; {%- if messages[0].role == 'system' %}&#10; {%- set content = render_content(messages[0].content, false, true)|trim %}&#10; {%- if content %}&#10; {{- '\n\n' + content }}&#10; {%- endif %}&#10; {%- endif %}&#10; {{- '&lt;|im_end|>\n' }}&#10;{%- else %}&#10; {%- if messages[0].role == 'system' %}&#10; {%- set content = render_content(messages[0].content, false, true)|trim %}&#10; {{- '&lt;|im_start|>system\n' + content + '&lt;|im_end|>\n' }}&#10; {%- endif %}&#10;{%- endif %}&#10;{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}&#10;{%- for message in messages[::-1] %}&#10; {%- set index = (messages|length - 1) - loop.index0 %}&#10; {%- if ns.multi_step_tool and message.role == &quot;user&quot; %}&#10; {%- set content = render_content(message.content, false)|trim %}&#10; {%- if not(content.startswith('&lt;tool_response>') and content.endswith('&lt;/tool_response>')) %}&#10; {%- set ns.multi_step_tool = false %}&#10; {%- set ns.last_query_index = index %}&#10; {%- endif %}&#10; {%- endif %}&#10;{%- endfor %}&#10;{%- if ns.multi_step_tool %}&#10; {{- raise_exception('No user query found in messages.') }}&#10;{%- endif %}&#10;{%- for message in messages %}&#10; {%- set content = render_content(message.content, true)|trim %}&#10; {%- if message.role == &quot;system&quot; %}&#10; {%- if not loop.first %}&#10; {{- raise_exception('System message must be at the beginning.') }}&#10; {%- endif %}&#10; {%- elif message.role == &quot;user&quot; %}&#10; {{- '&lt;|im_start|>' + message.role + '\n' + content + '&lt;|im_end|>' + '\n' }}&#10; {%- elif message.role == &quot;assistant&quot; %}&#10; {%- set reasoning_content = '' %}&#10; {%- if message.reasoning_content is string %}&#10; {%- set reasoning_content = message.reasoning_content %}&#10; {%- else %}&#10; {%- if '&lt;/think>' in content %}&#10; {%- set reasoning_content = content.split('&lt;/think>')[0].rstrip('\n').split('&lt;think>')[-1].lstrip('\n') %}&#10; {%- set content = content.split('&lt;/think>')[-1].lstrip('\n') %}&#10; {%- endif %}&#10; {%- endif %}&#10; {%- set reasoning_content = reasoning_content|trim %}&#10; {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}&#10; {{- '&lt;|im_start|>' + message.role + '\n&lt;think>\n' + reasoning_content + '\n&lt;/think>\n\n' + content }}&#10; {%- else %}&#10; {{- '&lt;|im_start|>' + message.role + '\n' + content }}&#10; {%- endif %}&#10; {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}&#10; {%- for tool_call in message.tool_calls %}&#10; {%- if tool_call.function is defined %}&#10; {%- set tool_call = tool_call.function %}&#10; {%- endif %}&#10; {%- if loop.first %}&#10; {%- if content|trim %}&#10; {{- '\n\n&lt;tool_call>\n&lt;function=' + tool_call.name + '>\n' }}&#10; {%- else %}&#10; {{- '&lt;tool_call>\n&lt;function=' + tool_call.name + '>\n' }}&#10; {%- endif %}&#10; {%- else %}&#10; {{- '\n&lt;tool_call>\n&lt;function=' + tool_call.name + '>\n' }}&#10; {%- endif %}&#10; {%- if tool_call.arguments is defined %}&#10; {%- for args_name, args_value in tool_call.arguments|items %}&#10; {{- '&lt;parameter=' + args_name + '>\n' }}&#10; {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}&#10; {{- args_value }}&#10; {{- '\n&lt;/parameter>\n' }}&#10; {%- endfor %}&#10; {%- endif %}&#10; {{- '&lt;/function>\n&lt;/tool_call>' }}&#10; {%- endfor %}&#10; {%- endif %}&#10; {{- '&lt;|im_end|>\n' }}&#10; {%- elif message.role == &quot;tool&quot; %}&#10; {%- if loop.previtem and loop.previtem.role != &quot;tool&quot; %}&#10; {{- '&lt;|im_start|>user' }}&#10; {%- endif %}&#10; {{- '\n&lt;tool_response>\n' }}&#10; {{- content }}&#10; {{- '\n&lt;/tool_response>' }}&#10; {%- if not loop.last and loop.nextitem.role != &quot;tool&quot; %}&#10; {{- '&lt;|im_end|>\n' }}&#10; {%- elif loop.last %}&#10; {{- '&lt;|im_end|>\n' }}&#10; {%- endif %}&#10; {%- else %}&#10; {{- raise_exception('Unexpected message role.') }}&#10; {%- endif %}&#10;{%- endfor %}&#10;{%- if add_generation_prompt %}&#10; {{- '&lt;|im_start|>assistant\n' }}&#10; {%- if enable_thinking is defined and enable_thinking is false %}&#10; {{- '&lt;think>\n\n&lt;/think>\n\n' }}&#10; {%- else %}&#10; {{- '&lt;think>\n' }}&#10; {%- endif %}&#10;{%- endif %}" />
198
+ <clean_up_tokenization_spaces />
199
+ <detokenizer_input_type value="i64" />
200
+ <eos_token_id value="248046" />
201
+ <handle_special_tokens_with_re />
202
+ <max_length />
203
+ <number_of_inputs value="1" />
204
+ <openvino_tokenizers_version value="2026.2.1.0-682-6da74a793f0" />
205
+ <openvino_version value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
206
+ <original_post_processor_template value="{&quot;type&quot;: &quot;ByteLevel&quot;, &quot;add_prefix_space&quot;: false, &quot;trim_offsets&quot;: false, &quot;use_regex&quot;: false}" />
207
+ <original_tokenizer_class value="&lt;class 'transformers.tokenization_utils_tokenizers.TokenizersBackend'>" />
208
+ <pad_token_id value="248044" />
209
+ <processed_post_processor_template value="{&quot;single&quot;: {&quot;ids&quot;: [-1], &quot;type_ids&quot;: [0]}, &quot;pair&quot;: {&quot;ids&quot;: [-1, -2], &quot;type_ids&quot;: [0, 0]}}" />
210
+ <sentencepiece_version value="0.2.2" />
211
+ <skip_special_tokens value="True" />
212
+ <streaming_detokenizer value="False" />
213
+ <tiktoken_version value="0.13.0" />
214
+ <tokenizer_output_type value="i64" />
215
+ <tokenizers_version value="0.22.2" />
216
+ <transformers_version value="5.2.0" />
217
+ <use_max_padding value="False" />
218
+ <use_sentencepiece_backend value="False" />
219
+ <utf8_replace_mode value="replace" />
220
+ <with_detokenizer value="True" />
221
+ </rt_info>
222
+ </net>
openvino_language_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ab655e335a0410c51f5658e94da62e5b6b88718f16a95cb38046b26fb153d8b8
3
+ size 13216987954
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:a398f6cbd7efdbe66c0ef95f4c500df67cd4fec8465646013afa958748083f9e
3
+ size 1271895044
openvino_text_embeddings_model.xml ADDED
@@ -0,0 +1,180 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="Model105" 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="i8" shape="248320, 5120" offset="0" size="1271398400" />
15
+ <output>
16
+ <port id="0" precision="I8">
17
+ <dim>248320</dim>
18
+ <dim>5120</dim>
19
+ </port>
20
+ </output>
21
+ </layer>
22
+ <layer id="2" name="Convert_3364733" type="Convert" version="opset1">
23
+ <data destination_type="f16" />
24
+ <input>
25
+ <port id="0" precision="I8">
26
+ <dim>248320</dim>
27
+ <dim>5120</dim>
28
+ </port>
29
+ </input>
30
+ <output>
31
+ <port id="1" precision="FP16">
32
+ <dim>248320</dim>
33
+ <dim>5120</dim>
34
+ </port>
35
+ </output>
36
+ </layer>
37
+ <layer id="3" name="self.weight/scale" type="Const" version="opset1">
38
+ <data element_type="f16" shape="248320, 1" offset="1271398400" size="496640" />
39
+ <output>
40
+ <port id="0" precision="FP16">
41
+ <dim>248320</dim>
42
+ <dim>1</dim>
43
+ </port>
44
+ </output>
45
+ </layer>
46
+ <layer id="4" name="self.weight/fq_weights_0" type="Multiply" version="opset1">
47
+ <data auto_broadcast="numpy" />
48
+ <input>
49
+ <port id="0" precision="FP16">
50
+ <dim>248320</dim>
51
+ <dim>5120</dim>
52
+ </port>
53
+ <port id="1" precision="FP16">
54
+ <dim>248320</dim>
55
+ <dim>1</dim>
56
+ </port>
57
+ </input>
58
+ <output>
59
+ <port id="2" precision="FP16">
60
+ <dim>248320</dim>
61
+ <dim>5120</dim>
62
+ </port>
63
+ </output>
64
+ </layer>
65
+ <layer id="5" name="ov_ext::embedding/Convert" type="Convert" version="opset1">
66
+ <data destination_type="f32" />
67
+ <rt_info>
68
+ <attribute name="decompression" version="0" />
69
+ </rt_info>
70
+ <input>
71
+ <port id="0" precision="FP16">
72
