attnres 0.1.1

Attention Residuals (MoonshotAI/Kimi) implementation in Rust using burn
Documentation 
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
222

/// Integration tests for attnres.
///
/// Tests end-to-end model behavior including forward passes,
/// autodiff gradient flow, and model configuration.
use attnres::{causal_mask, AttnResConfig, AttnResTransformer};
use burn::backend::Autodiff;
use burn::backend::NdArray;
use burn::config::Config;
use burn::prelude::*;

type TestBackend = NdArray;
type AutodiffBackend = Autodiff<TestBackend>;

#[test]
fn test_full_model_forward_backward() {
    let device = Default::default();
    let config = AttnResConfig::new(32, 4, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<AutodiffBackend> = config.init_model(&device);

    let input_ids = Tensor::<AutodiffBackend, 2, Int>::zeros([1, 8], &device);
    let logits = model.forward(input_ids, None);

    // Compute a simple loss and backward
    let loss = logits.mean();
    let _grads = loss.backward();
    // If we get here without panic, gradients flow correctly
}

#[test]
fn test_model_with_mask_forward_backward() {
    let device = Default::default();
    let config = AttnResConfig::new(32, 4, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<AutodiffBackend> = config.init_model(&device);

    let mask = causal_mask::<AutodiffBackend>(2, 8, &device);
    let input_ids = Tensor::<AutodiffBackend, 2, Int>::zeros([2, 8], &device);
    let logits = model.forward(input_ids, Some(&mask));
    let loss = logits.mean();
    let _grads = loss.backward();
}

#[test]
fn test_full_attnres_config() {
    // Full AttnRes: num_blocks = num_layers
    let device = Default::default();
    let config = AttnResConfig::new(32, 4, 4)
        .with_num_heads(4)
        .with_vocab_size(50);

    assert!(config.is_full());

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);
    let input = Tensor::<TestBackend, 2, Int>::zeros([1, 4], &device);
    let out = model.forward(input, None);
    assert_eq!(out.dims(), [1, 4, 50]);
}

#[test]
fn test_block_attnres_config() {
    // Block AttnRes: num_blocks < num_layers
    let device = Default::default();
    let config = AttnResConfig::new(32, 8, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    assert!(!config.is_full());
    assert_eq!(config.block_size(), 4);

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);
    let input = Tensor::<TestBackend, 2, Int>::zeros([1, 8], &device);
    let out = model.forward(input, None);
    assert_eq!(out.dims(), [1, 8, 50]);
}

#[test]
fn test_deterministic_forward() {
    let device = Default::default();
    let config = AttnResConfig::new(32, 4, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);

    let input = Tensor::<TestBackend, 2, Int>::zeros([1, 8], &device);

    // Same input should produce same output
    let out1 = model.forward(input.clone(), None);
    let out2 = model.forward(input, None);

    let diff: f32 = (out1 - out2).abs().max().into_scalar();
    assert!(
        diff < 1e-6,
        "Same input should produce identical output, diff={diff}"
    );
}

// ========================
// Two-phase Inference Integration Tests
// ========================

#[test]
fn test_two_phase_matches_standard_integration() {
    // Test with a deeper model to exercise multiple block boundaries
    let device = Default::default();
    let config = AttnResConfig::new(32, 12, 3)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);
    let input = Tensor::<TestBackend, 2, Int>::zeros([2, 8], &device);

    let standard = model.forward(input.clone(), None);
    let two_phase = model.forward_two_phase(input, None);

    let diff: f32 = (standard - two_phase).abs().max().into_scalar();
    assert!(
        diff < 1e-3,
        "Two-phase should match standard for deep model, diff={diff}"
    );
}

#[test]
fn test_two_phase_with_causal_mask() {
    let device = Default::default();
    let config = AttnResConfig::new(32, 8, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);
    let mask = causal_mask::<TestBackend>(1, 8, &device);
    let input = Tensor::<TestBackend, 2, Int>::zeros([1, 8], &device);

    let standard = model.forward(input.clone(), Some(&mask));
    let two_phase = model.forward_two_phase(input, Some(&mask));

    let diff: f32 = (standard - two_phase).abs().max().into_scalar();
    assert!(
        diff < 1e-3,
        "Two-phase with mask should match standard, diff={diff}"
    );
}

#[test]
fn test_two_phase_with_odd_block_size() {
    // 6 sublayers / 2 blocks => block size 3, which forces a boundary inside
    // a transformer layer (between attention and MLP).
    let device = Default::default();
    let config = AttnResConfig::new(32, 6, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);
    let input = Tensor::<TestBackend, 2, Int>::zeros([1, 8], &device);

    let standard = model.forward(input.clone(), None);
    let two_phase = model.forward_two_phase(input, None);

    let diff: f32 = (standard - two_phase).abs().max().into_scalar();
    assert!(
        diff < 1e-3,
        "Two-phase should match standard when block boundaries occur before MLP, diff={diff}"
    );
}

// ========================
// Serialization Integration Tests
// ========================

#[test]
fn test_save_load_preserves_forward() {
    let device = Default::default();
    let config = AttnResConfig::new(32, 4, 2)
        .with_num_heads(4)
        .with_vocab_size(50);

    let model: AttnResTransformer<TestBackend> = config.init_model(&device);
    let input = Tensor::<TestBackend, 2, Int>::zeros([1, 8], &device);
    let out_before = model.forward(input.clone(), None);

    let path = std::env::temp_dir().join("attnres_integ_save_load");
    let path_str = path.to_str().unwrap();
    model.save(path_str, &device).expect("Failed to save");

    let loaded: AttnResTransformer<TestBackend> =
        AttnResTransformer::load(path_str, &config, &device).expect("Failed to load");
    let out_after = loaded.forward(input, None);

    let diff: f32 = (out_before - out_after).abs().max().into_scalar();
    assert!(
        diff < 1e-6,
        "Loaded model should match saved model, diff={diff}"
    );

    let _ = std::fs::remove_file(format!("{path_str}.mpk"));
}

#[test]
fn test_config_save_load_then_init_model() {
    let device = Default::default();
    let config = AttnResConfig::new(32, 8, 4)
        .with_num_heads(4)
        .with_vocab_size(100)
        .with_dropout(0.1);

    let path = std::env::temp_dir().join("attnres_integ_config.json");
    config.save(&path).expect("Failed to save config");

    let loaded_config = AttnResConfig::load(&path).expect("Failed to load config");
    let _ = loaded_config.init_model::<TestBackend>(&device);

    assert_eq!(config.d_model, loaded_config.d_model);
    assert_eq!(config.num_layers, loaded_config.num_layers);
    assert_eq!(config.num_blocks, loaded_config.num_blocks);

    let _ = std::fs::remove_file(&path);
}