oarust 0.8.6

GPU-first semantic computing engine written in Rust
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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221

// @generated by tool/gen/fn/generate.py; DO NOT EDIT.
// schema=ml_training.json schema_version=1 generator_version=64
// schema_sha256=6678961ee57f24310158676c7b4fe0821e21ca48db7edcb20d2a749a164976ae
// Saved-state and backward recipes; lifecycle remains tape-owned.
use crate::ml::autograd::node::operation::BackwardStatus;
use crate::ml::autograd::tape::backward::accumulate_value_gradient;
use crate::ml::matrix as ml_matrix;
use crate::{Matrix, Result};
use std::collections::HashMap;

pub(in crate::ml::autograd) struct GradNodeGelu {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeGelu {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self { input, output_id } = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::gelu_backward(input, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeSilu {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeSilu {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self { input, output_id } = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::silu_backward(input, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeRelu {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeRelu {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self {
			input,
			output,
			output_id,
		} = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::relu_backward(output, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeTanh {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeTanh {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self {
			input,
			output,
			output_id,
		} = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::tanh_backward(output, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeSigmoid {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeSigmoid {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self {
			input,
			output,
			output_id,
		} = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::sigmoid_backward(output, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeLeakyRelu {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) alpha: f32,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeLeakyRelu {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self {
			input,
			alpha,
			output_id,
		} = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::leaky_relu_backward(input, &output_gradient, *alpha)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeElu {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output: Matrix,
	pub(in crate::ml::autograd) alpha: f32,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeElu {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self {
			input,
			output,
			alpha,
			output_id,
		} = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::elu_backward(output, &output_gradient, *alpha)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeMish {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeMish {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self { input, output_id } = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::mish_backward(input, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}

pub(in crate::ml::autograd) struct GradNodeSoftplus {
	pub(in crate::ml::autograd) input: Matrix,
	pub(in crate::ml::autograd) output: Matrix,
	pub(in crate::ml::autograd) output_id: u64,
}

impl GradNodeSoftplus {
	pub(in crate::ml::autograd) fn backward(
		&self,
		gradients: &mut HashMap<u64, Matrix>,
	) -> Result<BackwardStatus> {
		let Self {
			input,
			output,
			output_id,
		} = self;
		let Some(output_gradient) = gradients.remove(output_id) else {
			return Ok(BackwardStatus::Skipped);
		};
		let input_gradient = ml_matrix::softplus_backward(output, &output_gradient)?;
		accumulate_value_gradient(gradients, input.value_id(), input_gradient)?;
		Ok(BackwardStatus::Recorded)
	}
}