kgrabko commited on
Commit
51df9a4
Β·
verified Β·
1 Parent(s): dd9c19b

README: v2 layout (vocab 128256, tags in reserved slots, think/tool tokens)

Browse files
Files changed (1) hide show
  1. README.md +296 -269
README.md CHANGED
@@ -1,269 +1,296 @@
1
- ---
2
- license: apache-2.0
3
- language:
4
- - multilingual
5
- tags:
6
- - tokenizer
7
- - bpe
8
- - byte-level-bpe
9
- - chatml
10
- - routing
11
- - moe
12
- - robotics
13
- - jiarck
14
- - llama
15
- - llama3
16
- - meta
17
- ---
18
-
19
- # πŸ’Ž Llama Robotics & Routing Tokenizer
20
- - Llama Pro Tokenizer from JiRack team . Fully compatible with Llama3 models from META Inc.
21
- - Size {
22
- Vocab size: 128365
23
- pad_token_id: 128001
24
- eos_token_id: 128001
25
- }
26
- - Compatible with Llama 3 β€’ Optimized for Code . Use resize function without adaptation, see examples below
27
- - It needs 100k example to fully adapt routing features for RAG Routing model . So check llama modification tokenizer rules
28
- - A Llama 3-based tokenizer enhanced with FIM markers from Microsoft datasets (
29
- <|fim_prefix|>,
30
- <|fim_middle|>,
31
- <f|im_suffix|>
32
- )
33
- - Added Robotics & Embodiment tags (
34
- "<|action_start|>",
35
- "<|action_end|>",
36
- "<|trajectory_start|>",
37
- "<|trajectory_end|>",
38
- "<|joint_start|>",
39
- "<|joint_end|>",
40
- "<|sensor_start|>",
41
- "<|sensor_end|>",
42
- "<|command_start|>",
43
- "<|command_end|>",
44
- "<|state_start|>",
45
- "<|state_end|>",
46
- "<|pose|>",
47
- "<|velocity|>",
48
- "<|force|>",
49
- "<|torque|>",
50
- "<|gripper|>",
51
- "<|navigation|>",
52
- "<|obstacle|>",
53
- "<|task_start|>",
54
- "<|task_end|>",
55
- "<|plan_start|>",
56
- "<|plan_end|>",
57
- "<|behavior_start|>",
58
- "<|behavior_end|>",
59
- "<|skill_start|>",
60
- "<|skill_end|>",
61
- "<|motor|>",
62
- "<|servo|>",
63
- "<|imu|>",
64
- "<|lidar|>",
65
- "<|camera|>",
66
- "<|depth|>",
67
- "<|waypoint|>",
68
- "<|path|>",
69
- "<|collision|>",
70
- "<|grasp|>",
71
- "<|release|>",
72
- "<|homing|>",
73
- "<|emergency_stop|>",
74
- "<|calibration|>",
75
- "<|manipulation|>",
76
- "<|locomotion|>",
77
- "<|feedback|>",
78
- "<|control_loop|>",)
79
-
80
- - Added Multi models support (
81
- "<|image|>",
82
- "<|video|>",
83
- "<|sound|>",
84
- "<|voice|>",
85
- "<|listening|>",
86
- "<|vision|>",)
87
-
88
- - Added Human mood tags (
89
- "<|mood_happy|>",
90
- "<|mood_sad|>",
91
- "<|mood_angry|>",
92
- "<|mood_neutral|>",
93
- )
94
-
95
- - Added RAG routing tags for RAG MoE Systems (
96
- "__SCIENCE__",
97
- "__CODING__",
98
- "__STOCK_EXCHANGE__",
99
- "__MEDICINE__",
100
- "__GOVERNMENT__",
101
- "__NEWS__",
102
- "__GENERAL__",
103
- "__MATERIAL_SCIENCE__",
104
- "__ELECTRONICS__",
105
- "__MICROELECTRONICS__",
106
- "__ENGINEERING__",
107
- "__ROBOTICS__",
108
- "__ENERGY__",
109
- "__AUTOMOTIVE__",
110
- "__AVIATION__",
111
- "__MATH__",
112
- "__PYTHON__",
113
- "__C__",
114
- "__CPP__",
115
- "__C_SHARP__",
116
- "__JAVA__",
117
- "__JAVASCRIPT__",
118
- "__TYPESCRIPT__",
119
- "__RUST__",
120
- "__GO__",
121
- "__RUBY__",
122
- "__PHP__",
123
- "__SWIFT__",
124
- "__KOTLIN__",
125
- "__BASH__",
126
- "__SQL__",
127
- "__ASSEMBLY__",
128
- "__PHILOSOPHY__",
129
- "__LITERATURE__",
130
- "__SOCIOLOGY__",
131
- "__PSYCHOLOGY__",
132
- "__POLITICAL_SCIENCE__",
133
- "__CULTURAL_STUDIES__",
134
- "__ETHNOGRAPHY__",
135
- "__HUMAN_RIGHTS__",
136
- "__COMPLIANCE__",
137
- "__MILITARY__",
138
- "__BANKING__",
139
- "__OIL_INDUSTRY__",
140
- "__LIGHT_INDUSTRY__",
141
- "__NATURE__",
142
- "__OCEAN__",
143
- "__SPORT__",
144
- "__CULINARY__",
145
- "__TRAVEL__",
146
- "__HOBBY__"
147
- )
148
-
149
- - Fully compatible with Microsoft BigCode datasets including The Stack, StarCoder, and NextCoder.