+ <dim>248320</dim>
73
+ <dim>5120</dim>
74
+ </port>
75
+ </input>
76
+ <output>
77
+ <port id="1" precision="FP32">
78
+ <dim>248320</dim>
79
+ <dim>5120</dim>
80
+ </port>
81
+ </output>
82
+ </layer>
83
+ <layer id="6" name="ov_ext::embedding/Convert_1" type="Convert" version="opset1">
84
+ <data destination_type="i32" />
85
+ <input>
86
+ <port id="0" precision="I64">
87
+ <dim>-1</dim>
88
+ <dim>-1</dim>
89
+ </port>
90
+ </input>
91
+ <output>
92
+ <port id="1" precision="I32">
93
+ <dim>-1</dim>
94
+ <dim>-1</dim>
95
+ </port>
96
+ </output>
97
+ </layer>
98
+ <layer id="7" name="ov_ext::embedding/Constant" type="Const" version="opset1">
99
+ <data element_type="i32" shape="" offset="1271895040" size="4" />
100
+ <output>
101
+ <port id="0" precision="I32" />
102
+ </output>
103
+ </layer>
104
+ <layer id="8" name="ov_ext::embedding/Gather" type="Gather" version="opset8">
105
+ <data batch_dims="0" />
106
+ <input>
107
+ <port id="0" precision="FP32">
108
+ <dim>248320</dim>
109
+ <dim>5120</dim>
110
+ </port>
111
+ <port id="1" precision="I32">
112
+ <dim>-1</dim>
113
+ <dim>-1</dim>
114
+ </port>
115
+ <port id="2" precision="I32" />
116
+ </input>
117
+ <output>
118
+ <port id="3" precision="FP32" names="inputs_embeds">
119
+ <dim>-1</dim>
120
+ <dim>-1</dim>
121
+ <dim>5120</dim>
122
+ </port>
123
+ </output>
124
+ </layer>
125
+ <layer id="9" name="Result_454912" type="Result" version="opset1" output_names="inputs_embeds">
126
+ <input>
127
+ <port id="0" precision="FP32">
128
+ <dim>-1</dim>
129
+ <dim>-1</dim>
130
+ <dim>5120</dim>
131
+ </port>
132
+ </input>
133
+ </layer>
134
+ </layers>
135
+ <edges>
136
+ <edge from-layer="0" from-port="0" to-layer="6" to-port="0" />
137
+ <edge from-layer="1" from-port="0" to-layer="2" to-port="0" />
138
+ <edge from-layer="2" from-port="1" to-layer="4" to-port="0" />
139
+ <edge from-layer="3" from-port="0" to-layer="4" to-port="1" />
140
+ <edge from-layer="4" from-port="2" to-layer="5" to-port="0" />
141
+ <edge from-layer="5" from-port="1" to-layer="8" to-port="0" />
142
+ <edge from-layer="6" from-port="1" to-layer="8" to-port="1" />
143
+ <edge from-layer="7" from-port="0" to-layer="8" to-port="2" />
144
+ <edge from-layer="8" from-port="3" to-layer="9" to-port="0" />
145
+ </edges>
146
+ <rt_info>
147
+ <info name="OpenVINO Runtime" value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
148
+ <Runtime_version value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
149
+ <conversion_parameters>
150
+ <framework value="pytorch" />
151
+ <is_python_object value="True" />
152
+ </conversion_parameters>
153
+ <nncf>
154
+ <friendly_names_were_updated value="True" />
155
+ <version value="3.2.0" />
156
+ <weight_compression>
157
+ <advanced_parameters value="{'statistics_path': None, 'lora_adapter_rank': 256, 'group_size_fallback_mode': 'error', 'min_adjusted_group_size': 32, 'awq_params': {'subset_size': 32, 'percent_to_apply': 0.002, 'alpha_min': 0.0, 'alpha_max': 1.0, 'steps': 100, 'prefer_data_aware_scaling': True}, 'scale_estimation_params': {'subset_size': 64, 'initial_steps': 5, 'scale_steps': 5, 'weight_penalty': -1.0}, 'gptq_params': {'damp_percent': 0.1, 'block_size': 128, 'subset_size': 128}, 'lora_correction_params': {'adapter_rank': 8, 'num_iterations': 3, 'apply_regularization': True, 'subset_size': 128, 'use_int8_adapters': True}, 'backend_params': {}, 'codebook': None, 'adaptive_codebook_params': {'value_type': 'f8e4m3', 'across_blocks': False, 'num_elements': 16}}" />
158
+ <all_layers value="False" />
159
+ <awq value="False" />
160
+ <backup_mode value="int8_asym" />
161
+ <compression_format value="dequantize" />
162
+ <gptq value="False" />
163
+ <group_size value="-1" />
164
+ <ignored_scope value="[]" />
165
+ <lora_correction value="False" />
166
+ <mode value="int8_sym" />
167
+ <ratio value="1.0" />
168
+ <scale_estimation value="False" />
169
+ <sensitivity_metric value="weight_quantization_error" />
170
+ </weight_compression>
171
+ </nncf>
172
+ <optimum>
173
+ <nncf_version value="3.2.0" />
174
+ <optimum_intel_version value="2.1.0.dev0+109314c" />
175
+ <optimum_version value="2.2.0.dev0" />
176
+ <pytorch_version value="2.11.0+cu128" />
177
+ <transformers_version value="5.2.0" />
178
+ </optimum>
179
+ </rt_info>
180
+ </net>
openvino_tokenizer.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b67896f9f7b0ad421b9edb1bdd909a8e2bc2059230f48699e8e9585fa8f0e900
3
+ size 9631155
openvino_tokenizer.xml ADDED
@@ -0,0 +1,773 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="tokenizer" version="11">
3
+ <layers>
4
+ <layer id="0" name="Parameter_4168620" type="Parameter" version="opset1">
5
+ <data shape="?" element_type="string" />
6
+ <output>
7
+ <port id="0" precision="STRING" names="Parameter_4168620">
8
+ <dim>-1</dim>
9
+ </port>
10
+ </output>
11
+ </layer>
12
+ <layer id="1" name="Constant_4168626" type="Const" version="opset1">
13
+ <data element_type="i64" shape="" offset="0" size="8" />
14
+ <output>
15
+ <port id="0" precision="I64" />
16
+ </output>
17
+ </layer>
18
+ <layer id="2" name="StringTensorUnpack_4168621" type="StringTensorUnpack" version="opset15">
19
+ <input>
20
+ <port id="0" precision="STRING">
21
+ <dim>-1</dim>
22
+ </port>
23
+ </input>
24
+ <output>
25
+ <port id="1" precision="I32">
26
+ <dim>-1</dim>
27
+ </port>
28
+ <port id="2" precision="I32">
29
+ <dim>-1</dim>
30
+ </port>
31
+ <port id="3" precision="U8">
32
+ <dim>-1</dim>
33
+ </port>
34
+ </output>
35
+ </layer>
36
+ <layer id="3" name="ShapeOf_4168622" type="ShapeOf" version="opset3">
37
+ <data output_type="i64" />
38
+ <input>
39
+ <port id="0" precision="I32">
40
+ <dim>-1</dim>
41
+ </port>
42
+ </input>
43
+ <output>
44
+ <port id="1" precision="I64">
45
+ <dim>1</dim>
46
+ </port>
47
+ </output>
48
+ </layer>
49
+ <layer id="4" name="Constant_4168623" type="Const" version="opset1">
50
+ <data element_type="i64" shape="" offset="0" size="8" />
51
+ <output>
52
+ <port id="0" precision="I64" />
53
+ </output>
54
+ </layer>
55
+ <layer id="5" name="Constant_4168624" type="Const" version="opset1">
56
+ <data element_type="i64" shape="" offset="0" size="8" />
57
+ <output>
58
+ <port id="0" precision="I64" />
59
+ </output>
60
+ </layer>
61
+ <layer id="6" name="Gather_4168625" type="Gather" version="opset8">
62
+ <data batch_dims="0" />
63
+ <input>
64
+ <port id="0" precision="I64">
65
+ <dim>1</dim>
66
+ </port>
67
+ <port id="1" precision="I64" />
68
+ <port id="2" precision="I64" />
69
+ </input>
70
+ <output>
71
+ <port id="3" precision="I64" />
72
+ </output>
73
+ </layer>
74
+ <layer id="7" name="Constant_4168627" type="Const" version="opset1">
75
+ <data element_type="i64" shape="" offset="8" size="8" />
76
+ <output>
77
+ <port id="0" precision="I64" />
78
+ </output>
79
+ </layer>
80
+ <layer id="8" name="Range_4168628" type="Range" version="opset4">
81
+ <data output_type="i32" />
82
+ <input>
83
+ <port id="0" precision="I64" />
84
+ <port id="1" precision="I64" />
85
+ <port id="2" precision="I64" />
86
+ </input>
87
+ <output>
88
+ <port id="3" precision="I32">
89
+ <dim>-1</dim>
90
+ </port>
91
+ </output>
92
+ </layer>
93
+ <layer id="9" name="Constant_4168629" type="Const" version="opset1">
94
+ <data element_type="i64" shape="" offset="8" size="8" />
95
+ <output>
96
+ <port id="0" precision="I64" />
97
+ </output>
98
+ </layer>
99
+ <layer id="10" name="Constant_4168630" type="Const" version="opset1">
100
+ <data element_type="i64" shape="" offset="8" size="8" />
101
+ <output>
102
+ <port id="0" precision="I64" />
103
+ </output>
104
+ </layer>
105
+ <layer id="11" name="Add_4168631" type="Add" version="opset1">
106
+ <data auto_broadcast="numpy" />
