You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
**SKaiNET** is an open-source deep learning framework written in Kotlin Multiplatform, designed with developers in mind to enable the creation modern AI powered applications with ease.
7
7
8
+
## Key features at a glance
9
+
10
+
### SKaiNET is Data
11
+
12
+
- Data loaders (MNIST, JSON datasets, image helpers)
13
+
- Type‑safe tensors across JVM/JS/Native
14
+
15
+
```kotlin
16
+
// data loaders
17
+
val ds =MNIST.load(train =true)
18
+
val (x, y) = ds.nextBatch(64)
19
+
20
+
// Type-safe tensor creation via tensor DSL
21
+
val ctx =DefaultNeuralNetworkExecutionContext()
22
+
val mask = data<FP32, Float>(ctx) {
23
+
tensor{
24
+
shape(3, 3) {
25
+
from(
26
+
1f, 0f, 0f,
27
+
1f, 1f, 0f,
28
+
1f, 1f, 1f,
29
+
)
30
+
}
31
+
}
32
+
}
33
+
34
+
val t = tensor<FP32, Float>(ctx, FP32::class) {
35
+
tensor {
36
+
shape(2, 3) {
37
+
from(
38
+
0f, 1f, 2f,
39
+
10f, 11f, 12f
40
+
)
41
+
}
42
+
}
43
+
}
44
+
println("shape=${t.shape} first=${t.data[0,0]}")
45
+
```
46
+
47
+
### SKaiNET is Language
48
+
49
+
- Kotlin DSLs for Data, Neural Nets, and Pipelines
50
+
51
+
```kotlin
52
+
// Neural network DSL
53
+
val model = nn {
54
+
input(28*28)
55
+
dense(out=128)
56
+
relu()
57
+
dense(out=10)
58
+
}
59
+
```
60
+
61
+
### SKaiNET is Tools
62
+
63
+
- Kotlin Notebook support Explorer and Notebook-friendly APIs
64
+
65
+
```kotlin
66
+
// Works smoothly in Kotlin Notebooks
67
+
display(model.summary())
68
+
println(ds.describe())
69
+
```
70
+
71
+
### SKaiNET is Compiler
72
+
73
+
- MLIR/StableHLO based lowering (modules provided in `SKaiNET-compile-*`)
74
+
75
+
```kotlin
76
+
// Illustrative: export graph to JSON/StableHLO IR
77
+
val ir =Compile.toStableHlo(model)
78
+
println(ir.pretty())
79
+
```
80
+
81
+
### SKaiNET is for Developers
82
+
83
+
- Clean APIs, growing docs, Maven Central artifacts
0 commit comments