150
- - Enables efficient training on large-scale coding data for superior code generation and understanding.
151
-
152
-
153
- **Inventor:** Konstantin Vladimirovich Grabko
154
- **Organization:** CMS Manhattan JiRack Technology
155
- **Official Site:** [www.cmsmanhattan.com](http://www.cmsmanhattan.com)
156
- Designed for Banking and Fintech Institutions
157
-
158
- **Banks and Fintech** JiRack Architecture: Build Sovereign Financial Models from Scratch
159
- - Leveraging the JiRack Tokenizer and our Open Dataset, we enable financial institutions to develop secure, internal AI models from the ground up. This approach ensures maximum data privacy and model sovereignty for high-stakes banking operations.
160
- - There is fix price for FinTech
161
- - I recommend initializing the model with a 4K context window for initial stability, followed by scaling to 8K context using specialized JiRack 8K datasets. This two-stage approach ensures robust positional encoding before extending the model's long-range dependency.
162
-
163
-
164
- **JiRack Corp Tokenizer solution**
165
- - Use JiRack models with trusted, high-quality coding datasets while maintaining full control over your code and data privacy.
166
- - Excellent fit for Banks, Fintech companies, and any organization that requires strict data confidentiality and security.
167
- - Update JiRack model for corp privacy coding.
168
-
169
- **JiRack Tokenizer Subcription**
170
- - All subscribed members will receive regular tokenizer updates optimized for the latest high-quality coding datasets.
171
-
172
- ### Open Robot platform
173
-
174
- - **Tiangong** : https://english.www.gov.cn/english.www.gov.cn/news/202411/13/content_WS673406e2c6d0868f4e8ece33.html
175
- - **Unitree g1** https://a.co/d/0e4A8YVc
176
- - **LimX Oli** https://www.limxdynamics.com/en/products/oli?channel=option_google_advertising__c-
177
- - **ubtrobot** https://www.ubtrobot.com/en/
178
- - **x-humanoid** https://www.x-humanoid.com/detail/hskw.html
179
-
180
-
181
- ## ROBOS HUMANOID Project
182
-
183
- - https://github.com/cmsmanhattan/ROBOS
184
-
185
-
186
- ### Robot dataset example
187
-
188
- ```bash
189
- {"messages":[{"role":"system","content":"You are the robot's brain. Use tags and . ROBOTICS"},{"role":"user","content":"Scene: a table in front of me, with a red mug on it. Command: pick up the mug."},{"role":"assistant","content":"reach(x=0.4,y=0.0,z=0.3)"}]}
190
- {"messages":[{"role":"system","content":"You are the robot's brain. Use tags and .ROBOTICS"},{"role":"user","content":"Scene: hand near the mug, distance 2 cm. Command: grasp."},{"role":"assistant","content":"grasp(force=60)"}]}
191
- {"messages":[{"role":"system","content":"You are the robot's brain. Use tags and . ROBOTICS"},{"role":"user","content":"Scene: mug in hand, stable. Command: lift 20 cm."},{"role":"assistant","content":"lift(z=0.2)"}]}
192
-
193
- ```
194
-
195
- ```bash
196
- {
197
- "messages": [
198
- {
199
- "role": "system",
200
- "content": "You are a robot. Use tags and . ROBOTICS"
201
- },
202
- {
203
- "role": "user",
204
- "content": "Scene: a table in front of me, with a red mug on it. Command: pick up the mug."