107
+ <input>
108
+ <port id="0" precision="I64" />
109
+ <port id="1" precision="I64" />
110
+ </input>
111
+ <output>
112
+ <port id="2" precision="I64" />
113
+ </output>
114
+ </layer>
115
+ <layer id="12" name="Constant_4168632" type="Const" version="opset1">
116
+ <data element_type="i64" shape="" offset="8" size="8" />
117
+ <output>
118
+ <port id="0" precision="I64" />
119
+ </output>
120
+ </layer>
121
+ <layer id="13" name="Range_4168633" type="Range" version="opset4">
122
+ <data output_type="i32" />
123
+ <input>
124
+ <port id="0" precision="I64" />
125
+ <port id="1" precision="I64" />
126
+ <port id="2" precision="I64" />
127
+ </input>
128
+ <output>
129
+ <port id="3" precision="I32">
130
+ <dim>-1</dim>
131
+ </port>
132
+ </output>
133
+ </layer>
134
+ <layer id="14" name="Constant_4168697" type="Const" version="opset1">
135
+ <data element_type="u8" shape="588" offset="16" size="588" />
136
+ <output>
137
+ <port id="0" precision="U8">
138
+ <dim>588</dim>
139
+ </port>
140
+ </output>
141
+ </layer>
142
+ <layer id="15" name="SpecialTokensSplit_4168698" type="SpecialTokensSplit" version="extension">
143
+ <input>
144
+ <port id="0" precision="I32">
145
+ <dim>-1</dim>
146
+ </port>
147
+ <port id="1" precision="I32">
148
+ <dim>-1</dim>
149
+ </port>
150
+ <port id="2" precision="I32">
151
+ <dim>-1</dim>
152
+ </port>
153
+ <port id="3" precision="I32">
154
+ <dim>-1</dim>
155
+ </port>
156
+ <port id="4" precision="U8">
157
+ <dim>-1</dim>
158
+ </port>
159
+ <port id="5" precision="U8">
160
+ <dim>588</dim>
161
+ </port>
162
+ </input>
163
+ <output>
164
+ <port id="6" precision="I32">
165
+ <dim>-1</dim>
166
+ </port>
167
+ <port id="7" precision="I32">
168
+ <dim>-1</dim>
169
+ </port>
170
+ <port id="8" precision="I32">
171
+ <dim>-1</dim>
172
+ </port>
173
+ <port id="9" precision="I32">
174
+ <dim>-1</dim>
175
+ </port>
176
+ <port id="10" precision="U8">
177
+ <dim>-1</dim>
178
+ </port>
179
+ <port id="11" precision="BOOL">
180
+ <dim>-1</dim>
181
+ </port>
182
+ </output>
183
+ </layer>
184
+ <layer id="16" name="CharsMapNormalization_4168699" type="CharsMapNormalization" version="extension">
185
+ <data add_dummy_prefix="false" remove_extra_whitespaces="false" escape_whitespaces="false" normalization_form="nfc" case_fold="false" nmt="false" />
186
+ <input>
187
+ <port id="0" precision="I32">
188
+ <dim>-1</dim>
189
+ </port>
190
+ <port id="1" precision="I32">
191
+ <dim>-1</dim>
192
+ </port>
193
+ <port id="2" precision="U8">
194
+ <dim>-1</dim>
195
+ </port>
196
+ <port id="3" precision="BOOL">
197
+ <dim>-1</dim>
198
+ </port>
199
+ </input>
200
+ <output>
201
+ <port id="4" precision="I32">
202
+ <dim>-1</dim>
203
+ </port>
204
+ <port id="5" precision="I32">
205
+ <dim>-1</dim>
206
+ </port>
207
+ <port id="6" precision="U8">
208
+ <dim>-1</dim>
209
+ </port>
210
+ <port id="7" precision="BOOL">
211
+ <dim>-1</dim>
212
+ </port>
213
+ </output>
214
+ </layer>
215
+ <layer id="17" name="Constant_4168701" type="Const" version="opset1">
216
+ <data element_type="u8" shape="122" offset="604" size="122" />
217
+ <output>
218
+ <port id="0" precision="U8">
219
+ <dim>122</dim>
220
+ </port>
221
+ </output>
222
+ </layer>
223
+ <layer id="18" name="RegexSplit_4168702" type="RegexSplit" version="extension">
224
+ <data behaviour="isolate" invert="false" max_splits="-1" />
225
+ <input>
226
+ <port id="0" precision="I32">
227
+ <dim>-1</dim>
228
+ </port>
229
+ <port id="1" precision="I32">
230
+ <dim>-1</dim>
231
+ </port>
232
+ <port id="2" precision="I32">
233
+ <dim>-1</dim>
234
+ </port>
235
+ <port id="3" precision="I32">
236
+ <dim>-1</dim>
237
+ </port>
238
+ <port id="4" precision="U8">
239
+ <dim>-1</dim>
240
+ </port>
241
+ <port id="5" precision="BOOL">
242
+ <dim>-1</dim>
243
+ </port>
244
+ <port id="6" precision="U8">
245
+ <dim>122</dim>
246
+ </port>
247
+ </input>
248
+ <output>
249
+ <port id="7" precision="I32">
250
+ <dim>-1</dim>
251
+ </port>
252
+ <port id="8" precision="I32">
253
+ <dim>-1</dim>
254
+ </port>
255
+ <port id="9" precision="I32">
256
+ <dim>-1</dim>
257
+ </port>
258
+ <port id="10" precision="I32">
259
+ <dim>-1</dim>
260
+ </port>
261
+ <port id="11" precision="U8">
262
+ <dim>-1</dim>
263
+ </port>
264
+ <port id="12" precision="BOOL">
265
+ <dim>-1</dim>
266
+ </port>
267
+ </output>
268
+ </layer>
269
+ <layer id="19" name="Constant_4168704" type="Const" version="opset1">
270
+ <data element_type="i32" shape="248077" offset="726" size="992308" />
271
+ <output>
272
+ <port id="0" precision="I32">
273
+ <dim>248077</dim>
274
+ </port>
275
+ </output>
276
+ </layer>
277
+ <layer id="20" name="Constant_4168706" type="Const" version="opset1">
278
+ <data element_type="i32" shape="248077" offset="993034" size="992308" />
279
+ <output>
280
+ <port id="0" precision="I32">
281
+ <dim>248077</dim>
282
+ </port>
283
+ </output>
284
+ </layer>
285
+ <layer id="21" name="Constant_4168708" type="Const" version="opset1">
286
+ <data element_type="u8" shape="1843484" offset="1985342" size="1843484" />
287
+ <output>
288
+ <port id="0" precision="U8">
289
+ <dim>1843484</dim>
290
+ </port>
291
+ </output>
292
+ </layer>
293
+ <layer id="22" name="Constant_4168716" type="Const" version="opset1">
294
+ <data element_type="i32" shape="247587" offset="3828826" size="990348" />
295
+ <output>
296
+ <port id="0" precision="I32">
297
+ <dim>247587</dim>
298
+ </port>
299
+ </output>
300
+ </layer>
301
+ <layer id="23" name="Constant_4168718" type="Const" version="opset1">
302
+ <data element_type="i32" shape="247587" offset="4819174" size="990348" />
303
+ <output>
304
+ <port id="0" precision="I32">
305
+ <dim>247587</dim>
306
+ </port>
307
+ </output>
308
+ </layer>
309
+ <layer id="24" name="Constant_4168720" type="Const" version="opset1">
310
+ <data element_type="u8" shape="938547" offset="5809522" size="938547" />
311
+ <output>
312
+ <port id="0" precision="U8">
313
+ <dim>938547</dim>
314
+ </port>
315
+ </output>
316
+ </layer>
317
+ <layer id="25" name="Constant_4168722" type="Const" version="opset1">
318
+ <data element_type="i32" shape="247587" offset="6748069" size="990348" />
319
+ <output>
320
+ <port id="0" precision="I32">
321
+ <dim>247587</dim>
322
+ </port>
323
+ </output>
324
+ </layer>
325
+ <layer id="26" name="Constant_4168724" type="Const" version="opset1">
326
+ <data element_type="i32" shape="247587" offset="7738417" size="990348" />
327
+ <output>
328
+ <port id="0" precision="I32">
329
+ <dim>247587</dim>
330
+ </port>
331
+ </output>
332
+ </layer>
333
+ <layer id="27" name="Constant_4168726" type="Const" version="opset1">
334
+ <data element_type="u8" shape="901525" offset="8728765" size="901525" />
335
+ <output>
336
+ <port id="0" precision="U8">
337
+ <dim>901525</dim>
338
+ </port>
339
+ </output>
340
+ </layer>
341
+ <layer id="28" name="Constant_4168710" type="Const" version="opset1">
342
+ <data element_type="i32" shape="33" offset="9630290" size="132" />
343
+ <output>
344
+ <port id="0" precision="I32">
345
+ <dim>33</dim>
346
+ </port>
347
+ </output>
348
+ </layer>
349
+ <layer id="29" name="Constant_4168712" type="Const" version="opset1">
350
+ <data element_type="i32" shape="33" offset="9630422" size="132" />
351
+ <output>
352
+ <port id="0" precision="I32">
353
+ <dim>33</dim>
354
+ </port>
355
+ </output>
356
+ </layer>
357
+ <layer id="30" name="Constant_4168714" type="Const" version="opset1">
358
+ <data element_type="u8" shape="439" offset="9630554" size="439" />
359
+ <output>
360
+ <port id="0" precision="U8">
361
+ <dim>439</dim>
362
+ </port>
363
+ </output>
364
+ </layer>
365
+ <layer id="31" name="Constant_4168727" type="Const" version="opset1">
