vkml 0.0.3

High-level Vulkan-based machine learning library
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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172

use onnx_extractor::DataType;

use crate::{instruction, tensor::TensorDesc, utils::error::VKMLError};

use super::{Layer, execution::LayerExecution};

pub struct LinearLayer {
    pub in_features: i64,
    pub out_features: i64,
    pub bias: bool,
}

impl LinearLayer {
    pub fn new(in_features: i64, out_features: i64) -> Self {
        Self {
            in_features,
            out_features,
            bias: true,
        }
    }

    pub fn new_with(in_features: i64, out_features: i64, bias: bool) -> Self {
        Self {
            in_features,
            out_features,
            bias,
        }
    }
}

impl Layer for LinearLayer {
    fn output_shapes(
        &self,
        batch_size: i64,
        input_shapes: &[&TensorDesc],
    ) -> Result<Vec<TensorDesc>, VKMLError> {
        if input_shapes.len() != 1 {
            return Err(VKMLError::Layer(format!(
                "Linear layer requires exactly 1 input, got {}",
                input_shapes.len()
            )));
        }

        let input_shape = input_shapes[0];

        // Check if it's a 2D tensor (batch, features)
        if input_shape.ndim() != 2 {
            return Err(VKMLError::Layer(format!(
                "Linear layer requires matrix input, got tensor with {} dimensions",
                input_shape.ndim()
            )));
        }

        let cols = input_shape.dims()[1];
        if cols != self.in_features {
            return Err(VKMLError::Layer(format!(
                "Linear layer expected {} input features, got {}",
                self.in_features, cols
            )));
        }

        // Output shape is [batch_size, out_features]
        Ok(vec![TensorDesc::new(
            vec![batch_size, self.out_features],
            DataType::Float,
        )])
    }

    fn parameter_shapes(&self, _input_shapes: &[&TensorDesc]) -> Option<(TensorDesc, TensorDesc)> {
        let weights = TensorDesc::new(vec![self.out_features, self.in_features], DataType::Float);
        let biases = TensorDesc::new(vec![self.out_features], DataType::Float);

        Some((weights, biases))
    }

    fn parameter_count(&self, _batch_size: i64, _input_shapes: &[&TensorDesc]) -> i64 {
        let weight_params = self.in_features * self.out_features;
        let bias_params = if self.bias { self.out_features } else { 0 };

        weight_params + bias_params
    }

    fn input_requirements(&self) -> (usize, Option<usize>) {
        (1, Some(1))
    }

    fn name(&self) -> String {
        "Linear".to_string()
    }

    fn config_string(&self) -> Option<String> {
        if self.bias {
            Some("bias=true".to_string())
        } else {
            Some("bias=false".to_string())
        }
    }

    fn in_features(&self) -> i64 {
        self.in_features
    }

    fn out_features(&self) -> i64 {
        self.out_features
    }

    fn build_layer_exec(
        &self,
        batch_size: i64,
        input_shapes: &[&TensorDesc],
    ) -> Result<LayerExecution, VKMLError> {
        if input_shapes.is_empty() {
            return Err(VKMLError::Layer(
                "LinearLayer requires at least one input".to_string(),
            ));
        }

        let input_shape = input_shapes[0]; // Use the first input shape

        if input_shape.ndim() != 2 {
            return Err(VKMLError::Layer("Linear layer expects matrix input".into()));
        }

        let cols = input_shape.dims()[1];
        if cols != self.in_features {
            return Err(VKMLError::Layer(format!(
                "Linear layer expects {} input features, got {}",
                self.in_features, cols
            )));
        }

        let mut tensors = Vec::new();
        let mut instructions = Vec::new();

        // input = 0
        tensors.push(input_shape.clone());

        // weights = 1
        tensors.push(TensorDesc::new(
            vec![self.out_features, self.in_features],
            DataType::Float,
        ));

        // output = 2
        tensors.push(TensorDesc::new(
            vec![batch_size, self.out_features],
            DataType::Float,
        ));

        // Create MatMul instruction
        instructions.push(instruction::matmul(0, 1, 2));

        // If using bias, add it
        if self.bias {
            // bias = 3
            tensors.push(TensorDesc::new(vec![self.out_features], DataType::Float));

            // TODO: Add back add_inplace instruction
            //instructions.push(instruction::add_inplace(2, 3));
        }

        // Get input mappings using the trait method
        let input_mappings = self.map_input_tensors(input_shapes.len());

        Ok(LayerExecution {
            tensors,
            instructions,
            outputs: vec![2],
            input_mappings,
        })
    }
}