205
- },
206
- {
207
- "role": "assistant",
208
- "content": "reach(x=0.4,y=0.0,z=0.3)"
209
- }
210
- ]
211
- }
212
- ```
213
-
214
- ### Key Features
215
-
216
- - **Algorithm**: Byte-Level BPE
217
- - **Vocabulary Size**: **128,000** tokens β€” excellent balance between precision and efficiency
218
- - **Multilingual & Technical Strength**: Optimized for English, Russian, code, scientific literature, and technical documentation
219
- - **Domain Specialization**: Strong performance on programming languages, engineering, robotics, and scientific texts
220
-
221
- ### Special Tokens Support
222
-
223
- - Full **Llama3 compatible format** dialogue format
224
- - FIM (Fill-in-the-Middle) support for code generation
225
- - Rich set of domain routing tokens (`__CODING__`, `__PYTHON__`, `__ROBOTICS__`, `__SCIENCE__`, etc.)
226
- - Extended robotics and control tokens
227
-
228
- ### CMS Manhattan Service & Support
229
-
230
- - Jirack patent guards your technology for competitors
231
- - Redesign Llama , Qwen , Gemma to Ternary model
232
- - Re-tain and replace embeddings for Llama , Qwen , Gemma to extend langeages to 347
233
- - Accelerate inference via high compression 256K tokenizer and replace multiplication with sum operations via Ternary weights
234
-
235
- **Install for Llamma compatible models in your chat script**
236
-
237
- - from transformers import AutoModelForCausalLM
238
-
239
- - model = AutoModelForCausalLM.from_pretrained(your model)
240
-
241
- - # The Must !
242
- - model.resize_token_embeddings(len(tokenizer.tokenizer)) # ΠΈΠ»ΠΈ просто len(tokenizer.tokenizer)
243
-
244
- - print("New Embedding size for you chat script:", model.get_input_embeddings().weight.shape[0])
245
-
246
- - # Tesr Tokenizer size !
247
- (venv_ji) root@jirack2:# python -c '
248
- from transformers import AutoTokenizer
249
- tok = AutoTokenizer.from_pretrained("./jirack_code_tokenizer_fixed")
250
- print("Vocab size:", len(tok))
251
- print("pad_token_id:", tok.pad_token_id)
252
- print("eos_token_id:", tok.eos_token_id)
253
- '
254
-
255
- Vocab size: 128365
256
- pad_token_id: 128001
257
- eos_token_id: 128001
258
-
259
-
260
- ## πŸ“§ Contact & Licensing
261
-
262
- For joint ventures, hardware integration, or licensing inquiries:
263
-
264
- - **Email:** grabko@cmsmanhattan.com
265
- - **Phone:** +1 (516) 777-0945
266
- - **Location:** New York, USA
267
-
268
- ## πŸ“§ Copyright 2026 CMS Manhattan . All rights reserved
269
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - multilingual
5
+ tags:
6
+ - tokenizer
7
+ - bpe
8
+ - byte-level-bpe
9
+ - chatml
10
+ - routing
11
+ - moe
12
+ - robotics
13
+ - jiarck
14
+ - llama
15
+ - llama3
16
+ - meta
17
+ ---
18
+
19
+ # πŸ’Ž Llama Robotics & Routing Tokenizer
20
+ - Llama Pro Tokenizer from JiRack team . Fully compatible with Llama3 models from META Inc.