366
+ <data element_type="i32" shape="33" offset="9630993" size="132" />
367
+ <output>
368
+ <port id="0" precision="I32">
369
+ <dim>33</dim>
370
+ </port>
371
+ </output>
372
+ </layer>
373
+ <layer id="32" name="BPETokenizer_4168728" type="BPETokenizer" version="extension">
374
+ <data unk_token="" fuse_unk="false" suffix_indicator="" end_suffix="" byte_fallback="false" cache_capacity="49608" />
375
+ <input>
376
+ <port id="0" precision="I32">
377
+ <dim>-1</dim>
378
+ </port>
379
+ <port id="1" precision="I32">
380
+ <dim>-1</dim>
381
+ </port>
382
+ <port id="2" precision="I32">
383
+ <dim>-1</dim>
384
+ </port>
385
+ <port id="3" precision="I32">
386
+ <dim>-1</dim>
387
+ </port>
388
+ <port id="4" precision="U8">
389
+ <dim>-1</dim>
390
+ </port>
391
+ <port id="5" precision="I32">
392
+ <dim>248077</dim>
393
+ </port>
394
+ <port id="6" precision="I32">
395
+ <dim>248077</dim>
396
+ </port>
397
+ <port id="7" precision="U8">
398
+ <dim>1843484</dim>
399
+ </port>
400
+ <port id="8" precision="I32">
401
+ <dim>247587</dim>
402
+ </port>
403
+ <port id="9" precision="I32">
404
+ <dim>247587</dim>
405
+ </port>
406
+ <port id="10" precision="U8">
407
+ <dim>938547</dim>
408
+ </port>
409
+ <port id="11" precision="I32">
410
+ <dim>247587</dim>
411
+ </port>
412
+ <port id="12" precision="I32">
413
+ <dim>247587</dim>
414
+ </port>
415
+ <port id="13" precision="U8">
416
+ <dim>901525</dim>
417
+ </port>
418
+ <port id="14" precision="I32">
419
+ <dim>33</dim>
420
+ </port>
421
+ <port id="15" precision="I32">
422
+ <dim>33</dim>
423
+ </port>
424
+ <port id="16" precision="U8">
425
+ <dim>439</dim>
426
+ </port>
427
+ <port id="17" precision="I32">
428
+ <dim>33</dim>
429
+ </port>
430
+ </input>
431
+ <output>
432
+ <port id="18" precision="I32">
433
+ <dim>-1</dim>
434
+ </port>
435
+ <port id="19" precision="I32">
436
+ <dim>-1</dim>
437
+ </port>
438
+ <port id="20" precision="I32">
439
+ <dim>-1</dim>
440
+ </port>
441
+ </output>
442
+ </layer>
443
+ <layer id="33" name="Constant_4168729" type="Const" version="opset1">
444
+ <data element_type="i32" shape="" offset="9631125" size="4" />
445
+ <output>
446
+ <port id="0" precision="I32" />
447
+ </output>
448
+ </layer>
449
+ <layer id="34" name="Constant_4168731" type="Const" version="opset1">
450
+ <data element_type="u8" shape="5" offset="9631129" size="5" />
451
+ <output>
452
+ <port id="0" precision="U8">
453
+ <dim>5</dim>
454
+ </port>
455
+ </output>
456
+ </layer>
457
+ <layer id="35" name="Constant_4168733" type="Const" version="opset1">
458
+ <data element_type="u8" shape="13" offset="9631134" size="13" />
459
+ <output>
460
+ <port id="0" precision="U8">
461
+ <dim>13</dim>
462
+ </port>
463
+ </output>
464
+ </layer>
465
+ <layer id="36" name="Truncate_4168734" type="Truncate" version="extension">
466
+ <data m_num_inputs="1" />
467
+ <input>
468
+ <port id="0" precision="I32">
469
+ <dim>-1</dim>
470
+ </port>
471
+ <port id="1" precision="I32">
472
+ <dim>-1</dim>
473
+ </port>
474
+ <port id="2" precision="I32">
475
+ <dim>-1</dim>
476
+ </port>
477
+ <port id="3" precision="I32" />
478
+ <port id="4" precision="U8">
479
+ <dim>5</dim>
480
+ </port>
481
+ <port id="5" precision="U8">
482
+ <dim>13</dim>
483
+ </port>
484
+ </input>
485
+ <output>
486
+ <port id="6" precision="I32">
487
+ <dim>-1</dim>
488
+ </port>
489
+ <port id="7" precision="I32">
490
+ <dim>-1</dim>
491
+ </port>
492
+ <port id="8" precision="I32">
493
+ <dim>-1</dim>
494
+ </port>
495
+ </output>
496
+ </layer>
497
+ <layer id="37" name="Constant_4168735" type="Const" version="opset1">
498
+ <data element_type="i32" shape="1" offset="9631147" size="4" />
499
+ <output>
500
+ <port id="0" precision="I32">
501
+ <dim>1</dim>
502
+ </port>
503
+ </output>
504
+ </layer>
505
+ <layer id="38" name="CombineSegments_4168736" type="CombineSegments" version="extension">
506
+ <input>
507
+ <port id="0" precision="I32">
508
+ <dim>-1</dim>
509
+ </port>
510
+ <port id="1" precision="I32">
511
+ <dim>-1</dim>
512
+ </port>
513
+ <port id="2" precision="I32">
514
+ <dim>-1</dim>
515
+ </port>
516
+ <port id="3" precision="I32">
517
+ <dim>1</dim>
518
+ </port>
519
+ </input>
520
+ <output>
521
+ <port id="4" precision="I32">
522
+ <dim>-1</dim>
523
+ </port>
524
+ <port id="5" precision="I32">
525
+ <dim>-1</dim>
526
+ </port>
527
+ <port id="6" precision="I32">
528
+ <dim>-1</dim>
529
+ </port>
530
+ <port id="7" precision="I32">
531
+ <dim>-1</dim>
532
+ </port>
533
+ <port id="8" precision="I32">
534
+ <dim>-1</dim>
535
+ </port>
536
+ <port id="9" precision="I32">
537
+ <dim>-1</dim>
538
+ </port>
539
+ </output>
540
+ </layer>
541
+ <layer id="39" name="Subtract_4168737" type="Subtract" version="opset1">
542
+ <data auto_broadcast="numpy" />
543
+ <input>
544
+ <port id="0" precision="I32">
545
+ <dim>-1</dim>
546
+ </port>
547
+ <port id="1" precision="I32">
548
+ <dim>-1</dim>
549
+ </port>
550
+ </input>
551
+ <output>
552
+ <port id="2" precision="I32">
553
+ <dim>-1</dim>
554
+ </port>
555
+ </output>
556
+ </layer>
557
+ <layer id="40" name="Constant_4168738" type="Const" version="opset1">
558
+ <data element_type="i32" shape="" offset="9631147" size="4" />
559
+ <output>
560
+ <port id="0" precision="I32" />
561
+ </output>
562
+ </layer>
563
+ <layer id="41" name="ReduceMax_4168739" type="ReduceMax" version="opset1">
564
+ <data keep_dims="false" />
565
+ <input>
566
+ <port id="0" precision="I32">
567
+ <dim>-1</dim>
568
+ </port>
569
+ <port id="1" precision="I32" />
570
+ </input>
571
+ <output>
572
+ <port id="2" precision="I32" />
573
+ </output>
574
+ </layer>
575
+ <layer id="42" name="Constant_4168740" type="Const" version="opset1">
576
+ <data element_type="i32" shape="" offset="9631151" size="4" />
577
+ <output>
578
+ <port id="0" precision="I32" />
579
+ </output>
580
+ </layer>
581
+ <layer id="43" name="RaggedToDense_4168741" type="RaggedToDense" version="extension">
582
+ <data pad_right="true" m_pad_max_length="false" />
583
+ <input>
584
+ <port id="0" precision="I32">
585
+ <dim>-1</dim>
586
+ </port>
587
+ <port id="1" precision="I32">
588
+ <dim>-1</dim>
589
+ </port>
590
+ <port id="2" precision="I32">
591
+ <dim>-1</dim>
592
+ </port>
593
+ <port id="3" precision="I32" />
594
+ <port id="4" precision="I32" />
595
+ </input>
596
+ <output>
597
+ <port id="5" precision="I32">
598
+ <dim>-1</dim>
599
+ <dim>-1</dim>
600
+ </port>
601
+ <port id="6" precision="BOOL">
602
+ <dim>-1</dim>
603
+ <dim>-1</dim>
604
+ </port>
605
+ </output>
606
+ </layer>
607
+ <layer id="44" name="Convert_4168742" type="Convert" version="opset1">
608
+ <data destination_type="i32" />
609
+ <input>
610
+ <port id="0" precision="BOOL">
611
+ <dim>-1</dim>
612
+ <dim>-1</dim>
613
+ </port>
614
+ </input>
615
+ <output>
616
+ <port id="1" precision="I32">
617
+ <dim>-1</dim>
618
+ <dim>-1</dim>
619
+ </port>
620
+ </output>
621
+ </layer>
622
+ <layer id="45" name="Convert_4168742.0" type="Convert" version="opset1">
623
+ <data destination_type="i64" />
624
+ <input>
625
+ <port id="0" precision="I32">
626
+ <dim>-1</dim>
627
+ <dim>-1</dim>
628
+ </port>
629
+ </input>
630
+ <output>
631
+ <port id="1" precision="I64" names="attention_mask">
632
+ <dim>-1</dim>
633
+ <dim>-1</dim>
634
+ </port>
635
+ </output>
636
+ </layer>
637
+ <layer id="46" name="RaggedToDense_4168741.0" type="Convert" version="opset1">
638
+ <data destination_type="i64" />
639
+ <input>
640
+ <port id="0" precision="I32">
641
+ <dim>-1</dim>
642
+ <dim>-1</dim>
643
+ </port>
644
+ </input>
645
+ <output>
646
+ <port id="1" precision="I64" names="input_ids">
647
+ <dim>-1</dim>
648
