File size: 4,797 Bytes
67fe046
 
 
 
dd9d28c
67fe046
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
806ff63
67fe046
806ff63
 
 
 
 
 
 
67fe046
e8daec3
67fe046
 
806ff63
 
 
 
 
 
 
67fe046
 
 
 
 
 
 
 
 
 
 
b878b86
6f6b163
b878b86
 
67fe046
 
b878b86
 
 
 
67fe046
b878b86
 
67fe046
b878b86
 
 
 
67fe046
 
 
 
6f6b163
 
b878b86
 
 
 
 
 
 
 
67fe046
b878b86
 
67fe046
b878b86
67fe046
b878b86
 
67fe046
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e8daec3
 
67fe046
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
# πŸ›°οΈ ZETIC.ai β€” On-Device AI for Every Device

**Build. Deploy. Run. Anywhere.**  
ZETIC.ai helps AI engineers deploy models on *any* mobile device β€” without cloud GPU servers.  
We transform your existing AI models into **NPU-optimized, on-device runtimes** in **under 6 hours** including from global device benchmark to runtime source code generation.

---

## πŸš€ What We Do

**ZETIC.MLange** β€” our core platform β€” enables **serverless AI** by:
- **Automated Conversion**: Convert your PyTorch, ONNX, or TFLite model into a device-specific NPU library.
- **Peak Performance**: Up to **60Γ— faster** than GPU cloud inference, with zero accuracy loss.
- **Broad Compatibility**: Supports Android, iOS, Linux; MediaTek, Qualcomm, Apple NPUs β€” more coming soon.
- **End-to-End SDK**: From model optimization to app integration β€” no extra engineering required.

---

## πŸ›  Key Features

- **Zero GPU Costs** β€” Replace expensive GPU cloud servers with *free* NPU power in devices.
- **Full Privacy & Security** β€” Data never leaves the device.
- **Ultra-Low Latency** β€” Real-time AI experiences, even offline.
- **Cross-Platform** β€” One model β†’ All devices β†’ Same performance.

---

## πŸ“¦ Example Use Cases

- πŸŽ™ **Speech Recognition (Whisper)** β€” Real-time, offline transcription on mobile.
- 🦷 **Dental AI Diagnostics** β€” Instant tooth condition analysis via smartphone camera.
- 🏌️ **Sports AI** β€” On-device golf swing analytics.
- πŸ€– **On-Device LLMs** β€” Chat & reasoning models running entirely offline.

---

## πŸ“Š Benchmarks

| Device | Task | Cloud GPU | On-Device NPU | Speedup |
|--------|------|-----------|---------------|---------|
| iPhone 16 Pro | Whisper-Small | 1.2s | 0.07s | **Γ—17** |
| Galaxy S24 Ultra | LLaMA-3-8B | 2.4s/token | 0.09s/token | **Γ—26** |

[πŸ”— See more benchmarks Β»](https://mlange.zetic.ai)



### YOLOv8n β€” NPU Latency (ms)
| Device | Manufacturer | CPU | GPU | CPU/GPU | NPU |
|--------|--------------|-----|-----|---------|-----|
| Apple iPhone 16 | Apple | 126.27 | - | 8.98 | **2.03** |
| Apple iPhone 16 Pro | Apple | 122.23 | - | 7.54 | **1.69** |
| Samsung Galaxy S24+ | Qualcomm | 69.79 | 24.38 | 618.05 | **3.85** |
| Samsung Galaxy Tab S9 | Qualcomm | 107.78 | 30.39 | 344.42 | **5.21** |
| Samsung Galaxy S22 Ultra 5G | Qualcomm | 103.40 | 39.73 | 100.34 | **7.41** |

---

### Whisper-tiny-encoder β€” NPU Latency (ms)
| Device | Manufacturer | CPU | GPU | CPU/GPU | NPU |
|--------|--------------|-----|-----|---------|-----|
| Apple iPhone 16 | Apple | 552.13 | - | 44.49 | **19.01** |
| Apple iPhone 15 Pro | Apple | 527.78 | - | 43.13 | **19.40** |
} Samsung Galaxy S23  | Qualcomm | 290.62ms | 169.82ms | 2,795.18ms | **86.88** |
| Samsung Galaxy S24+ | Qualcomm | 278.78 | 133.48 | 2619.56 | **106.44** |
| Samsung Galaxy S23 Ultra | Qualcomm | 308.82 | 170.08 | 2688.97 | **68.34** |


- **You can get runtime source code and benchmark report of your model with [ZETIC.MLange](https://mlange.zetic.ai)**

---

## πŸ‘¨πŸ»β€πŸ’» Plug-and-play To Your App

- The runtime SDK is also provided for your AI model with ZETIC.MLange

- **iOS Integration** (Swift)

``` swift

// import
import ZeticMLange

// ...

// (1) Load Zetic MLange model
let model = try ZeticMLangeModel("MLANGE_PROJECT_API_KEY")

// (2) Run model after preparing model inputs
let inputs: [Data] = [] // Prepare your inputs

try model.run(inputs)

// (3) Get output data array
let outputs = model.getOutputDataArray()

```

- **Android Integration** (Kotlin, Java)

``` kotlin
// import
import com.zeticai.mlange.core.model.Target
import com.zeticai.mlange.core.model.ZeticMLangeModel

// ...

// (1) Load Zetic MLange model
val model = ZeticMLangeModel(this, "MLANGE_PROJECT_API_KEY")

// (2) Run model after preparing model inputs
val inputs: Array<ByteBuffer> = // Prepare your inputs

model.run(inputs)

// (3) Get output buffers of the model
val outputs = model.outputBuffers
```


## πŸ“₯ Try It Now

- **MLange Dashboard**: [https://mlange.zetic.ai](https://mlange.zetic.ai)  
- **Demo Apps**: [App Store](https://apps.apple.com/app/zeticapp/id6739862746) / [Google Play](https://play.google.com/store/apps/details?id=com.zeticai.zeticapp)

---

## 🧭 Supported Targets

- **OS**: Android, iOS, Linux
- **NPUs**: MediaTek, Qualcomm, Apple (more coming)
- **Frameworks In**: PyTorch, ONNX, TFLite
- **Artifacts Out**: NPU-optimized runtime libraries + SDK bindings (Kotlin, Java, Swift, Flutter, React Native)

## πŸ“¬ Contact Us

- **Website**: [https://zetic.ai](https://zetic.ai)
- **Email**: [email protected]
- **LinkedIn**: [linkedin.com/company/zetic-ai](https://linkedin.com/company/zetic-ai)

---

**ZETIC.ai** β€” AI for All, Anytime, Anywhere.  
Run your AI where it matters: **on the device.**