21
+ - Size {
22
+ Vocab size: 128256 (same as Llama 3, no embedding resize needed)
23
+ bos_token_id: 128000 (<|begin_of_text|>)
24
+ eos_token_id: 128009 (<|eot_id|>)
25
+ }
26
+ - Compatible with Llama 3 β€’ Optimized for Code . All tags live in Llama 3 reserved slots, so the vocabulary keeps the native 128256 size
27
+ - It needs 100k example to fully adapt routing features for RAG Routing model . So check llama modification tokenizer rules
28
+ - A Llama 3-based tokenizer enhanced with FIM markers from Microsoft datasets (
29
+ <|fim_prefix|>,
30
+ <|fim_middle|>,
31
+ <f|im_suffix|>
32
+ )
33
+ - Added Robotics & Embodiment tags (
34
+ "<|action_start|>",
35
+ "<|action_end|>",
36
+ "<|trajectory_start|>",
37
+ "<|trajectory_end|>",
38
+ "<|joint_start|>",
39
+ "<|joint_end|>",
40
+ "<|sensor_start|>",
41
+ "<|sensor_end|>",
42
+ "<|command_start|>",
43
+ "<|command_end|>",
44
+ "<|state_start|>",
45
+ "<|state_end|>",
46
+ "<|pose|>",
47
+ "<|velocity|>",
48
+ "<|force|>",
49
+ "<|torque|>",
50
+ "<|gripper|>",
51
+ "<|navigation|>",
52
+ "<|obstacle|>",
53
+ "<|task_start|>",
54
+ "<|task_end|>",
55
+ "<|plan_start|>",
56
+ "<|plan_end|>",
57
+ "<|behavior_start|>",
58
+ "<|behavior_end|>",
59
+ "<|skill_start|>",
60
+ "<|skill_end|>",
61
+ "<|motor|>",
62
+ "<|servo|>",
63
+ "<|imu|>",
64
+ "<|lidar|>",
65
+ "<|camera|>",
66
+ "<|depth|>",
67
+ "<|waypoint|>",
68
+ "<|path|>",
69
+ "<|collision|>",
70
+ "<|grasp|>",
71
+ "<|release|>",
72
+ "<|homing|>",
73
+ "<|emergency_stop|>",
74
+ "<|calibration|>",
75
+ "<|manipulation|>",
76
+ "<|locomotion|>",
77
+ "<|feedback|>",
78
+ "<|control_loop|>",)
79
+
80
+ - Added Multi models support (
81
+ "<|image|>",
82
+ "<|video|>",
83
+ "<|sound|>",
84
+ "<|voice|>",
85
+ "<|listening|>",
86
+ "<|vision|>",)
87
+
88
+ - Added Human mood tags (
89
+ "<|mood_happy|>",
90
+ "<|mood_sad|>",
91
+ "<|mood_angry|>",
92
+ "<|mood_neutral|>",
93
+ )
94
+
95
+ - Added RAG routing tags for RAG MoE Systems (
96
+ "__SCIENCE__",
97
+ "__CODING__",
98
+ "__STOCK_EXCHANGE__",
99
+ "__MEDICINE__",
100
+ "__GOVERNMENT__",
101
+ "__NEWS__",
102
+ "__GENERAL__",
103
+ "__MATERIAL_SCIENCE__",
104
+ "__ELECTRONICS__",
105
+ "__MICROELECTRONICS__",
106
+ "__ENGINEERING__",
107
+ "__ROBOTICS__",
108
+ "__ENERGY__",
109
+ "__AUTOMOTIVE__",
110
+ "__AVIATION__",
111
+ "__MATH__",
112
+ "__PYTHON__",
113
+ "__C__",
114
+ "__CPP__",
115
+ "__C_SHARP__",
116
+ "__JAVA__",
117
+ "__JAVASCRIPT__",
118
+ "__TYPESCRIPT__",
119
+ "__RUST__",
120
+ "__GO__",
121
+ "__RUBY__",
122
+ "__PHP__",
123
+ "__SWIFT__",
124
+ "__KOTLIN__",
125
+ "__BASH__",
126
+ "__SQL__",
127
+ "__ASSEMBLY__",
128
+ "__PHILOSOPHY__",
129
+ "__LITERATURE__",
130
+ "__SOCIOLOGY__",
131
+ "__PSYCHOLOGY__",
132
+ "__POLITICAL_SCIENCE__",
133
+ "__CULTURAL_STUDIES__",
134
+ "__ETHNOGRAPHY__",
135
+ "__HUMAN_RIGHTS__",
136
+ "__COMPLIANCE__",
137
+ "__MILITARY__",
138
+ "__BANKING__",
139
+ "__OIL_INDUSTRY__",
140
+ "__LIGHT_INDUSTRY__",
141
+ "__NATURE__",
142
+ "__OCEAN__",
143
+ "__SPORT__",
144
+ "__CULINARY__",
145
+ "__TRAVEL__",
146
+ "__HOBBY__"
147
+ )
148
+
149
+ - Fully compatible with Microsoft BigCode datasets including The Stack, StarCoder, and NextCoder.