+ <dim>-1</dim>
649
+ </port>
650
+ </output>
651
+ </layer>
652
+ <layer id="47" name="Result_4168745" type="Result" version="opset1" output_names="input_ids">
653
+ <input>
654
+ <port id="0" precision="I64">
655
+ <dim>-1</dim>
656
+ <dim>-1</dim>
657
+ </port>
658
+ </input>
659
+ </layer>
660
+ <layer id="48" name="Result_4168747" type="Result" version="opset1" output_names="attention_mask">
661
+ <input>
662
+ <port id="0" precision="I64">
663
+ <dim>-1</dim>
664
+ <dim>-1</dim>
665
+ </port>
666
+ </input>
667
+ </layer>
668
+ </layers>
669
+ <edges>
670
+ <edge from-layer="0" from-port="0" to-layer="2" to-port="0" />
671
+ <edge from-layer="1" from-port="0" to-layer="8" to-port="0" />
672
+ <edge from-layer="2" from-port="1" to-layer="3" to-port="0" />
673
+ <edge from-layer="2" from-port="3" to-layer="15" to-port="4" />
674
+ <edge from-layer="2" from-port="2" to-layer="15" to-port="3" />
675
+ <edge from-layer="2" from-port="1" to-layer="15" to-port="2" />
676
+ <edge from-layer="3" from-port="1" to-layer="6" to-port="0" />
677
+ <edge from-layer="4" from-port="0" to-layer="6" to-port="1" />
678
+ <edge from-layer="5" from-port="0" to-layer="6" to-port="2" />
679
+ <edge from-layer="6" from-port="3" to-layer="11" to-port="0" />
680
+ <edge from-layer="6" from-port="3" to-layer="8" to-port="1" />
681
+ <edge from-layer="7" from-port="0" to-layer="8" to-port="2" />
682
+ <edge from-layer="8" from-port="3" to-layer="15" to-port="0" />
683
+ <edge from-layer="9" from-port="0" to-layer="13" to-port="0" />
684
+ <edge from-layer="10" from-port="0" to-layer="11" to-port="1" />
685
+ <edge from-layer="11" from-port="2" to-layer="13" to-port="1" />
686
+ <edge from-layer="12" from-port="0" to-layer="13" to-port="2" />
687
+ <edge from-layer="13" from-port="3" to-layer="15" to-port="1" />
688
+ <edge from-layer="14" from-port="0" to-layer="15" to-port="5" />
689
+ <edge from-layer="15" from-port="10" to-layer="16" to-port="2" />
690
+ <edge from-layer="15" from-port="7" to-layer="18" to-port="1" />
691
+ <edge from-layer="15" from-port="6" to-layer="18" to-port="0" />
692
+ <edge from-layer="15" from-port="11" to-layer="16" to-port="3" />
693
+ <edge from-layer="15" from-port="9" to-layer="16" to-port="1" />
694
+ <edge from-layer="15" from-port="8" to-layer="16" to-port="0" />
695
+ <edge from-layer="16" from-port="4" to-layer="18" to-port="2" />
696
+ <edge from-layer="16" from-port="5" to-layer="18" to-port="3" />
697
+ <edge from-layer="16" from-port="6" to-layer="18" to-port="4" />
698
+ <edge from-layer="16" from-port="7" to-layer="18" to-port="5" />
699
+ <edge from-layer="17" from-port="0" to-layer="18" to-port="6" />
700
+ <edge from-layer="18" from-port="7" to-layer="32" to-port="0" />
701
+ <edge from-layer="18" from-port="8" to-layer="32" to-port="1" />
702
+ <edge from-layer="18" from-port="9" to-layer="32" to-port="2" />
703
+ <edge from-layer="18" from-port="10" to-layer="32" to-port="3" />
704
+ <edge from-layer="18" from-port="11" to-layer="32" to-port="4" />
705
+ <edge from-layer="19" from-port="0" to-layer="32" to-port="5" />
706
+ <edge from-layer="20" from-port="0" to-layer="32" to-port="6" />
707
+ <edge from-layer="21" from-port="0" to-layer="32" to-port="7" />
708
+ <edge from-layer="22" from-port="0" to-layer="32" to-port="8" />
709
+ <edge from-layer="23" from-port="0" to-layer="32" to-port="9" />
710
+ <edge from-layer="24" from-port="0" to-layer="32" to-port="10" />
711
+ <edge from-layer="25" from-port="0" to-layer="32" to-port="11" />
712
+ <edge from-layer="26" from-port="0" to-layer="32" to-port="12" />
713
+ <edge from-layer="27" from-port="0" to-layer="32" to-port="13" />
714
+ <edge from-layer="28" from-port="0" to-layer="32" to-port="14" />
715
+ <edge from-layer="29" from-port="0" to-layer="32" to-port="15" />
716
+ <edge from-layer="30" from-port="0" to-layer="32" to-port="16" />
717
+ <edge from-layer="31" from-port="0" to-layer="32" to-port="17" />
718
+ <edge from-layer="32" from-port="18" to-layer="36" to-port="0" />
719
+ <edge from-layer="32" from-port="19" to-layer="36" to-port="1" />
720
+ <edge from-layer="32" from-port="20" to-layer="36" to-port="2" />
721
+ <edge from-layer="33" from-port="0" to-layer="36" to-port="3" />
722
+ <edge from-layer="34" from-port="0" to-layer="36" to-port="4" />
723
+ <edge from-layer="35" from-port="0" to-layer="36" to-port="5" />
724
+ <edge from-layer="36" from-port="6" to-layer="38" to-port="0" />
725
+ <edge from-layer="36" from-port="8" to-layer="38" to-port="2" />
726
+ <edge from-layer="36" from-port="7" to-layer="38" to-port="1" />
727
+ <edge from-layer="37" from-port="0" to-layer="38" to-port="3" />
728
+ <edge from-layer="38" from-port="5" to-layer="39" to-port="0" />
729
+ <edge from-layer="38" from-port="4" to-layer="39" to-port="1" />
730
+ <edge from-layer="38" from-port="4" to-layer="43" to-port="0" />
731
+ <edge from-layer="38" from-port="5" to-layer="43" to-port="1" />
732
+ <edge from-layer="38" from-port="6" to-layer="43" to-port="2" />
733
+ <edge from-layer="39" from-port="2" to-layer="41" to-port="0" />
734
+ <edge from-layer="40" from-port="0" to-layer="41" to-port="1" />
735
+ <edge from-layer="41" from-port="2" to-layer="43" to-port="3" />
736
+ <edge from-layer="42" from-port="0" to-layer="43" to-port="4" />
737
+ <edge from-layer="43" from-port="6" to-layer="44" to-port="0" />
738
+ <edge from-layer="43" from-port="5" to-layer="46" to-port="0" />
739
+ <edge from-layer="44" from-port="1" to-layer="45" to-port="0" />
740
+ <edge from-layer="45" from-port="1" to-layer="48" to-port="0" />
741
+ <edge from-layer="46" from-port="1" to-layer="47" to-port="0" />
742
+ </edges>
743
+ <rt_info>
744
+ <info name="OpenVINO Runtime" value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
745
+ <add_attention_mask value="True" />
746
+ <add_prefix_space />
747
+ <add_special_tokens value="True" />
748
+ <chat_template value="{%- set image_count = namespace(value=0) %}&#10;{%- set video_count = namespace(value=0) %}&#10;{%- macro render_content(content, do_vision_count, is_system_content=false) %}&#10; {%- if content is string %}&#10; {{- content }}&#10; {%- elif content is iterable and content is not mapping %}&#10; {%- for item in content %}&#10; {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}&#10; {%- if is_system_content %}&#10; {{- raise_exception('System message cannot contain images.') }}&#10; {%- endif %}&#10; {%- if do_vision_count %}&#10; {%- set image_count.value = image_count.value + 1 %}&#10; {%- endif %}&#10; {%- if add_vision_id %}&#10; {{- 'Picture ' ~ image_count.value ~ ': ' }}&#10; {%- endif %}&#10; {{- '&lt;|vision_start|>&lt;|image_pad|>&lt;|vision_end|>' }}&#10; {%- elif 'video' in item or item.type == 'video' %}&#10; {%- if is_system_content %}&#10; {{- raise_exception('System message cannot contain videos.') }}&#10; {%- endif %}&#10; {%- if do_vision_count %}&#10; {%- set video_count.value = video_count.value + 1 %}&#10; {%- endif %}&#10; {%- if add_vision_id %}&#10; {{- 'Video ' ~ video_count.value ~ ': ' }}&#10; {%- endif %}&#10; {{- '&lt;|vision_start|>&lt;|video_pad|>&lt;|vision_end|>' }}&#10; {%- elif 'text' in item %}&#10; {{- item.text }}&#10; {%- else %}&#10; {{- raise_exception('Unexpected item type in content.') }}&#10; {%- endif %}&#10; {%- endfor %}&#10; {%- elif content is none or content is undefined %}&#10; {{- '' }}&#10; {%- else %}&#10; {{- raise_exception('Unexpected content type.') }}&#10; {%- endif %}&#10;{%- endmacro %}&#10;{%- if not messages %}&#10; {{- raise_exception('No messages provided.') }}&#10;{%- endif %}&#10;{%- if tools and tools is iterable and tools is not mapping %}&#10; {{- '&lt;|im_start|>system\n' }}&#10; {{- &quot;# Tools\n\nYou have access to the following functions:\n\n&lt;tools>&quot; }}&#10; {%- for tool in tools %}&#10; {{- &quot;\n&quot; }}&#10; {{- tool | tojson }}&#10; {%- endfor %}&#10; {{- &quot;\n&lt;/tools>&quot; }}&#10; {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n&lt;tool_call>\n&lt;function=example_function_name>\n&lt;parameter=example_parameter_1>\nvalue_1\n&lt;/parameter>\n&lt;parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n&lt;/parameter>\n&lt;/function>\n&lt;/tool_call>\n\n&lt;IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner &lt;function=...