150
+ - Enables efficient training on large-scale coding data for superior code generation and understanding.
151
+
152
+
153
+ **Inventor:** Konstantin Vladimirovich Grabko
154
+ **Organization:** CMS Manhattan JiRack Technology
155
+ **Official Site:** [www.cmsmanhattan.com](http://www.cmsmanhattan.com)
156
+ Designed for Banking and Fintech Institutions
157
+
158
+ **Banks and Fintech** JiRack Architecture: Build Sovereign Financial Models from Scratch
159
+ - Leveraging the JiRack Tokenizer and our Open Dataset, we enable financial institutions to develop secure, internal AI models from the ground up. This approach ensures maximum data privacy and model sovereignty for high-stakes banking operations.
160
+ - There is fix price for FinTech
161
+ - I recommend initializing the model with a 4K context window for initial stability, followed by scaling to 8K context using specialized JiRack 8K datasets. This two-stage approach ensures robust positional encoding before extending the model's long-range dependency.
162
+
163
+
164
+ **JiRack Corp Tokenizer solution**
165
+ - Use JiRack models with trusted, high-quality coding datasets while maintaining full control over your code and data privacy.
166
+ - Excellent fit for Banks, Fintech companies, and any organization that requires strict data confidentiality and security.
167
+ - Update JiRack model for corp privacy coding.
168
+
169
+ **JiRack Tokenizer Subcription**
170
+ - All subscribed members will receive regular tokenizer updates optimized for the latest high-quality coding datasets.
171
+
172
+ ### Open Robot platform
173
+
174
+ - **Tiangong** : https://english.www.gov.cn/english.www.gov.cn/news/202411/13/content_WS673406e2c6d0868f4e8ece33.html
175
+ - **Unitree g1** https://a.co/d/0e4A8YVc
176
+ - **LimX Oli** https://www.limxdynamics.com/en/products/oli?channel=option_google_advertising__c-
177
+ - **ubtrobot** https://www.ubtrobot.com/en/
178
+ - **x-humanoid** https://www.x-humanoid.com/detail/hskw.html
179
+
180
+
181
+ ## ROBOS HUMANOID Project
182
+
183
+ - https://github.com/cmsmanhattan/ROBOS
184
+
185
+
186
+ ### Robot dataset example
187
+
188
+ ```bash
189
+ {"messages":[{"role":"system","content":"You are the robot's brain. Use tags and . ROBOTICS"},{"role":"user","content":"Scene: a table in front of me, with a red mug on it. Command: pick up the mug."},{"role":"assistant","content":"reach(x=0.4,y=0.0,z=0.3)"}]}
190
+ {"messages":[{"role":"system","content":"You are the robot's brain. Use tags and .ROBOTICS"},{"role":"user","content":"Scene: hand near the mug, distance 2 cm. Command: grasp."},{"role":"assistant","content":"grasp(force=60)"}]}
191
+ {"messages":[{"role":"system","content":"You are the robot's brain. Use tags and . ROBOTICS"},{"role":"user","content":"Scene: mug in hand, stable. Command: lift 20 cm."},{"role":"assistant","content":"lift(z=0.2)"}]}
192
+
193
+ ```
194
+
195
+ ```bash
196
+ {
197
+ "messages": [
198
+ {
199
+ "role": "system",
200
+ "content": "You are a robot. Use tags and . ROBOTICS"
201
+ },
202
+ {
203
+ "role": "user",
204
+ "content": "Scene: a table in front of me, with a red mug on it. Command: pick up the mug."
205
+ },
206
+ {
207
+ "role": "assistant",
208
+ "content": "reach(x=0.4,y=0.0,z=0.3)"
209
+ }
210
+ ]
211
+ }
212
+ ```
213
+
214
+ ### Version 2 (current): tags in Llama 3 reserved slots
215
+
216
+ Since commit `dd9c19ba` all tags are placed into the Llama 3 reserved special-token slots instead of being appended
217
+ after the vocabulary. The vocabulary stays at **128,256** β€” exactly the Llama 3 embedding size, so any Llama 3 model
218
+ loads it without `resize_token_embeddings`.