>&lt;/function> block must be nested within &lt;tool_call>&lt;/tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n&lt;/IMPORTANT>' }}&#10; {%- if messages[0].role == 'system' %}&#10; {%- set content = render_content(messages[0].content, false, true)|trim %}&#10; {%- if content %}&#10; {{- '\n\n' + content }}&#10; {%- endif %}&#10; {%- endif %}&#10; {{- '&lt;|im_end|>\n' }}&#10;{%- else %}&#10; {%- if messages[0].role == 'system' %}&#10; {%- set content = render_content(messages[0].content, false, true)|trim %}&#10; {{- '&lt;|im_start|>system\n' + content + '&lt;|im_end|>\n' }}&#10; {%- endif %}&#10;{%- endif %}&#10;{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}&#10;{%- for message in messages[::-1] %}&#10; {%- set index = (messages|length - 1) - loop.index0 %}&#10; {%- if ns.multi_step_tool and message.role == &quot;user&quot; %}&#10; {%- set content = render_content(message.content, false)|trim %}&#10; {%- if not(content.startswith('&lt;tool_response>') and content.endswith('&lt;/tool_response>')) %}&#10; {%- set ns.multi_step_tool = false %}&#10; {%- set ns.last_query_index = index %}&#10; {%- endif %}&#10; {%- endif %}&#10;{%- endfor %}&#10;{%- if ns.multi_step_tool %}&#10; {{- raise_exception('No user query found in messages.') }}&#10;{%- endif %}&#10;{%- for message in messages %}&#10; {%- set content = render_content(message.content, true)|trim %}&#10; {%- if message.role == &quot;system&quot; %}&#10; {%- if not loop.first %}&#10; {{- raise_exception('System message must be at the beginning.') }}&#10; {%- endif %}&#10; {%- elif message.role == &quot;user&quot; %}&#10; {{- '&lt;|im_start|>' + message.role + '\n' + content + '&lt;|im_end|>' + '\n' }}&#10; {%- elif message.role == &quot;assistant&quot; %}&#10; {%- set reasoning_content = '' %}&#10; {%- if message.reasoning_content is string %}&#10; {%- set reasoning_content = message.reasoning_content %}&#10; {%- else %}&#10; {%- if '&lt;/think>' in content %}&#10; {%- set reasoning_content = content.split('&lt;/think>')[0].rstrip('\n').split('&lt;think>')[-1].lstrip('\n') %}&#10; {%- set content = content.split('&lt;/think>')[-1].lstrip('\n') %}&#10; {%- endif %}&#10; {%- endif %}&#10; {%- set reasoning_content = reasoning_content|trim %}&#10; {%- if (preserve_thinking is defined and preserve_thinking is true) or (loop.index0 > ns.last_query_index) %}&#10; {{- '&lt;|im_start|>' + message.role + '\n&lt;think>\n' + reasoning_content + '\n&lt;/think>\n\n' + content }}&#10; {%- else %}&#10; {{- '&lt;|im_start|>' + message.role + '\n' + content }}&#10; {%- endif %}&#10; {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}&#10; {%- for tool_call in message.tool_calls %}&#10; {%- if tool_call.function is defined %}&#10; {%- set tool_call = tool_call.function %}&#10; {%- endif %}&#10; {%- if loop.first %}&#10; {%- if content|trim %}&#10; {{- '\n\n&lt;tool_call>\n&lt;function=' + tool_call.name + '>\n' }}&#10; {%- else %}&#10; {{- '&lt;tool_call>\n&lt;function=' + tool_call.name + '>\n' }}&#10; {%- endif %}&#10; {%- else %}&#10; {{- '\n&lt;tool_call>\n&lt;function=' + tool_call.name + '>\n' }}&#10; {%- endif %}&#10; {%- if tool_call.arguments is defined %}&#10; {%- for args_name, args_value in tool_call.arguments|items %}&#10; {{- '&lt;parameter=' + args_name + '>\n' }}&#10; {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}&#10; {{- args_value }}&#10; {{- '\n&lt;/parameter>\n' }}&#10; {%- endfor %}&#10; {%- endif %}&#10; {{- '&lt;/function>\n&lt;/tool_call>' }}&#10; {%- endfor %}&#10; {%- endif %}&#10; {{- '&lt;|im_end|>\n' }}&#10; {%- elif message.role == &quot;tool&quot; %}&#10; {%- if loop.previtem and loop.previtem.role != &quot;tool&quot; %}&#10; {{- '&lt;|im_start|>user' }}&#10; {%- endif %}&#10; {{- '\n&lt;tool_response>\n' }}&#10; {{- content }}&#10; {{- '\n&lt;/tool_response>' }}&#10; {%- if not loop.last and loop.nextitem.role != &quot;tool&quot; %}&#10; {{- '&lt;|im_end|>\n' }}&#10; {%- elif loop.last %}&#10; {{- '&lt;|im_end|>\n' }}&#10; {%- endif %}&#10; {%- else %}&#10; {{- raise_exception('Unexpected message role.') }}&#10; {%- endif %}&#10;{%- endfor %}&#10;{%- if add_generation_prompt %}&#10; {{- '&lt;|im_start|>assistant\n' }}&#10; {%- if enable_thinking is defined and enable_thinking is false %}&#10; {{- '&lt;think>\n\n&lt;/think>\n\n' }}&#10; {%- else %}&#10; {{- '&lt;think>\n' }}&#10; {%- endif %}&#10;{%- endif %}" />
749
+ <clean_up_tokenization_spaces />
750
+ <detokenizer_input_type value="i64" />
751
+ <eos_token_id value="248046" />
752
+ <handle_special_tokens_with_re />
753
+ <max_length />
754
+ <number_of_inputs value="1" />
755
+ <openvino_tokenizers_version value="2026.2.1.0-682-6da74a793f0" />
756
+ <openvino_version value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
757
+ <original_post_processor_template value="{&quot;type&quot;: &quot;ByteLevel&quot;, &quot;add_prefix_space&quot;: false, &quot;trim_offsets&quot;: false, &quot;use_regex&quot;: false}" />
758
+ <original_tokenizer_class value="&lt;class 'transformers.tokenization_utils_tokenizers.TokenizersBackend'>" />
759
+ <pad_token_id value="248044" />
760
+ <processed_post_processor_template value="{&quot;single&quot;: {&quot;ids&quot;: [-1], &quot;type_ids&quot;: [0]}, &quot;pair&quot;: {&quot;ids&quot;: [-1, -2], &quot;type_ids&quot;: [0, 0]}}" />
761
+ <sentencepiece_version value="0.2.2" />
762
+ <skip_special_tokens value="True" />
763
+ <streaming_detokenizer value="False" />
764
+ <tiktoken_version value="0.13.0" />
765
+ <tokenizer_output_type value="i64" />
766
+ <tokenizers_version value="0.22.2" />
767
+ <transformers_version value="5.2.0" />
768
+ <use_max_padding value="False" />
769
+ <use_sentencepiece_backend value="False" />
770
+ <utf8_replace_mode value="replace" />
771
+ <with_detokenizer value="True" />
772
+ </rt_info>
773
+ </net>
openvino_vision_embeddings_merger_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3986801bf93dbf0871da05be045429cbadfffbc5b6006d042a3e4961089b85a
3
+ size 457529956
openvino_vision_embeddings_merger_model.xml ADDED
The diff for this file is too large to render. See raw diff
 
openvino_vision_embeddings_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7af806e6ebf378e26ad3c34258999b30b100c55cfea1f4dc2b38f1537425919f
3
+ size 1776440
openvino_vision_embeddings_model.xml ADDED
@@ -0,0 +1,291 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="Model99" version="11">
3
+ <layers>
4
+ <layer id="0" name="hidden_states" type="Parameter" version="opset1">
5
+ <data shape="?,?" element_type="f32" />
6
+ <output>
7
+ <port id="0" precision="FP32" names="hidden_states">
8
+ <dim>-1</dim>
9
+ <dim>-1</dim>
10
+ </port>
11
+ </output>
12
+ </layer>
13
+ <layer id="1" name="Constant_379065" type="Const" version="opset1">
14
+ <data element_type="i64" shape="5" offset="0" size="40" />
15
+ <output>
16
+ <port id="0" precision="I64">
17
+ <dim>5</dim>
18
+ </port>
19
+ </output>
20
+ </layer>
21
+ <layer id="2" name="aten::view/Reshape" type="Reshape" version="opset1">
22
+ <data special_zero="false" />
23
+ <input>
24
+ <port id="0" precision="FP32">
25
+ <dim>-1</dim>
26
+ <dim>-1</dim>
27
+ </port>
28
+ <port id="1" precision="I64">
29
+ <dim>5</dim>
30
+ </port>
31
+ </input>
32
+ <output>
33
+ <port id="2" precision="FP32" names="14,9,hidden_states_1,input">
34
+ <dim>-1</dim>
35
+ <dim>3</dim>
36
+ <dim>2</dim>
37
+ <dim>16</dim>
38
+ <dim>16</dim>
39
+ </port>
40
+ </output>
41
+ </layer>
42
+ <layer id="3" name="self.proj.weight" type="Const" version="opset1">
43
+ <data element_type="i8" shape="1152, 3, 2, 16, 16" offset="40" size="1769472" />
44
+ <output>
45
+ <port id="0" precision="I8">
46
+ <dim>1152</dim>
47
+ <dim>3</dim>
48
+ <dim>2</dim>
49
+ <dim>16</dim>
50
+ <dim>16</dim>
51
+ </port>
52
+ </output>
53
+ </layer>
54
+ <layer id="4" name="Convert_3369892" type="Convert" version="opset1">
55
+ <data destination_type="f16" />
56
+ <input>
57
+ <port id="0" precision="I8">
58
+ <dim>1152</dim>
59
+ <dim>3</dim>
60
+ <dim>2</dim>
61
+ <dim>16</dim>
62
+ <dim>16</dim>
63
+ </port>
64
+ </input>
65
+ <output>
66
+ <port id="1" precision="FP16">
67
+ <dim>1152</dim>
68
+ <dim>3</dim>
69
+ <dim>2</dim>
70
+ <dim>16</dim>
71
+ <dim>16</dim>
72
+ </port>
73
+ </output>
74
+ </layer>
75
+ <layer id="5" name="self.proj.weight/scale" type="Const" version="opset1">
76
+ <data element_type="f16" shape="1152, 1, 1, 1, 1" offset="1769512" size="2304" />
77
+ <output>
78
+ <port id="0" precision="FP16">
79
+ <dim>1152</dim>
80
+ <dim>1</dim>
81
+ <dim>1</dim>
82
+ <dim>1</dim>
83
+ <dim>1</dim>
84
+ </port>
85
+ </output>
86
+ </layer>
87
+ <layer id="6" name="self.proj.weight/fq_weights_1" type="Multiply" version="opset1">
88
+ <data auto_broadcast="numpy" />
89
+ <input>
90
+ <port id="0" precision="FP16">
91
+ <dim>1152</dim>
92
+ <dim>3</dim>
93
+ <dim>2</dim>
94
+ <dim>16</dim>
95
+ <dim>16</dim>
96
+ </port>
97
+ <port id="1" precision="FP16">
98
+ <dim>1152</dim>
99
+ <dim>1</dim>
100
+ <dim>1</dim>
101
+ <dim>1</dim>
102
+ <dim>1</dim>
103
+ </port>
104
+ </input>
105
+ <output>
106
+ <port id="2" precision="FP16">
107
+ <dim>1152</dim>
108
+ <dim>3</dim>
109
+ <dim>2</dim>
110
+ <dim>16</dim>
111
+ <dim>16</dim>
112
+ </port>
113
+ </output>
114
+ </layer>
115
+ <layer id="7" name="self.proj.weight/fq_weights_1/convert" type="Convert" version="opset1">
116
+ <data destination_type="f32" />
117
+ <input>
118
+ <port id="0" precision="FP16">
119
+ <dim>1152</dim>
120
+ <dim>3</dim>
121
+ <dim>2</dim>
122
+ <dim>16</dim>
123
+ <dim>16</dim>
124
+ </port>
125
+ </input>
126
+ <output>
127
+ <port id="1" precision="FP32">
128
+ <dim>1152</dim>
129
+ <dim>3</dim>
130
+ <dim>2</dim>
131
+ <dim>16</dim>
132
+ <dim>16</dim>
133
+ </port>
134
+ </output>
135
+ </layer>
136
+ <layer id="8" name="__module.proj/aten::_convolution/Convolution" type="Convolution" version="opset1">
137
+ <data strides="2, 16, 16" dilations="1, 1, 1" pads_begin="0, 0, 0" pads_end="0, 0, 0" auto_pad="explicit" />
138
+ <input>
139
+ <port id="0" precision="FP32">
140
+ <dim>-1</dim>
141
+ <dim>3</dim>
142
+ <dim>2</dim>
143
+ <dim>16</dim>
144
+ <dim>16</dim>
145
+ </port>
146
+ <port id="1" precision="FP32">
147
+ <dim>1152</dim>
148
+ <dim>3</dim>
149
+ <dim>2</dim>
150
+ <dim>16</dim>
151
+ <dim>16</dim>
152
+ </port>
153
+ </input>
154
+ <output>
155
+ <port id="2" precision="FP32">
156
+ <dim>-1</dim>
157
+ <dim>1152</dim>
158
+ <dim>1</dim>
159
+ <dim>1</dim>
160
+ <dim>1</dim>
161
+ </port>
162
+ </output>
163
+ </layer>
164
+ <layer id="9" name="__module.proj/aten::_convolution/Reshape" type="Const" version="opset1">
165
+ <data element_type="f32" shape="1, 1152, 1, 1, 1" offset="1771816" size="4608" />
166
+ <output>
167
+ <port id="0" precision="FP32">
168
+ <dim>1</dim>
169
+ <dim>1152</dim>
170
+ <dim>1</dim>
171
+ <dim>1</dim>
172
+ <dim>1</dim>
173
+ </port>
174
+ </output>
175
+ </layer>
176
+ <layer id="10" name="__module.proj/aten::_convolution/Add" type="Add" version="opset1">
177
+ <data auto_broadcast="numpy" />
178
+ <input>
179
+ <port id="0" precision="FP32">
180
+ <dim>-1</dim>
181
+ <dim>1152</dim>
182
+ <dim>1</dim>
183
+ <dim>1</dim>
184
+ <dim>1</dim>
185
+ </port>
186
+ <port id="1" precision="FP32">
187
+ <dim>1</dim>
188
+ <dim>1152</dim>
189
+ <dim>1</dim>
190
+ <dim>1</dim>
191
+ <dim>1</dim>
192
+ </port>
193
+ </input>
194
+ <output>
195
+ <port id="2" precision="FP32" names="33">
196
+ <dim>-1</dim>
197
+ <dim>1152</dim>
198
+ <dim>1</dim>
199
+ <dim>1</dim>
200
+ <dim>1</dim>
201
+ </port>
202
+ </output>
203
+ </layer>
204
+ <layer id="11" name="Constant_379138" type="Const" version="opset1">
205
+ <data element_type="i64" shape="2" offset="1776424" size="16" />
206
+ <output>
207
+ <port id="0" precision="I64">
208
+ <dim>2</dim>
209
+ </port>
210
+ </output>
211
+ </layer>
212
+ <layer id="12" name="aten::view/Reshape_1" type="Reshape" version="opset1">
213
+ <data special_zero="false" />
214
+ <input>
215
+ <port id="0" precision="FP32">
216
+ <dim>-1</dim>
217
+ <dim>1152</dim>
218
+ <dim>1</dim>
219
+ <dim>1</dim>
220
+ <dim>1</dim>
221
+ </port>
222
+ <port id="1" precision="I64">
223
+ <dim>2</dim>
224
+ </port>
225
+ </input>
226
+ <output>
227
+ <port id="2" precision="FP32" names="last_hidden_state">
228
+ <dim>-1</dim>
229
+ <dim>1152</dim>
230
+ </port>
231
+ </output>
232
+ </layer>
233
+ <layer id="13" name="Result_379166" type="Result" version="opset1" output_names="last_hidden_state">
234
+ <input>
235
+ <port id="0" precision="FP32">
236
+ <dim>-1</dim>
237
+ <dim>1152</dim>
238
+ </port>
239
+ </input>
240
+ </layer>
241
+ </layers>
242
+ <edges>
243
+ <edge from-layer="0" from-port="0" to-layer="2" to-port="0" />
244
+ <edge from-layer="1" from-port="0" to-layer="2" to-port="1" />
245
+ <edge from-layer="2" from-port="2" to-layer="8" to-port="0" />
246
+ <edge from-layer="3" from-port="0" to-layer="4" to-port="0" />
247
+ <edge from-layer="4" from-port="1" to-layer="6" to-port="0" />
248
+ <edge from-layer="5" from-port="0" to-layer="6" to-port="1" />
249
+ <edge from-layer="6" from-port="2" to-layer="7" to-port="0" />
250
+ <edge from-layer="7" from-port="1" to-layer="8" to-port="1" />
251
+ <edge from-layer="8" from-port="2" to-layer="10" to-port="0" />
252
+ <edge from-layer="9" from-port="0" to-layer="10" to-port="1" />
253
+ <edge from-layer="10" from-port="2" to-layer="12" to-port="0" />
254
+ <edge from-layer="11" from-port="0" to-layer="12" to-port="1" />
255
+ <edge from-layer="12" from-port="2" to-layer="13" to-port="0" />
256
+ </edges>
257
+ <rt_info>
258
+ <info name="OpenVINO Runtime" value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
259
+ <Runtime_version value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
260
+ <conversion_parameters>
261
+ <framework value="pytorch" />
262
+ <is_python_object value="True" />
263
+ </conversion_parameters>
264
+ <nncf>
265
+ <friendly_names_were_updated value="True" />
266
+ <version value="3.2.0" />
267
+ <weight_compression>
268
+ <advanced_parameters value="{'statistics_path': None, 'lora_adapter_rank': 256, 'group_size_fallback_mode': 'error', 'min_adjusted_group_size': 32, 'awq_params': {'subset_size': 32, 'percent_to_apply': 0.002, 'alpha_min': 0.0, 'alpha_max': 1.0, 'steps': 100, 'prefer_data_aware_scaling': True}, 'scale_estimation_params': {'subset_size': 64, 'initial_steps': 5, 'scale_steps': 5, 'weight_penalty': -1.0}, 'gptq_params': {'damp_percent': 0.1, 'block_size': 128, 'subset_size': 128}, 'lora_correction_params': {'adapter_rank': 8, 'num_iterations': 3, 'apply_regularization': True, 'subset_size': 128, 'use_int8_adapters': True}, 'backend_params': {}, 'codebook': None, 'adaptive_codebook_params': {'value_type': 'f8e4m3', 'across_blocks': False, 'num_elements': 16}}" />