219
+
220
+ | IDs | Tokens |
221
+ |---|---|
222
+ | 128141 – 128146 | multimodal: `<\|image\|>` … `<\|vision\|>` |
223
+ | 128147 – 128150 | mood: `<\|mood_happy\|>` … `<\|mood_neutral\|>` |
224
+ | 128151 – 128153 | FIM: `<fim_prefix>`, `<fim_middle>`, `<fim_suffix>` |
225
+ | 128154 – 128198 | robotics: `<\|action_start\|>` … `<\|control_loop\|>` |
226
+ | 128199 – 128249 | RAG routing: `__SCIENCE__` … `__HOBBY__` |
227
+ | 128250 / 128251 | `<think>` / `</think>` |
228
+ | 128252 / 128253 | `<tool_call>` / `</tool_call>` |
229
+ | 128254 / 128255 | `<tool_response>` / `</tool_response>` |
230
+
231
+ Used by **Llama3Ternary 2B** ([CMSManhattan/Llama3TernaryPro_2b](https://huggingface.co/CMSManhattan/Llama3TernaryPro_2b)).
232
+
233
+ The previous version (vocab 128,365, tags appended at IDs 128256+, needs `resize_token_embeddings`) is still
234
+ available as revision `703ce8f1`:
235
+
236
+ ```python
237
+ tok = AutoTokenizer.from_pretrained("CMSManhattan/LlamaRoboticsTokenizer", revision="703ce8f1")
238
+ ```
239
+
240
+ ### Key Features
241
+
242
+ - **Algorithm**: Byte-Level BPE
243
+ - **Vocabulary Size**: **128,256** tokens β€” excellent balance between precision and efficiency
244
+ - **Multilingual & Technical Strength**: Optimized for English, Russian, code, scientific literature, and technical documentation
245
+ - **Domain Specialization**: Strong performance on programming languages, engineering, robotics, and scientific texts
246
+
247
+ ### Special Tokens Support
248
+
249
+ - Full **Llama3 compatible format** dialogue format
250
+ - FIM (Fill-in-the-Middle) support for code generation
251
+ - Rich set of domain routing tokens (`__CODING__`, `__PYTHON__`, `__ROBOTICS__`, `__SCIENCE__`, etc.)
252
+ - Extended robotics and control tokens
253
+
254
+ ### CMS Manhattan Service & Support
255
+
256
+ - Jirack patent guards your technology for competitors
257
+ - Redesign Llama , Qwen , Gemma to Ternary model
258
+ - Re-tain and replace embeddings for Llama , Qwen , Gemma to extend langeages to 347
259
+ - Accelerate inference via high compression 256K tokenizer and replace multiplication with sum operations via Ternary weights
260
+
261
+ **Install for Llamma compatible models in your chat script**
262
+
263
+ - from transformers import AutoModelForCausalLM
264
+
265
+ - model = AutoModelForCausalLM.from_pretrained(your model)
266
+
267
+ - # Version 2: no resize needed for Llama 3 models (vocab 128256). Only for the old revision 703ce8f1:
268
+ - # model.resize_token_embeddings(len(tokenizer))
269
+
270
+ - print("Embedding size:", model.get_input_embeddings().weight.shape[0])
271
+
272
+ - # Tesr Tokenizer size !
273
+ (venv_ji) root@jirack2:# python -c '
274
+ from transformers import AutoTokenizer
275
+ tok = AutoTokenizer.from_pretrained("CMSManhattan/LlamaRoboticsTokenizer")
276
+ print("Vocab size:", len(tok))
277
+ print("bos_token_id:", tok.bos_token_id)
278
+ print("eos_token_id:", tok.eos_token_id)
279
+ print("<tool_call>:", tok.convert_tokens_to_ids("<tool_call>"))
280
+ '
281
+
282
+ Vocab size: 128256
283
+ bos_token_id: 128000
284
+ eos_token_id: 128009
285
+ <tool_call>: 128252
286
+
287
+
288
+ ## πŸ“§ Contact & Licensing
289
+
290
+ For joint ventures, hardware integration, or licensing inquiries:
291
+
292
+ - **Email:** grabko@cmsmanhattan.com
293
+ - **Phone:** +1 (516) 777-0945
294
+ - **Location:** New York, USA
295
+
296
+ ## πŸ“§ Copyright 2026 CMS Manhattan . All rights reserved