269
+ <all_layers value="False" />
270
+ <awq value="False" />
271
+ <backup_mode value="int8_asym" />
272
+ <compression_format value="dequantize" />
273
+ <gptq value="False" />
274
+ <group_size value="-1" />
275
+ <ignored_scope value="[]" />
276
+ <lora_correction value="False" />
277
+ <mode value="int8_sym" />
278
+ <ratio value="1.0" />
279
+ <scale_estimation value="False" />
280
+ <sensitivity_metric value="weight_quantization_error" />
281
+ </weight_compression>
282
+ </nncf>
283
+ <optimum>
284
+ <nncf_version value="3.2.0" />
285
+ <optimum_intel_version value="2.1.0.dev0+109314c" />
286
+ <optimum_version value="2.2.0.dev0" />
287
+ <pytorch_version value="2.11.0+cu128" />
288
+ <transformers_version value="5.2.0" />
289
+ </optimum>
290
+ </rt_info>
291
+ </net>
openvino_vision_embeddings_pos_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:736234aba414e875f22c0e604188ee27c7667be5bebbcbb0c90a1969b3c6b2a5
3
+ size 2658820
openvino_vision_embeddings_pos_model.xml ADDED
@@ -0,0 +1,180 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <?xml version="1.0"?>
2
+ <net name="Model108" version="11">
3
+ <layers>
4
+ <layer id="0" name="input" type="Parameter" version="opset1">
5
+ <data shape="4,?" element_type="i64" />
6
+ <output>
7
+ <port id="0" precision="I64" names="input">
8
+ <dim>4</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="i8" shape="2304, 1152" offset="0" size="2654208" />
15
+ <output>
16
+ <port id="0" precision="I8">
17
+ <dim>2304</dim>
18
+ <dim>1152</dim>
19
+ </port>
20
+ </output>
21
+ </layer>
22
+ <layer id="2" name="Convert_3942764" type="Convert" version="opset1">
23
+ <data destination_type="f16" />
24
+ <input>
25
+ <port id="0" precision="I8">
26
+ <dim>2304</dim>
27
+ <dim>1152</dim>
28
+ </port>
29
+ </input>
30
+ <output>
31
+ <port id="1" precision="FP16">
32
+ <dim>2304</dim>
33
+ <dim>1152</dim>
34
+ </port>
35
+ </output>
36
+ </layer>
37
+ <layer id="3" name="self.weight/scale" type="Const" version="opset1">
38
+ <data element_type="f16" shape="2304, 1" offset="2654208" size="4608" />
39
+ <output>
40
+ <port id="0" precision="FP16">
41
+ <dim>2304</dim>
42
+ <dim>1</dim>
43
+ </port>
44
+ </output>
45
+ </layer>
46
+ <layer id="4" name="self.weight/fq_weights_0" type="Multiply" version="opset1">
47
+ <data auto_broadcast="numpy" />
48
+ <input>
49
+ <port id="0" precision="FP16">
50
+ <dim>2304</dim>
51
+ <dim>1152</dim>
52
+ </port>
53
+ <port id="1" precision="FP16">
54
+ <dim>2304</dim>
55
+ <dim>1</dim>
56
+ </port>
57
+ </input>
58
+ <output>
59
+ <port id="2" precision="FP16">
60
+ <dim>2304</dim>
61
+ <dim>1152</dim>
62
+ </port>
63
+ </output>
64
+ </layer>
65
+ <layer id="5" name="ov_ext::embedding/Convert" type="Convert" version="opset1">
66
+ <data destination_type="f32" />
67
+ <rt_info>
68
+ <attribute name="decompression" version="0" />
69
+ </rt_info>
70
+ <input>
71
+ <port id="0" precision="FP16">
72
+ <dim>2304</dim>
73
+ <dim>1152</dim>
74
+ </port>
75
+ </input>
76
+ <output>
77
+ <port id="1" precision="FP32">
78
+ <dim>2304</dim>
79
+ <dim>1152</dim>
80
+ </port>
81
+ </output>
82
+ </layer>
83
+ <layer id="6" name="ov_ext::embedding/Convert_1" type="Convert" version="opset1">
84
+ <data destination_type="i32" />
85
+ <input>
86
+ <port id="0" precision="I64">
87
+ <dim>4</dim>
88
+ <dim>-1</dim>
89
+ </port>
90
+ </input>
91
+ <output>
92
+ <port id="1" precision="I32">
93
+ <dim>4</dim>
94
+ <dim>-1</dim>
95
+ </port>
96
+ </output>
97
+ </layer>
98
+ <layer id="7" name="ov_ext::embedding/Constant" type="Const" version="opset1">
99
+ <data element_type="i32" shape="" offset="2658816" size="4" />
100
+ <output>
101
+ <port id="0" precision="I32" />
102
+ </output>
103
+ </layer>
104
+ <layer id="8" name="ov_ext::embedding/Gather" type="Gather" version="opset8">
105
+ <data batch_dims="0" />
106
+ <input>
107
+ <port id="0" precision="FP32">
108
+ <dim>2304</dim>
109
+ <dim>1152</dim>
110
+ </port>
111
+ <port id="1" precision="I32">
112
+ <dim>4</dim>
113
+ <dim>-1</dim>
114
+ </port>
115
+ <port id="2" precision="I32" />
116
+ </input>
117
+ <output>
118
+ <port id="3" precision="FP32" names="last_hidden_state">
119
+ <dim>4</dim>
120
+ <dim>-1</dim>
121
+ <dim>1152</dim>
122
+ </port>
123
+ </output>
124
+ </layer>
125
+ <layer id="9" name="Result_457229" type="Result" version="opset1" output_names="last_hidden_state">
126
+ <input>
127
+ <port id="0" precision="FP32">
128
+ <dim>4</dim>
129
+ <dim>-1</dim>
130
+ <dim>1152</dim>
131
+ </port>
132
+ </input>
133
+ </layer>
134
+ </layers>
135
+ <edges>
136
+ <edge from-layer="0" from-port="0" to-layer="6" to-port="0" />
137
+ <edge from-layer="1" from-port="0" to-layer="2" to-port="0" />
138
+ <edge from-layer="2" from-port="1" to-layer="4" to-port="0" />
139
+ <edge from-layer="3" from-port="0" to-layer="4" to-port="1" />
140
+ <edge from-layer="4" from-port="2" to-layer="5" to-port="0" />
141
+ <edge from-layer="5" from-port="1" to-layer="8" to-port="0" />
142
+ <edge from-layer="6" from-port="1" to-layer="8" to-port="1" />
143
+ <edge from-layer="7" from-port="0" to-layer="8" to-port="2" />
144
+ <edge from-layer="8" from-port="3" to-layer="9" to-port="0" />
145
+ </edges>
146
+ <rt_info>
147
+ <info name="OpenVINO Runtime" value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
148
+ <Runtime_version value="2026.2.1-21919-ede283a88e3-releases/2026/2" />
149
+ <conversion_parameters>
150
+ <framework value="pytorch" />
151
+ <is_python_object value="True" />
152
+ </conversion_parameters>
153
+ <nncf>
154
+ <friendly_names_were_updated value="True" />
155
+ <version value="3.2.0" />
156
+ <weight_compression>
157
+ <advanced_parameters value="{'statistics_path': None, 'lora_adapter_rank': 256, 'group_size_fallback_mode': 'error', 'min_adjusted_group_size': 32, 'awq_params': {'subset_size': 32, 'percent_to_apply': 0.002, 'alpha_min': 0.0, 'alpha_max': 1.0, 'steps': 100, 'prefer_data_aware_scaling': True}, 'scale_estimation_params': {'subset_size': 64, 'initial_steps': 5, 'scale_steps': 5, 'weight_penalty': -1.0}, 'gptq_params': {'damp_percent': 0.1, 'block_size': 128, 'subset_size': 128}, 'lora_correction_params': {'adapter_rank': 8, 'num_iterations': 3, 'apply_regularization': True, 'subset_size': 128, 'use_int8_adapters': True}, 'backend_params': {}, 'codebook': None, 'adaptive_codebook_params': {'value_type': 'f8e4m3', 'across_blocks': False, 'num_elements': 16}}" />
158
+ <all_layers value="False" />
159
+ <awq value="False" />
160
+ <backup_mode value="int8_asym" />
161
+ <compression_format value="dequantize" />
162
+ <gptq value="False" />
163
+ <group_size value="-1" />
164
+ <ignored_scope value="[]" />
165
+ <lora_correction value="False" />
166
+ <mode value="int8_sym" />
167
+ <ratio value="1.0" />
168
+ <scale_estimation value="False" />
169
+ <sensitivity_metric value="weight_quantization_error" />
170
+ </weight_compression>
171
+ </nncf>
172
+ <optimum>
173
+ <nncf_version value="3.2.0" />
174
+ <optimum_intel_version value="2.1.0.dev0+109314c" />
175
+ <optimum_version value="2.2.0.dev0" />
176
+ <pytorch_version value="2.11.0+cu128" />
177
+ <transformers_version value="5.2.0" />
178
+ </optimum>
179
+ </rt_info>
180
+ </net>
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87a7830d63fcf43bf241c3c5242e96e62dd3fdc29224ca26fed8ea333db72de4
3
+ size 19989343
tokenizer_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
25
+ "split_special_tokens": false,
26
+ "tokenizer_class": "TokenizersBackend",
27
+ "unk_token": null,
28
+ "video_token": "<|video_pad|>",
29
+ "vision_bos_token": "<|vision_start|>",
30
+ "vision_eos_token": "<|vision_end|>"
31
+ }